A census of gold's behaviour — counted, not storied. Foundations first, then the evidence, then the sketchbook.黄金行为普查——数出来的，不是编出来的。先基础，再证据，最后是素描本。
Foundations — the course in numbers基础——用数字看这门课
First learn how the market works. Then learn how each market behaves — counted, not guessed.先学会市场如何运行。再学会每个市场如何行为——用计数，不用猜测。
18
chapters, from zero to desk-ready章节，从零到能上交易台
523
behaviours counted across 4 markets四个市场共计数 523 个行为
3 yrs
of M15 data behind every verdict每个结论背后有 3 年 M15 数据
0
approved strategies — nothing here has passed the bar yet已批准策略 0 个——还没有任何东西通过门槛
Every behaviour, judged每一个行为，都被审判All four POSEIDON censuses combined: 523 behaviour rows四个 POSEIDON 普查合计：523 个行为行
151
101
207
64
151CONFIRMED — survived unseen data已确认——通过未见数据检验101WEAK — a flag, not a finding弱——只是标记，不是结论207DEAD — killed by the holdout已死亡——被留出数据杀掉64DESCRIPTIVE — measured, no direction claim描述性——只测量，不下方向结论
Each market’s volume (tabs 2, 4, 5, 6) opens with its own picture page like this one.每个市场的卷（页签 2、4、5、6）都以像这样的图像页开头。
Read the Market
What trading really is, how price moves, and how to read a chart like a professional.
This is not a strategy book. There are no "secret setups" here. This is the foundation underneath every strategy — the market itself. Read it in order. Each chapter builds on the last. When you finish, you'll understand why price does what it does, which is the only thing that makes any strategy worth trading.
Foundation Course · 基础课程 Part of the XAUUSD Trading Desk
1. What Is Trading?
Part I · The Market Itself
Strip away the charts, the jargon, the guru courses. Trading is one thing: buying something now to sell it later at a higher price, or selling now to buy it back cheaper. That's it. Everything else is detail.
The two directions
Long (buy): you profit if price goes up. You buy gold at $2,650, sell at $2,670, keep the $20 difference.
Short (sell): you profit if price goes down. You "sell" gold you don't own at $2,670 (your broker lends it), buy it back at $2,650, keep the $20.
Shorting confuses beginners, but it's just the mirror image. In forex and gold CFDs, going short is as easy as going long — one click either way.
Trading vs. investing
An investor buys gold because they believe it will be worth more in five years. A trader buys gold because they believe it will be worth more in five minutes — or five hours, or five days. The difference is time horizon, not intelligence. Investors can be wrong for years; traders find out in an afternoon.
What you're actually trading
When you trade XAUUSD on MT5, you don't own gold. You're trading a CFD (Contract for Difference) — a contract with your broker that pays you the difference between your entry and exit price. No vault, no bars, no delivery truck. This matters because:
You can trade tiny sizes (0.01 lots = 1 ounce).
You pay a spread (the broker's cut, the gap between buy and sell price) on every trade.
Your broker is your counterparty — choose one that won't hunt your stops.
Key idea · 核心
Trading is not predicting the future. It's placing bets where the potential reward justifies the risk, repeatedly, with discipline. Anyone promising certainty is selling something.
Prices don't move by magic. Every tick is a transaction — someone bought, someone sold. The question is who, because different players move price in different ways.
The players, biggest first
Player
What they do
How they move gold
Central banks
Hold gold as reserves; set interest rates
Rate decisions shift real yields → gold reprices in minutes
Institutions & funds
Manage billions; hedge portfolios
Large orders create the trends you ride (or get run over by)
Market makers / LPs
Provide liquidity, quote buy/sell all day
They don't bet on direction — they harvest the spread
Retail traders
That's you (and me)
Individually irrelevant; collectively, our stop-losses are liquidity for bigger players
The uncomfortable truth about retail
Your 0.10 lot buy order doesn't move gold. But thousands of retail stop-losses clustered below a swing low? That's a juicy pool of sell orders. When price dips to grab them and snaps back, that's not conspiracy — it's mechanics. Big players need someone to trade against. Your stops are their fills.
This is why "liquidity" matters more than any indicator. Price is drawn to where the orders are.
Key idea · 核心
You're not trading against "the market." You're trading in a river moved by institutions, where retail stop clusters are rocks the water flows around — and sometimes over.
Forget "bulls vs bears" as a metaphor. Prices move for one mechanical reason: imbalance between buy and sell orders at the current price.
The order book (simplified)
At any moment, there's a stack of buy orders (bids) below the price and sell orders (asks) above it. When aggressive buyers eat through all the asks at $2,650, price ticks to $2,650.10, then $2,650.20 — it must move up to find more sellers. That's all a rally is: buyers consuming available supply faster than it replenishes.
What creates imbalance?
News: CPI comes in hot → traders expect higher rates → real yields rise → gold less attractive → sell orders flood in. This repricing takes minutes, not hours. (Your desk's research confirmed: no reliable multi-hour post-news drift.)
Large institutional orders: a fund rebalancing billions can't hide. Their buying pressure lifts price as the book absorbs it.
Stop cascades: price breaks a level → stops trigger → those stops are market orders → more buying/selling pressure → price extends. This is why breakouts accelerate.
Session flows: London open brings European institutional volume; NY open brings American volume. More participants = more order flow = more movement. (This is why your 21:00–00:00 MYT window is the most volatile.)
What does NOT move prices
Your indicator crossing. RSI doesn't move gold; traders reacting to RSI might, collectively, a little.
Chart patterns "completing." A head-and-shoulders doesn't cause selling — it describes selling that already happened.
What someone on YouTube predicted.
Honest warning · 诚实提醒
Most "reasons" given for price moves after the fact are stories, not causes. "Gold fell on profit-taking" explains nothing — it's just words attached to a red candle. Focus on where orders are, not stories about why.
Key idea · 核心
Price moves when one side runs out. Your job isn't to predict why — it's to position where the imbalance is likely to resolve in your favor, with a stop when you're wrong.
Every chart you'll ever read is built from the same atom: the candlestick. One candle = one period of time (1 minute, 15 minutes, 1 day — your choice). It compresses everything that happened in that period into four numbers.
Anatomy of a candle
Open: price when the period started.
High: highest price reached during the period.
Low: lowest price reached.
Close: price when the period ended. The most important of the four — it's where the market "settled."
The thick part (the body) spans open to close. The thin lines (the wicks or shadows) stretch to the high and low. Green/white body = close above open (buyers won the period). Red/black = close below open (sellers won).
What candles actually tell you
Long body, tiny wicks: one side dominated. Conviction.
Long wicks, small body (doji-like): indecision. Both sides pushed, neither held ground. Often appears at turning points — but also in the middle of nowhere.
Long lower wick: sellers pushed down, buyers absorbed it all and pushed back. Rejection of lower prices.
Long upper wick: the mirror — buyers pushed up, sellers slammed it back down.
The trap
A single candle means almost nothing in isolation. A "hammer" in the middle of a range is noise. The same hammer at a major support level after a selloff? That's a story. Context gives candles meaning — this principle runs through the entire book.
Key idea · 核心
Learn to read what each candle says about the battle between buyers and sellers in that period. Then forget single candles and read them in groups — that's where price action lives.
4. 读懂 K 线
第二部分 · 读懂图表
你以后读的每一张图都由同一种原子构成：K 线（蜡烛图）。一根 K 线 = 一个时间周期（1 分钟、15 分钟、1 天——你自己选）。它把这个周期内发生的一切压缩成四个数字。
孤立的一根 K 线几乎毫无意义。区间中部的"锤子线"是噪音。同一根锤子线出现在大级别支撑位、一波抛售之后？那就是一个故事。背景赋予 K 线意义——这个原则贯穿整本书。
Key idea · 核心
学会读懂每根 K 线在说买方和卖方在这个周期的战斗结果。然后忘掉单根 K 线，去读 K 线组合——价格行为就住在那里。
5. Timeframes
Part II · Reading Charts
The same market looks completely different on different timeframes. Gold can be in a screaming uptrend on the daily chart and a brutal downtrend on the 5-minute chart — simultaneously. Both are true. This confuses every beginner.
What each timeframe shows
Timeframe
Shows you
Used for
Monthly / Weekly
The big tide
Where is gold in its multi-year cycle?
Daily (D1)
The trend
Are we generally going up or down this month?
H4 / H1
The structure
Where are the swings? Where's the bias?
M15 / M5
The execution
Where exactly do I enter? (Your desk lives here)
M1
Noise (mostly)
Burning out your eyes. Avoid until expert.
The fractal rule
Markets are fractal — the same patterns repeat at every scale. A trend on M15 looks like a trend on D1, just faster and noisier. This is powerful: learn to read structure once, apply it everywhere.
The golden rule of timeframes
Never trade against the timeframe above your execution chart. If you're entering on M15, check H1 first. If H1 is in a strong downtrend, your M15 long is swimming upstream — it can work, but the odds are worse. This single habit filters out half of all bad trades.
Honest warning · 诚实提醒
Beginners flip between timeframes to find the one that agrees with the trade they want to take. That's not analysis — that's shopping for confirmation. Pick your timeframes in advance and respect what they say.
"The trend is your friend" is the most repeated — and most misunderstood — sentence in trading. A trend isn't a line on a chart. It's a persistent imbalance: buyers keep overwhelming sellers (uptrend) or sellers keep overwhelming buyers (downtrend).
Dow's definition (still the best)
Uptrend: higher highs AND higher lows. Each rally goes further than the last; each dip stops shallower than the last. Buyers are in control.
Downtrend: lower highs AND lower lows. Sellers in control.
Range: neither. Price bounces between two levels. No one is in control — this is where most retail money dies, buying tops and selling bottoms.
You need both conditions. Higher highs with lower lows isn't an uptrend — it's widening volatility, often before a reversal.
Trendlines
Connect two or more swing lows in an uptrend (or swing highs in a downtrend). The more touches without breaking, the more significant the line. But remember: a trendline is a visual aid, not a wall. Price owes it nothing. The third touch isn't "guaranteed to hold" — it's just where other traders are also watching, which can become self-fulfilling.
Why trends exist
Trends persist because large institutional orders take time to fill. A fund buying $500M of gold can't do it in one click — they'd spike the price against themselves. So they buy steadily over days or weeks, creating persistent buying pressure. You're surfing their wake.
Key idea · 核心
Trade with the trend on your execution timeframe, in the direction of the trend on the timeframe above. Counter-trend trading is picking up pennies in front of a steamroller — occasionally profitable, eventually fatal.
Support is a price level where buying has repeatedly overwhelmed selling — price falls to it and bounces. Resistance is the mirror: selling repeatedly overwhelms buying — price rises to it and rejects.
Why do they work?
Three reasons, in order of importance:
Orders cluster there. Traders place buys at known support, sells at known resistance. More orders = more real buying/selling pressure.
Memory. Traders who missed the last bounce at $2,600 swear "next time I'll buy there." When price returns, they do. Self-fulfilling.
Stop placement. Breakout traders put stops just beyond S/R. When price breaks through, their stops trigger as market orders, accelerating the move. This is why S/R breaks often explode.
Role reversal (the most useful S/R concept)
When support breaks, it often becomes resistance. Why? Everyone who bought at support is now losing. When price rallies back to their entry, they sell to "get out even" — creating selling pressure at the old support. Old support → new resistance. Old resistance → new support. This single concept generates more trade ideas than any indicator.
Round numbers
$2,700. $2,650. $2,600. Humans love round numbers — we place orders at them, set targets at them, talk about them. In gold, every $10 and especially every $50 level acts as mini support/resistance. Your desk's research confirmed: stop clustering around round numbers is measured and real.
Zones, not lines
Draw S/R as zones ($2,648–$2,652), not razor lines ($2,650.00). Price rarely reverses to the exact cent. Beginners draw a line, price pierces it by $1, they panic — but the zone held. Give levels room to breathe.
Key idea · 核心
The more times a level is tested without breaking, the stronger it is — but also the more stops accumulate beyond it, making the eventual break more violent. Strength and fragility grow together.
Market structure is the skeleton underneath every chart. Learn to see it and you'll stop seeing "random candles" and start seeing the battle.
The vocabulary
HH (Higher High) / HL (Higher Low) — the bones of an uptrend.
LH (Lower High) / LL (Lower Low) — the bones of a downtrend.
BOS (Break of Structure) — price breaks the last swing high/low in the direction of the trend. Trend continuation signal.
CHoCH (Change of Character) — price breaks structure against the trend. The first warning that the trend may be ending.
How to read it live
In an uptrend, watch the HLs. As long as each dip holds above the last HL and breaks make new HHs, the trend is healthy. The moment price breaks below the last HL — that's a CHoCH. It doesn't mean "short immediately." It means: the buyers are tired. Pay attention.
Why this beats indicators
Structure is derived from price itself, not from a formula applied to price. An RSI can stay "overbought" for hours in a strong trend (your PB16 RSI strategy learned this the expensive way — 28 trades, thin sample, no edge). Structure tells you what buyers and sellers are actually doing: making progress, or stalling.
Key idea · 核心
BOS = trend alive. CHoCH = trend sick. Learn to mark every swing high and low on your chart until it becomes automatic. This is the single highest-value chart skill.
This is the chapter that explains what you've already seen in your backtests — especially PB7 and PB12.
What is liquidity?
Liquidity = resting orders. Every stop-loss, every buy-stop, every limit order sitting at a level is liquidity. Big players need liquidity to fill large orders without moving price against themselves. So price is magnetically drawn to where liquidity sits.
Where does liquidity hide?
Beyond swing highs/lows — breakout traders' stops live here.
Beyond session ranges — the Asian high/low, yesterday's high/low.
Round numbers — $2,700, $2,650.
Equal highs/lows — double tops/bottoms are stop magnets.
The sweep
A sweep (or "stop hunt") is when price briefly spikes beyond a level, triggers the resting stops, then reverses. The spike is the point — it wasn't "fake," it was the market collecting the orders it needed. Then, with stops filled, price is free to move the real direction.
Your PB7 (Sweep-Reclaim) tested exactly this: fade the sweep after price reclaims the level. It failed (PF 1.04) — not because sweeps aren't real, but because fading every sweep without confirmation is just guessing. The Aegis v2.2 you shared adds what PB7 lacked: rejection-candle confirmation + conviction scoring + session filtering.
How to use this
Before entering near a level, ask: where are the stops? If your stop is in the obvious place, it's in the danger zone.
A sweep + strong rejection + reclaim is meaningful. A sweep alone is just volatility.
Never place stops at the exact obvious level. Give them room beyond the liquidity pool — or accept you'll be swept sometimes.
Honest warning · 诚实提醒
"Smart money" narratives are mostly marketing. There is no villain twirling a mustache hunting your stop. It's mechanics: resting orders attract price the way gravity attracts water. No conspiracy required.
Chapter 5 introduced timeframes. Now the method: top-down analysis. You read from large to small, and only trade when they agree (or when you have a specific reason to trade the disagreement).
The three-layer read
Bias (D1/H4): What's the big direction? Uptrend, downtrend, or range? This is your permission — it tells you which direction you're allowed to trade.
Structure (H1): Where are the swings? Is price at support or resistance? This is your map — it tells you where opportunity lives.
Execution (M15/M5): What's the entry trigger? Sweep + rejection? BOS? This is your trigger — it tells you when.
Example: a real read
D1: uptrend, price pulled back to support. H1: bounced, making HH/HL. M15: swept the Asian low, printed a bullish engulfing, reclaimed. Three timeframes agree: long. That's a high-quality setup — not because of a pattern, but because of alignment.
When timeframes disagree
D1 uptrend but H1 in a deep pullback? Stand aside or trade smaller. M15 screaming long while H1 resistance looms 5 points above? The reward isn't worth it. Disagreement = reduced size or no trade. This one rule would have saved most of your killed playbooks from their worst quarters.
Key idea · 核心
Bias → Map → Trigger. Never take a trigger without checking the map. Never read the map without knowing the bias.
Gold doesn't move equally all day. It breathes — quiet in Asia, waking in London, roaring in the New York overlap, fading after. Your desk's research measured this: US hours have the highest gold volatility, Asia the lowest. Trade the breathing, not the flatline.
The sessions (in your time, MYT)
Session
MYT
Character
Asia / Tokyo
07:00–14:00
Quiet. Tight ranges. Builds the levels London will hunt.
London
15:00–18:00
Wakes up. Often hunts Asia's high/low, then picks a direction.
NY Overlap
21:00–00:00
The main event. Biggest volume, biggest moves. Your window.
Late NY
00:00–05:00
Winds down. Choppy, thin — amateurs give back profits here.
Why this matters more than patterns
The same setup at 10:00 MYT (dead Asia) and 22:00 MYT (NY overlap) has completely different odds. Volatility is the fuel; without it, even perfect structure goes nowhere. Every playbook you tested lived or died partly on when it traded. The Aegis v2.2's "Kill Zones" (only the first 45–90 min of a session) is this principle weaponized.
News: the volatility bomb
CPI, NFP, FOMC — these detonate inside sessions. Price can move $30 in a minute. Rules:
Don't trade into major news. Spreads widen, slippage explodes, stops get slipped.
Trade the aftermath, not the spike. Wait 15–30 min for the market to pick a direction, then trade with the post-news structure.
Your desk's scouter briefs these daily. Read it before your session.
Key idea · 核心
When you trade matters as much as what you trade. Fish when the fish are biting: London open and NY overlap. Sleep through Asia.
Engulfing. Hammer. Doji. Morning star. You've seen the posters. Here's the honest version.
What patterns really are
A candlestick pattern is shorthand for a short story about buyer-seller interaction. A bullish engulfing says: "sellers pushed, buyers overwhelmed them completely, and closed strong." That's useful information — but it's not a signal by itself.
The patterns worth knowing
Engulfing: body completely covers prior candle's body. Shows a decisive shift in that period.
Pin bar / hammer: long wick one side, small body. Rejection of prices on the wick side.
Inside bar: entire range inside prior candle. Compression — energy building for an expansion.
Doji: open ≈ close. Pure indecision.
The honest math
Tested alone, most candlestick patterns have no statistically significant edge. A hammer wins maybe 50–52% of the time — before costs. After spread, it's negative. Patterns work when they appear at the right place: a hammer at major support in an uptrend, an engulfing after a liquidity sweep, an inside bar at a session open about to expand.
Anyone selling a course based purely on candlestick patterns is selling hope. Patterns are vocabulary, not strategy. You need sentences (structure), paragraphs (context), and a thesis (bias) before individual words matter.
12. K 线形态
第四部分 · 背景
吞没、锤子、十字星、启明星。你都见过那些海报。下面是诚实版。
形态到底是什么
K 线形态是关于买卖双方互动的速记故事。看涨吞没在说："卖方发力了，买方彻底压倒了他们，并强势收盘。"这是有用的信息——但单独不是信号。
值得知道的形态
吞没：实体完全覆盖前一根实体。显示这个周期内决定性的转向。
pin bar / 锤子：一侧长影线、小实体。拒绝影线那一侧的价格。
内包线：整个区间在前一根之内。压缩——能量在为扩张蓄力。
十字星：开盘 ≈ 收盘。纯犹豫。
诚实的数学
单独测试，多数 K 线形态没有统计显著的优势。锤子线胜率大概 50–52%——还是扣成本之前。扣掉点差就是负的。形态在正确的位置出现时才有效：上升趋势中大支撑位的锤子、扫单后的吞没、时段开盘即将扩张时的内包线。
形态 + 位置 + 背景 = setup。单独的形态 = 抛硬币。
Honest warning · 诚实提醒
任何只卖 K 线形态的课程都是在卖希望。形态是词汇，不是策略。在单个词有意义之前，你需要句子（结构）、段落（背景）和论点（偏向）。
13. Risk Management
Part V · The Trader's Math
This chapter is worth more than all the others combined. Most traders don't fail because their analysis is wrong — they fail because one bad stretch wipes them out. Risk management is the difference between "wrong" and "ruined."
The R multiple
Stop thinking in dollars. Think in R — your risk unit. If you risk $10 on a trade, that's 1R. If the trade wins $20, that's +2R. If it loses, −1R. R lets you compare trades across sizes and see your true edge.
Position sizing (the only formula that matters)
Risk per trade: 1% of account. $1,000 account → risk $10 per trade. Your stop distance determines your size:
Position size = Risk $ ÷ Stop distance $
$10 risk ÷ $20 stop = 0.5 oz. That's it. Wide stop = small size. Tight stop = bigger size. Risk stays $10. Never adjust risk to "feel good" about a trade.
Expectancy: the one number
Expectancy = (Win% × Avg Win) − (Loss% × Avg Loss), in R.
Win 40% at +2R average, lose 60% at −1R: (0.40 × 2) − (0.60 × 1) = 0.8 − 0.6 = +0.2R per trade. That's a real edge — 100 trades = +20R = +20% at 1% risk. Notice: you can win less than half your trades and still print money, if your winners are bigger than your losers.
The kill rules (your desk already has these)
−3R daily stop. Lose 3R in a day → screens off. You're tilted, not analyzing.
−6R weekly stop. Lose 6R in a week → done for the week. The market will be there Monday.
Max 2 positions. More than that and you're not managing risk, you're gambling.
The math of ruin · 爆仓数学
Lose 50% of your account and you need +100% just to get back. Lose 20% three times and you're nearly halved. Capital preservation isn't conservative — it's the only way compounding works. Every trader who blew up knew this chapter and ignored it.
Your brain was built to survive on the savanna, not to trade gold. Every cognitive bias you have is a designed-in trading bug. Knowing them doesn't cure them — but it lets you build systems that route around them.
The big four
Loss aversion: losses hurt ~2× more than equal wins feel good. Result: you cut winners early ("lock it in!") and let losers run ("it'll come back"). The exact opposite of profitable trading.
Recency bias: three losses in a row and your brain screams "the strategy is broken!" — even with 200 trades of edge behind it. (Your desk killed 17 strategies; some may have died to recency bias rather than real failure. The kill rules exist to separate the two.)
FOMO: price rips without you, you chase the entry, buy the top. The market always gives another setup. Chased entries have no edge — you missed the risk/reward.
Revenge trading: after a loss, you trade bigger to "get it back." This is how −1R becomes −6R becomes a blown account. The daily kill rule exists for exactly this moment.
The fix isn't willpower
You don't beat psychology with discipline — you beat it with structure. Pre-written plans. Hard kill rules. Fixed risk per trade. A journal that forces you to write why before you click. The desk's game plan + monitor + reviewer crew is a psychological exoskeleton: it makes the right behavior the default.
The journal (non-negotiable)
Every trade: setup, entry, stop, target, reason, and how you felt. Review weekly. You'll discover your edge isn't what you think — and your leaks are embarrassingly consistent. The traders who journal survive. The ones who don't, don't.
Key idea · 核心
Amateurs focus on entries. Professionals focus on risk and psychology. The market doesn't reward being right — it rewards surviving long enough for your edge to play out.
Now we extend everything to your instrument. Gold isn't just "another chart" — it has specific drivers, and misunderstanding them is why generic strategies fail on XAUUSD.
The real drivers (in order)
Real yields. The #1 driver. When inflation-adjusted US bond yields rise, holding non-yielding gold costs more in opportunity → gold falls. When real yields fall → gold rises. Watch the 10-year TIPS yield, not just the Fed.
The US dollar. Gold is priced in dollars globally. Dollar up → gold more expensive for foreign buyers → downward pressure. But it's correlation, not causation — your desk tested DXY as a hard filter and it failed (PF 1.22). Use it as context, not a gate.
Risk sentiment. Panic → money flees to gold (safe haven). Euphoria → money rotates to stocks. Geopolitical shocks spike gold in minutes.
Central bank buying. Structural bid — central banks (especially non-Western) have been steady buyers for years. This puts a slow floor under major selloffs.
Positioning & flows. COMEX futures positioning, ETF flows. When everyone's already long, who's left to buy? Extremes in positioning often mark tops/bottoms.
What does NOT drive gold intraday
Jewelry demand (too slow to matter for your timeframe).
Mining supply (glacial).
Most "analyst forecasts" (stories, not orders).
Gold's personality
Gold trends harder and reverses more violently than most forex pairs. Daily ranges of $20–40 are normal; news days can print $60+. It's emotional — fear and greed show up faster here than in EURUSD. Respect the volatility: size for gold's range, not forex's.
Key idea · 核心
Gold = real yields + dollar + fear. Everything else is noise at your timeframe. When real yields spike against your long, don't argue with the bond market — get out.
Everything in this book converges here. A complete read of gold, top-down, the way a professional does it before risking a dollar.
The pre-session routine (15 minutes)
D1: Trend or range? Where's price relative to the last major swing? Mark the key S/R zones.
H1: What's the structure — HH/HL or LH/LL? Any BOS or CHoCH today? Where's price relative to H1 S/R?
News: What's on the calendar tonight? (Your scouter briefs this.) Any landmines in your window?
Levels: Yesterday's high/low, Asian high/low, round numbers ($10/$50), equal highs/lows. Mark them all — that's where liquidity sits.
Plan:If price does X at level Y, then I do Z. Write it. No plan = no trade.
The live read (your 21:00–00:00 window)
Watch the NY open. The first 30–60 minutes often set the tone: does price hunt the Asian high/low (sweep) and reverse, or does it break and go? Don't trade the first 15 minutes — spreads are wide, direction is unclear. Let the market show its hand, then trade the structure it reveals.
Putting the whole book on one chart
Trend (Ch 6) + structure (Ch 8) tell you the direction.
S/R zones (Ch 7) + liquidity map (Ch 9) tell you where.
Session (Ch 11) tells you when.
Candles (Ch 4, 12) tell you how price is behaving right now.
Risk rules (Ch 13) tell you how much.
Psychology (Ch 14) keeps you executing the plan instead of your emotions.
The whole game · 终极
Reading a chart isn't about finding the perfect entry. It's about understanding where you are (trend/structure), what's likely next (liquidity/session), and what you'll do if wrong (risk). Do that consistently and entries take care of themselves.
What real traders actually read — with honest reviews
No book will give you an edge. Books give you vocabulary, risk management, and the collected mistakes of people who blew up before you — so you can make new mistakes instead of repeating theirs. Read in this order.
1. Trading in the Zone — Mark Douglas
Psychology · ★★★★★
Read first. The single most important trading book. Douglas argues that consistent profitability is 80% psychology: accepting randomness, thinking in probabilities, and executing without fear. Every blown account is a psychology failure wearing an analysis costume. If you read one book, this is it.
2. Reminiscences of a Stock Operator — Edwin Lefèvre
Speculation wisdom · 1923 · ★★★★★
The fictionalized memoir of Jesse Livermore, the greatest speculator of the early 1900s. A century old and still painfully relevant: "It never was my thinking that made the big money for me. It always was my sitting." About patience, position sizing, and how markets never change because human nature never changes. Free — it's public domain.
3. Market Wizards — Jack Schwager
Interviews · ★★★★☆
Interviews with the best traders of the 1980s. No single system — that's the point. Every wizard trades differently, but ALL of them obsess over risk management and cut losses fast. Read it to kill the fantasy that there's one right way to trade.
4. Japanese Candlestick Charting Techniques — Steve Nison
Candlesticks · ★★★★☆
The book that brought candlesticks to the West. Thorough catalog of every pattern — but read it alongside Chapter 12 of this course: patterns are vocabulary, not edge. Useful as a reference, dangerous as a strategy.
5. Technical Analysis of the Financial Markets — John Murphy
Comprehensive TA · ★★★★☆
The textbook. Trends, patterns, indicators, intermarket analysis — everything in one place. Dry but complete. Use it as a reference manual, not a cover-to-cover read. The intermarket chapter (bonds vs stocks vs commodities vs dollar) is gold for understanding XAUUSD's drivers.
6. The Disciplined Trader — Mark Douglas
Psychology II · ★★★☆☆
Douglas's earlier, denser book. Same message as Trading in the Zone but more theoretical. Read it only if the Zone clicked and you want the deeper framework. Otherwise skip — diminishing returns.
What NOT to read · 别读什么
Any book promising "the secret," "90% win rate," or "trade like a bank." Anyone selling certainty is monetizing your hope. The books above are honest about difficulty — that's how you know they're real.
How to read them · 怎么读
One book per month, maximum. Take notes. Paper-trade the concepts for two weeks before judging them. And re-read Trading in the Zone every year — you'll understand a different chapter each time, because the chapter you need changes as you grow.
63CONFIRMED — survived unseen data已确认——通过未见数据检验31WEAK — a flag, not a finding弱——只是标记，不是结论72DEAD — killed by the holdout已死亡——被留出数据杀掉22DESCRIPTIVE — measured, no direction claim描述性——只测量，不下方向结论
How wide each session runs每个时段跑多宽Median session range in dollars — Tier A year. Asia is NOT quiet: it runs as wide as London.时段区间中位数（美元）——Tier A 年度。亚洲并不安静：它和伦敦一样宽。
Off-hours非交易时段
$17.0
London伦敦
$33.7
Overnight隔夜
$35.6
Asia亚洲
$36.7
New York纽约
$41.2
Overlap重叠
$44.9
The friction tax — cost of one round trip摩擦税——一次往返的成本Round-trip cost as a fraction of one median M15 candle (ATR). Smaller = cheaper to trade.往返成本占一根典型 M15 K 线（ATR）的比例。越小 = 交易成本越低。
Gold黄金
0.07×
NAS100纳斯达克100
0.096×
USD/JPY美元/日元
0.12×
EUR/USD欧元/美元
0.20×
One day on the clock时钟上的一天Session blocks on the desk clock (UTC). Edge segments — first and last 30 minutes — are where the highs and lows form.交易台时钟（UTC）的时段区块。边缘时段——最初和最后 30 分钟——是高低点形成的地方。
Off-hrs非交易时段
$17.0
20–23 UTC
Overnight隔夜
$35.6
23–02
Asia亚洲
$36.7
02–07
London伦敦
$33.7
07–12
Overlap重叠
$44.9
12–15
New York纽约
$41.2
15–20
Laws that survived all four markets在四个市场都活下来的规律
📰
The first spike lies第一波冲高会说谎
On 08:30 ET data days the first spike is faded in gold, EUR/USD and USD/JPY. Never chase it.在 08:30 ET 数据日，第一波冲高在黄金、欧元/美元和美元/日元中都会被反打。永远不要追它。
📉
Streaks snap back连涨连跌会弹回
Three same-direction days lean the other way over the next five — in all four markets.连续三天同方向，接下来五天会倾向反方向——四个市场全部如此。
🧭
Breaks continue — except EUR/USD突破会延续——欧元/美元除外
An accepted break of yesterday’s range keeps going in gold, USD/JPY and NAS100. In EUR/USD it reverses.被接受的昨日区间突破在黄金、美元/日元和纳斯达克100中会继续。在欧元/美元中它会反转。
⏱
Edges make the extremes边缘产生极值
Session highs and lows form in the first or last 30 minutes — everywhere, in every market.时段高低点在最初或最后 30 分钟形成——每个市场、每个地方都如此。
The headline behaviours — look first, read later头条行为——先看图，后读字
The break that sticks守得住的突破
A close beyond yesterday’s high or low, held outside, usually keeps going the same day. Follow the close, not the wick.收盘站到昨日高低点之外并守住，当天通常会继续。跟随收盘，不要跟随影线。
A big candle that swallows the last one — away from any level — leans continuation. Gold’s sturdiest candle behaviour, confirmed three times.一根吞没上一根的大 K 线——远离任何价位时——倾向延续。黄金最稳健的 K 线行为，被确认了三次。
London hands its direction to the Overlap and the Overlap keeps it — the most reliable session handoff in the gold census.伦敦把方向交给重叠时段，重叠时段会守住它——黄金普查中最可靠的时段接力。
CONFIRMED · +0.42 ATR median · both corpora已确认 · 中位数 +0.42 ATR · 两个语料库
The first news spike lies第一波新闻冲高会说谎
On 08:30 ET data days the first spike is followed only 43% of the time — on CPI days just 39%. The spike is the trap, not the trade.在 08:30 ET 数据日，第一波冲高只有 43% 的时候被跟随——CPI 日只有 39%。冲高是陷阱，不是交易。
The full chapters follow below — every number above is drawn from them, unchanged.完整章节在下方——上面的每一个数字都来自它们，未作任何改动。
Start Here从这里开始
How Gold Behaves · Tab 2
Everything in this tab was counted, not storied. A census engine scanned gold's recorded history, found every instance of every behaviour defined in advance — the beautiful ones and the boring failures alike — stacked them on top of each other, and measured what price did next. Where a behaviour survives that treatment, it is written here as a fact about gold. Where it dies, it is buried in the open, in the Graveyard chapter, with the cause of death on the stone.
Two disciplines hold the whole book together. First, eyes propose, numbers dispose: a pattern a human thinks they see is only a suggestion until every instance has been counted. Second, nothing here is a strategy. Every number is gross of trading costs, and the desk's readiness bar (≥50 trades, profit factor ≥1.3 out-of-sample, ≥2 market periods, costs included) stands between anything in this book and real money. The Approaches tab shows what these behaviours might become — labelled unvalidated, because they are.
Every behaviour was measured twice: on a discovery sample (the first 60% of instances, chronologically) and a holdout sample (the last 40%) it was not tuned on, and compared against a null — random timestamps matched on session and direction, which absorbs the sample's overall drift. The verdict rule was written down before any measurement:
One chart is an anecdote; a stack is a shape. Each stacked chart in this book overlays every instance of a behaviour, aligned so the event sits at zero, with price movement normalised in ATR units (one ATR ≈ one typical bar's range, so $2,000-era and $4,000-era moves can share a chart). The gold line is the median instance, the pale line the mean, and the shaded band holds the middle half of all instances. Hover to scrub values; use the toggles to isolate a line. If the median hugs zero and the band is symmetric, there is no behaviour — whatever the stories say.
Before asking what gold does at a level or after a pattern, ask a ruder question: what kind of day is it? Every trading day in ten years was mechanically classified by three numbers — how efficiently price travelled from open to close, where the close sat inside the day's range, and how wide the day was against its ATR. No human judgement, no hindsight. This is the population every other chapter samples from.
Ten years of gold days, classified (share of 2,514 days)十年黄金交易日的分类（占 2,514 天的比例）
The headline a beginner should memorise: most days are not trend days. Range days (31.1%) and quiet-other days (30.6%) make up almost two-thirds of gold's life. Clean trend-up days are 18.7%, trend-down 15.8%. And the dramatic spike-and-reverse day of trading folklore — a violent spike that fully reverses — fired its mechanical definition just 0.7% of the time combined. If your mental image of gold is mostly drama, your sample is biased: drama is what memory keeps.
The classes are not just labels — they lean forward differently. After a trend-up day, the class-direction median five days later is still +0.34 ATR: up days cluster into up weeks. After a trend-down day, the one-day median is −0.13 ATR continuation-signed — down days, in this decade, tended to bounce the next day rather than continue. Read that pair carefully: it is the whole decade's bull-market asymmetry in miniature, and it will reappear in chapter after chapter. The null baseline absorbs the average drift, but state-dependent drift — the market's mood after up days versus down days — is exactly what a census can see and a story cannot.
The spike-reversal definitions nearly never fired as specified (0.6% and 0.1%) — that is recorded, not retuned. Definitions that are loosened until the interesting class appears are how folklore is manufactured. Also: the intraday-derived day frequencies for 2025–26 (Range 31.7%, Trend-up 23.0%, Trend-down 15.5%) resemble the decade's mix, but one bull year is not a decade.
Levels are where trading stories concentrate: yesterday's high, last week's low, the range a session left behind. The census separates two events the stories merge — a level being accepted (price closes beyond it and stays out) and a level being rejected (price wicks beyond it and closes back inside). They are different animals, and gold treats them differently.
When gold closes a 15-minute bar beyond the prior day's high or low — the first close outside yesterday's range — the move tends, on median, to keep going over the next two hours rather than snap back. This confirmed twice: on the trusted 2025–26 year (+0.10 ATR discovery / +0.22 holdout versus a matched null) and again on the longer extended corpus (+0.16 / +0.14, n=291).
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
2025-12-17 13:30 UTC · 2026-03-05 15:00 UTC · 2026-09-22 05:30 UTC
2025-12-17 13:30 UTC · 2026-03-05 15:00 UTC · 2026-09-22 05:30 UTC
Honest limits诚实边界
The pooled median hides the truth's shape: the raw pooled median is ≈ 0 and the positive mean is right-tail driven. Split by side, up-breaks carry the entire effect (raw 2h medians +0.56 ATR trusted, +0.43 extended) while down-breaks did not follow through (−0.42 / −0.39). In a sample where gold rose from $3,865 to $4,200, that asymmetry may be the bull market itself wearing a costume. Treat as regime-coloured until it survives a non-bull sample. Second breaks of the same level showed nothing (medians ≈ 0 to −0.23).
On the daily scale, over ten years, the first touch of the prior week's high or low is a magnet touch, not a launch. When the touch is a rejection — price wicks beyond last week's extreme and closes back inside its range — gold then underperforms a drift-matched random baseline over the next five days (−0.11 ATR discovery / −0.28 holdout, CONFIRMED as a negative). The sharpest cell: rejections at the week high fail — price comes back up through the level, and fading them (shorting the rejection) lost, raw median −0.16 ATR.
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
2017-10-23 · 2021-07-06 · 2024-01-30 (daily bars)
2017-10-23 · 2021-07-06 · 2024-01-30（日线）
Honest limits诚实边界
The mirror event is no better: accepted breaks of prior-week levels are WEAK-negative too (−0.40 / −0.08, n=287), and accepted breaks below the week low reversed (raw −0.32). Both directions of trade at week levels lost to baseline in this decade. The sample is a decade-long bull market, so 'up moves persist, down moves bounce' is partly the drift showing through — the null absorbs the average drift, but not every timing confound.
A descriptive gradient sharpens the accepted-break story. For prior-day breaks, the closer the breaking close sits to the level, the better the continuation: raw 2h medians by tercile of break distance run +0.24 (near), +0.08 (mid), −0.15 (far). A close that has already travelled far beyond the level has, on median, spent its move — far-overshoot closes reverse. On the daily scale at prior-week levels there is no gradient (all three terciles negative, −0.05 / −0.27 / −0.13) — consistent with week levels being magnets at every distance.
The sweep that closes back, drifts back — slightly收回来的扫盘会微微漂回
WEAK弱
When London-hours price wicks beyond the completed Asia range and closes back inside it, fading that sweep (trading back toward the range) was slightly positive in both halves of the trusted year (+0.04 / +0.03 ATR). It is real and it is tiny — the smallest effect in this chapter, kept at WEAK by the verdict rule.
Frequency: ≈ 10.8 per month (≈ 130 per year) · Dukascopy M15 · n = 132频率：每月约 10.8 次（每年约 130 次）· Dukascopy M15 · n = 132
Stacked paths for Asia-range sweep rejections in London hours, Dukascopy M15.
伦敦时段亚洲区间扫盘拒绝的叠加路径，Dukascopy M15。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
Archetypes live in the census file for this behaviour (WEAK behaviours carry no book archetypes).
原型实例存放在该行为的普查文件中（弱行为不带书籍原型）。
Honest limits诚实边界
Its sibling — accepted Asia breaks in London — was CONFIRMED in batch 1 and then killed on the extended corpus (sign flip). The pair together are the book's standing warning about session-edge behaviours in a single bull year: one half's gold is the next half's fool's gold. Both are recorded; only the fade survives, at WEAK.
Textbook lore says candlestick patterns matter at levels — at support, at resistance, where the battle is. The census tested that claim directly, splitting every pin bar, engulfing bar and inside bar into 'at a level' and 'mid-range'. The result came out upside-down, and it survived every attempt to kill it. This chapter is the book's cleanest example of why counting beats remembering.
The engulfing that works is the one at no level有效的吞没不在任何价位处
CONFIRMED已确认
An engulfing bar — a bar whose body fully swallows the previous bar's body — that appears away from any listed level is followed, on median, by continuation in its own direction over the next two hours (+0.15 ATR discovery / +0.22 holdout vs null). This is the most-replicated intraday behaviour in the census: it confirmed on the trusted year (n=985), again on the extended corpus (+0.04 / +0.23, n=1,788), and a third time under the strict level definition (+0.06 / +0.25, n=1,975). Both directions contribute (bull engulfings raw +0.08, bear +0.16).
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
2025-10-01 17:30 UTC · 2026-01-19 10:00 UTC · 2026-04-30 18:30 UTC
2025-10-01 17:30 UTC · 2026-01-19 10:00 UTC · 2026-04-30 18:30 UTC
Honest limits诚实边界
Perspective before enthusiasm: the median effect is ≈ +0.2 ATR against a typical adverse excursion near 1 ATR — a population tilt, not a per-trade promise, and gross of costs. The at-level twin of this pattern is DEAD (Graveyard): the same bar at a level carries no such continuation. Location doesn't amplify this pattern; it abolishes it.
A real archetype instance, replayed bar by bar: a bullish engulfing bar at 10:00 UTC on 19 Jan 2026 (extended corpus), chosen because its path is the closest to the median of all 1,788 extended-corpus instances. This is what the median of a confirmed behaviour actually looks like — including the wobble.
Below is another real instance, frozen at the engulfing bar's close (19 Jan 2026, 10:00 UTC, trusted corpus). Before you reveal it: the census base rate says the next two hours finished in the engulfing's direction 53% of the time. Which way did this one go?
It went the other way: −0.66 ATR (−$4.40) over the next two hours. That is not the behaviour failing — it is the behaviour being a 53% tilt instead of a law. Nearly half of all instances do this. Anyone who shows you only the engulfings that worked is showing you a coin's one side.
Honesty required a disclosure and a retest. In batch 1, 'at a level' was defined broadly — within 0.25 ATR of any listed level, including the running session's own high/low — and 71% of all bars qualified, which made 'mid-range' the selective bucket and 'at level' close to meaningless. Batch 2 re-ran the location question with a strict definition (within 0.15 ATR of the prior day's or prior week's extremes, or a completed session's range; still 40% of bars — levels are everywhere on a 15-minute chart).
Pin bars at strict levels: WEAK, negative (−0.07 / −0.02) — bull pins at real levels slightly underperformed. The batch-1 'pin at level' confirmation did not survive its own re-examination and was removed (see Graveyard).
Engulfing at strict levels: DEAD (sign flip). Engulfing mid-range: CONFIRMED a third time.
Inside bars, both contexts: DEAD. An inside bar does not predict expansion — forward movement after one is at or below the random baseline.
Family C · Dukascopy M15 + derived H1 + GC=F daily
Momentum lore says strength begets strength. The census agrees — up to a count. Gold's short runs lean with the trend; its long runs lean against it. The chapters of this family are about finding exactly where one hands over to the other.
After four consecutive same-direction 15-minute closes, and again after five, gold's next two hours lean against the streak versus a matched null (−0.06 / −0.06 ATR for streak-4; −0.09 / −0.05 for streak-5, both CONFIRMED negative). Three closes is not enough — that version is DEAD intraday. The daily scale agrees and adds a split worth memorising: after four straight down-days, gold bounced (continuation-signed raw median −0.59 ATR); after four straight up-days it kept drifting, but damped (+0.19 raw). Down-moves exhaust harder than up-moves in this sample.
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
Archetypes live in the census file for this behaviour.
原型实例存放在该行为的普查文件中。
Honest limits诚实边界
These are tilts of ≈ 0.05–0.09 ATR — real, replicated across two timeframes, and far too small to trade alone against costs. Their desk value is as a veto: the fifth consecutive close in your direction is a statistically documented reason not to chase.
In a confirmed H1 uptrend (EMA stack aligned), the first pullback bar that touches EMA20 or EMA50 resumed the trend versus null by small margins (EMA20: +0.03 / +0.16 ⚠; EMA50: +0.10 / +0.05). The distributions are the honest part: at the 2h horizon, closes finished back above the entry-close in the trend's favour only 33% (EMA20) and 24% (EMA50) of the time, versus outright failure in 44% and 38%. Most pullbacks take longer than two hours to resolve — or fail. The daily-scale twin of this behaviour is DEAD: buying the dip to a daily EMA, as defined, had no holdout pulse.
Stacked paths for H1 EMA50 trend pullbacks, Dukascopy derived H1.
H1 EMA50 趋势回调的叠加路径，Dukascopy 衍生 H1。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
Archetypes live in the census file for this behaviour.
原型实例存放在该行为的普查文件中。
Honest limits诚实边界
The EMA20 version is flagged ⚠ fragile (its discovery half is +0.03 — the confirmation rests on the holdout). Both EMA runs cover one bull year on one feed; the resume-versus-fail split, not the headline effect, is the durable lesson.
EMA20 版本标注 ⚠ 脆弱（其发现半区只有 +0.03——确认靠的是留出半区）。两个 EMA 运行都只覆盖一个数据源的一个牛市年份；持久的教训是恢复与失败的比例，而不是标题效应。
After very big percentage days超大百分比日之后
Percentage moves tell a different story from ATR moves, and the census ran both. After a ≥2% down day, gold's next five days are relatively weak versus a state-matched signed null (+0.14 / +0.20, CONFIRMED) — yet the raw median is still up +0.09 ATR, because the baseline itself drifts up after down-days. Relative weakness is not continuation: shorting big red days still faced a positive raw median. After ≥1% up days the mirror holds weakly (−0.07 / −0.04, WEAK): relative softness, absolute drift still up. The ≥1% down and ≥2% up versions are DEAD (Graveyard).
Volatility is the most honest variable in this book: it clusters, it persists, and it does not care about your narrative. Direction after volatility events is another matter — gold's loud days, on the evidence, more often mark local exhaustion than the start of a run.
Big days echo in size, lean against in direction大日子在幅度上回响，在方向上反倾
CONFIRMED已确认
Define a monster: a daily range beyond 2× ATR(14), or a close beyond 1.5 ATR from the open. Two facts survive the decade. Size persists: the next five days' absolute movement runs hot versus baseline (+0.19 / +0.18 ATR on |return|). Direction leans against the monster (−0.21 / −0.23, CONFIRMED): downside monsters reversed upward (raw next-5d median −0.24 continuation-signed, n=55), while upside monsters kept drifting (+0.33 raw) but still less than their matched baseline — the up-drift after a giant up-day is smaller than the drift after an ordinary day.
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
2018-08-24 · 2020-09-21 · 2024-06-07 (daily bars)
2018-08-24 · 2020-09-21 · 2024-06-07（日线）
Honest limits诚实边界
n=95 over ten years is a small census cell, and the two halves of the effect (size vs direction) are different claims with different strengths — the size persistence is the sturdier one. Dollar effects here are price-era-sensitive and are quoted in the census files, not in this chapter.
Sort days by their ATR percentile and the forward picture is stable across both halves of the decade: days born in the lowest-volatility tercile travel further over the next five days in ATR units (median |return| 1.17 ATR) than days born in the highest tercile (0.88 ATR) — mean reversion of volatility itself, visible because the ruler (ATR) stretches in loud regimes. In dollars the ordering reverses, which is the whole lesson: volatility regimes are partly a property of your measuring stick. Signed drift barely differs by regime (+0.22 … +0.32 ATR across all three) — quiet does not mean safe to fade, and loud does not mean directional.
Two pre-registered 'compression' definitions met opposite fates, and the difference is instructive. The Bollinger squeeze (bandwidth in its lowest quintile of the trailing year) leaned toward expansion in both halves (+0.15 / +0.09 on forward |return|, WEAK) with 58% up-closes at five days and a signed median of +0.49 ATR — bull-flavoured, promising, unproven. The three-day range compression definition died in the holdout (+0.01). Same story-shaped idea, different mechanical definitions, opposite verdicts: another reminder that in this book, the definition is the behaviour.
Mean reversion is the intraday trader's favourite religion, and the census is its least sentimental examiner. Of everything tested in this family, exactly one cell survived — a specific break, in a specific window, on a specific side. Everything broader died or turned negative. Read the survivor's borders as carefully as the survivor.
Accepted breaks below the London low snap back伦敦低点下方的接受突破会弹回
CONFIRMED已确认⚠ FRAGILE脆弱
During US hours, when price accepts a break below the completed London session's low (closes beyond it), the break tends to fail: the pooled behaviour confirmed (+0.08 / +0.64 ATR versus null), and the batch-4 session census narrowed it to the Overlap window, downside only — that one cell's raw median reversion is +1.14 ATR with 60% of instances positive (n=63). By the New York window the same downside cell is 0.00 — the snapback, if it comes, comes early. The other three cells refuse to cooperate: downside rejections kept falling (reversion-signed −0.73 — fading them loses), upside accepted breaks continued modestly (−0.31 reversion-signed), upside rejections drifted back (+0.27).
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
2025-12-09 15:45 UTC · 2026-04-16 13:45 UTC · 2026-09-23 13:30 UTC
2025-12-09 15:45 UTC · 2026-04-16 13:45 UTC · 2026-09-23 13:30 UTC
Honest limits诚实边界
Flagged ⚠ fragile: the pooled discovery half is ≈ +0.08 — nearly all the strength sits in the recent holdout. One cell of four carries the confirmation; the pooled frequency (≈21/mo) is not the tradable cell's frequency (≈63 instances in the year for downside-Overlap). A behaviour this localised must re-confirm on new data before it earns anything.
A real downside instance: price has just accepted a break below the London low inside the Overlap (16 Apr 2026, 13:45 UTC). The census base rate for this exact cell: 60% of instances finished the next two hours back in the reversion direction, median +1.14 ATR. Did this one snap back?
Yes — and by more than the median: +1.31 ATR (+$18.40) back upward over the next two hours. Remember the other 40% while you enjoy this one: two in five of these breaks simply kept going, and the pooled behaviour across all four cells is a coin flip (50% positive) — the edge lives only inside this one cell, in this one window.
When daily RSI(14) closes beyond 70 or below 30, the textbook says fade it. The decade says the fade loses: −0.12 / −0.37 ATR versus the signed null, CONFIRMED negative. The damage is one-sided — it lives in the overbought cell: fading strength after RSI>70 faced a fade-signed raw median of −0.33 ATR (n=59), and only 37% of fades finished positive at five days. The oversold cell is merely unhelpful (≈ +0.06 raw, n=24). In plain language: in 2016–2026, gold that looked 'overbought' on a daily chart tended to become more so, or stay so, long enough to hurt the fader.
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
2021-03-01 · 2024-09-20 · 2026-06-10 (daily bars)
2021-03-01 · 2024-09-20 · 2026-06-10（日线）
Honest limits诚实边界
The dollar figures in the census (−$1.80 / −$12.20 medians) are era-sensitive — early-sample instances traded at a third of today's price. The ATR verdict is the durable one. RSI(2) extremes, the faster cousin, are DEAD (sign flip) — see the Graveyard.
Gold trades around the clock, but it does not trade the same around the clock. The census cut every day into six fixed UTC windows — Overnight, Asia, London, Overlap, New York, Off-Hours — and gave each session its own full census: its anatomy (size, shape, when its extremes form), its internal behaviours (what its first hour's breakouts and fakeouts do), and its handoffs (what the next session does with this session's direction and range). Fixed UTC means daylight-saving transitions blur the edges by an hour part of the year; that blur is accepted, not fudged.
黄金全天候交易，但它在一天里并不是同一种交易。普查把每一天切成六个固定 UTC 窗口——隔夜、亚洲、伦敦、重叠、纽约、盘后——并给每个时段做了一套完整普查：它的解剖（大小、形状、极值何时形成）、它的内部行为（它第一小时的突破和假突破会怎样）、以及它的交接（下一时段如何对待本时段的方向和区间）。固定 UTC 意味着夏令时切换会让边界在一年中部分时间模糊一小时；这种模糊被接受，不作掩饰。
A reading note for this chapter: handoff and within-session behaviours are measured to the next session's close, so effects are a session long, not two hours. And anatomy is not a finding — the shape of a session is a fact to plan around, not an edge to trade.
The 24-hour volatility curve — median M15 bar range by hour (UTC), two corpora24 小时波动曲线——按小时（UTC）的 M15 K线幅度中位数，两个语料
Peak: 13:00–14:00 UTC (≈ $10.2 median bar) — the US data window inside the Overlap. The secondary 01:00 UTC peak is the Overnight session's final-hour swell (see Overnight, below). The quietest hours are the Off-Hours seam and mid-Asia.
峰值：13:00–14:00 UTC（K线中位数约 $10.2）——重叠时段内的美国数据窗口。01:00 UTC 的次峰是隔夜时段最后一小时的膨胀（见下文隔夜部分）。最安静的时段是盘后接缝和亚洲中段。
Session anatomy — median session range, trusted corpus (US$ and ATR)时段解剖——时段区间中位数，可信语料（美元与 ATR）
The Overlap is the biggest session ($44.9, 5.19 ATR); Off-Hours is barely a session at all ($17.0). London, folklore's opener, is the smallest of the four big windows in this sample — Asia's range matches it. New York's dollar range ($41.2) is large but its ATR range (3.52) is middling: by NY hours, the day's ATR yardstick has already stretched.
Fifteen adjacent pairs were measured (each session's completed direction → the next session's move, close-anchored). The map, honestly coloured:
共测量了十五对相邻交接（每一时段已完成的方向 → 下一时段的运动，以收盘锚定）。地图如实着色：
Handoff交接
Verdict判定
What the census found普查发现
Overnight → Asia
CONFIRMED已确认
Continuation, both corpora; strongest for mid-range overnight closes (+0.58 / +0.51).
延续，两个语料；隔夜收在区间中部时最强（+0.58 / +0.51）。
Asia → London
NOT ROBUST不稳健
Reversal confirmed on the trusted year, DEAD on the extended corpus. Not bankable — see London.
在可信年份确认反转，在扩展语料上已死。不可靠——见伦敦部分。
London → Overlap
CONFIRMED已确认
Continuation — the most reliable handoff in the book: raw median +0.42 ATR on BOTH corpora.
延续——全书最可靠的交接：两个语料的原始中位数都是 +0.42 ATR。
Overlap → New York
SPLIT分裂
Pooled: dead. Split by data-day proxy: continues on data days (+0.18 / +0.56), reverses on quiet days. See New York.
汇总：已死。按数据日代理拆分：数据日延续（+0.18 / +0.56），安静日反转。见纽约部分。
New York → Off-Hours
CONFIRMED · NEGLIGIBLE确认 · 可忽略
Continuation by +0.01–0.11 ATR — economically almost nothing.
以 +0.01–0.11 ATR 延续——经济上几乎为零。
Off-Hours → Overnight
CONFIRMED — NEGATIVE已确认 — 负向
Reversal, down-side driven: after down off-hours closes, overnight recovers (bottom-third closes −0.98 / −0.90 continuation-signed).
反转，由下行驱动：盘后收跌后，隔夜回升（底部三分位收盘延续签署 −0.98 / −0.90）。
Overnight (22:00–01:30 UTC)隔夜（22:00–01:30 UTC）
The session folklore ignores turns out to have the book's most obedient internals. Its first hour's direction continues to the session close (raw median +0.32 ATR; CONFIRMED on both corpora, +0.44 / +0.38 and +0.50 / +0.25 on the accepted-break version). Its own accepted breaks continue; its fakeouts are the exception — first-hour rejections claimed continuation in batch terms but flipped sign across corpora and are buried in the Graveyard.
Anatomy: the high forms late (52% of instances make their high in the final hour) while the low forms early (28% in the first 30 minutes), and median bar range swells to $13.0 in the +120–150 minute bucket — this is batch 2's unexplained 01:00 UTC volatility peak, identified: it is the overnight session winding up into the Asia open, not a mystery event. Two more confirmed facts point into this session: accepted New York closes get reversed here (−0.26 / −0.81 trusted, −0.53 / −1.02 extended — both sides), as do accepted Overlap closes (−0.27 / −0.94). Overnight is where the US day's leftovers go to be undone. Fading overnight sweeps during London, however, lost (WEAK-negative −0.27 / −0.50) — do not fade this session's range from the next one.
Dukascopy's feed has a daily maintenance hour, and it rotates with US daylight saving: on the trusted corpus the absent hour is 21:00 UTC in EST months (83 days) and 22:00 UTC in EDT months (177 days); on the extended corpus, 158 vs 278 days. Crossing the break is not a gap event — the step across it has a median of $3.77 versus $3.42 for ordinary steps — but the first bar after the break is the seam's widest ($10.2 median range): stale books re-price at once, then calm returns.
Asia is not the sleepy stand-in of the lore. Its median range ($36.7) matches London's in this sample — it just travels less efficiently (path efficiency 0.44). Its accepted breaks continue to its own close on both corpora (+0.12 / +0.38; +0.25 / +0.18; raw +0.18 ATR), and it faithfully continues the Overnight session's direction (see the map). Its extremes form at both ends of the session (25% of highs/lows in the first 30 minutes; the final bucket makes 20%/17%), volatility sags to $5.5 mid-session and rebuilds to $9.7 by 05:30–06:00. Some 58% of the Asia range forms inside the overnight range, yet Asia closes outside that range slightly more often than not.
Asia's width is information: terciles of Asia width predict the whole day's range in steps (0.77 → 0.86 → 1.21 ATR). Counter-intuitively, narrow-Asia days close outside the Asia range 75% of the time with the largest range multiple (3.3× the Asia range, vs 2.0× for wide-Asia days) — a narrow Asia is unspent fuel, not a cap on the day. What Asia does not do: its first-hour breakouts are a coin flip (rejection share 49%/48%, aftermath ≈ 0 on both corpora), and its range as a level-set for the next overnight session is dead. Fading New York sweep-rejections during Asia confirmed on the extended corpus but died on the trusted one — a corpus conflict, recorded, not bankable.
Asia width terciles vs the full day's range and close location (descriptive anatomy).
亚洲宽度三分位与全天区间及收盘位置（描述性解剖）。
London (07:00–12:30 UTC)伦敦（07:00–12:30 UTC）
London's chapter is the book's standing lesson in corpus humility. On the trusted year, a beautiful story confirmed: London's accepted breaks reversed (faded) and its rejections continued, Asia's direction reversed into London, London was the fakeout session. On the extended corpus, that story largely died — accepted breaks still fade, but only just (−0.29 / −0.15, raw −0.13; −0.21 on the trusted year too), rejections lean continuation (+0.05 / +0.26), and Asia→London reversal is DEAD. What survives both corpora, robustly: London's first half leans against itself by the session's end (WEAK-negative, −0.08 / −0.14 ⚠ discovery-zero), and London's accepted closes get continued into both the Overlap (+0.20 / +0.77) and New York (+0.16 / +0.16) — the day belongs to whoever inherits London, in London's direction.
Anatomy explains the fade-the-open instinct without endorsing the trade: London has the flattest volatility curve of any session ($6.6–8.2 in every bucket — no explosion at the open), its extremes form early and late (24%/22% in the first bucket), and at 36% RANGE it is the most range-like of the big sessions. London's range, however, is a dead level-set for the overnight session, and its first-hour breakout complex (the Sb family) remains corpus-split — confirmed-fade on extended, confirmed-negative on trusted — which is precisely why the Approaches tab treats 'fade London's first hour' as a visual concept standing on a confirmed session-shape fact, not on a confirmed trade.
Accepted breaks of London's first-hour range: faded on the extended corpus (left/top stack), the batch-4 session-census version.
伦敦第一小时区间的接受突破：在扩展语料上被淡（叠加图），第四批时段普查版本。
Overlap (12:30–16:30 UTC)重叠（12:30–16:30 UTC）
The Overlap — London and New York open together, and in summer the 08:30 ET data lands here — is gold's main event: the largest median range ($44.9, 5.19 ATR), volatility peaking at $14.0 in the 13:30–14:00 bucket, extremes forming late (27%/27% in the final 30 minutes). Its confirmed behaviours are the chapter's strongest set:
It continues London. The London→Overlap handoff is the most reliable in the book (raw +0.42 ATR on both corpora).
Its first half leans against itself into the close, on both corpora (−0.07 / −0.05; −0.30 / −0.18; raw −0.15 / −0.13) — early Overlap thrusts are spent by the end of the window.
It hosts the snapback. Accepted breaks below the London low inside this window reverse (+1.14 ATR raw median in the cell) — the Mean Reversion chapter's one surviving cell lives here, and by the New York window it is gone.
It undoes its own accepted closes later: the Overlap's accepted direction is reversed by the Overnight session on the trusted corpus (−0.27 / −0.94).
Two warnings complete the portrait. The Overlap's own first-hour breakouts are WEAK-to-DEAD — the biggest session does not reward chasing its own open either. And fading Asia sweep-rejections inside the Overlap lost heavily (CONFIRMED negative, −0.01 ⚠ / −0.89): levels from the quiet session do not get respected in the loud one.
The London→Overlap handoff: stacked paths of the Overlap session after London closes, direction-signed.
伦敦→重叠交接：伦敦收盘后重叠时段的叠加路径，按方向签署。
New York (16:30–21:00 UTC)纽约（16:30–21:00 UTC）
New York is the mirror session: its extremes form early (33%/33% in the first 30 minutes — exactly where the Overlap forms its late), and its volatility decays steadily ($12.4 down to $6.7) as the day winds down. Behaviours pointing into it are contrarian: accepted Asia closes reverse here on both corpora (−0.30 / −0.26, symmetric both sides), accepted Overnight closes reverse here too (−0.24 / −0.13), and the Overlap's direction leans reversal by NY's close on the extended corpus (−0.08 / −0.15; WEAK on trusted). Only London's direction survives the trip (+0.16 / +0.16).
Pooled, the Overlap→New York handoff is a coin flip (≈49% continuation share) — dead. But batch 4 split the instances using each corpus's own release-bucket volatility as a rough data-day proxy, and the pooled zero split in two: on proxy data days, New York's close continued the Overlap's direction (median +0.18 / +0.56 across the halves); on no-data days it reversed (−0.16 / −0.14). Continuation into the NY close looks like a data-day phenomenon. This is a proxy split, labelled as such — the real-calendar version belongs to the Macro chapter's engine, which found the intraday yield-link invisible at 2h. Both facts stand; neither inflates the other.
New York's own first hour rewards nobody: its breakouts are DEAD on both corpora, and its fakeouts confirmed on the trusted year but died on the extended — the session's tradable content is what it does to other sessions' moves, not its own open.
New York's treatment of accepted Asia closes: reversal on both corpora.
纽约对被接受的亚洲收盘的态度：两个语料上都是反转。
Off-Hours (21:00–22:00 UTC)盘后（21:00–22:00 UTC）
The seam hour: smallest window ($17.0 median range, 1.89 ATR), extremes concentrated in its first bucket (31%/38%) and again at the 22:00 reopen (26%/17%, with a $7.1 volatility bump as books refresh). It is also the least 'complete' session — only 77%/65% of instances ever touch one of its first-hour extremes, versus 92–98% elsewhere. Its one confirmed handoff is a lesson in statistical honesty: New York→Off-Hours continuation confirmed on the extended corpus at +0.01 / +0.09 ATR with a raw median of +0.11 / +0.02 — a confirmation that means almost nothing economically, kept in the book precisely to show that 'confirmed' and 'matters' are different words. Its accepted breaks confirmed trivially small on the trusted corpus (+0.09 / +0.08) and died on the extended. One disclosed scope gap: the S-d cross-session pairs were not run for Off-Hours — the chapter says so rather than implying coverage.
Family F · 560 scheduled events, 2016–2026 + intraday 2025–26
The desk's event calendar holds 560 scheduled US releases and FOMC decisions across eleven years — every CPI, NFP, PPI and Retail Sales print at 08:30 ET, every FOMC statement at 14:00 ET — assembled from BLS and Census release calendars (including archived captures for the 2025–26 government-shutdown months), Federal Reserve statement PDFs, and FRED macro series. This chapter answers, with counts, the questions traders argue about every month.
交易台的事件日历收录了十一年间 560 次预定的美国数据发布与 FOMC 决议——每一次 08:30 ET 的 CPI、非农、PPI 和零售销售，每一次 14:00 ET 的 FOMC 声明——来源包括 BLS 和人口普查局的发布日历（含 2025–26 政府停摆月份的存档抓取）、美联储声明 PDF 以及 FRED 宏观序列。本章用计数回答交易者每月争论的问题。
The first-spike answer第一波冲高的答案
The first 15-minute spike is faded最初 15 分钟的冲高会被淡
CONFIRMED已确认
Pool every 08:30 ET release (2025–26 intraday corpus, n=76): the spike — the extreme of the first 15-minute bar — is more often the end of the move than the start. Two hours later, price sits on the opposite side of the pre-release price relative to the spike's direction, by a median of −0.61 ATR (discovery) / −0.35 (holdout) versus a session-matched null. The follow-rate — the share of spikes whose side price still held after 2h — is 43%. The median spike is $7.1 (0.97 ATR); the median CPI spike is $11.7 (1.23 ATR).
Frequency: ≈ 3.7 releases per month · Dukascopy M15 · n = 76 pooled releases频率：每月约 3.7 次发布 · Dukascopy M15 · n = 76 次汇总发布
Stacked paths after 08:30 ET releases, signed by the first bar's spike direction, ATR units. Release at 0. Dukascopy M15, n=76.
08:30 ET 发布之后的叠加路径，按第一根K线冲高方向签署，ATR 单位。发布在 0 点。Dukascopy M15，n=76。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
Archetypes: pooled instances listed in the census file; the replay below is a CPI instance (12 Aug 2026).
原型：汇总实例列于普查文件；下方回放是一个 CPI 实例（2026 年 8 月 12 日）。
Honest limits诚实边界
Split by release type before quoting: CPI follow-rate 39% (median −0.35 ATR, −$3.4) and PPI 33% (−0.59, −$3.9) drive the pooled result and are individually only WEAK (holdout n of 6–8). NFP and Retail Sales are coin flips (50%, +0.15 / +0.11). The FOMC 14:00 window is a separate, unstable cell (n=14, DEAD) — its large median is not quotable. And nothing here says the fade is tradeable: the spike extreme is only knowable after the first bar closes.
引用前先按发布类型拆分：CPI 跟随率 39%（中位数 −0.35 ATR，−$3.4）和 PPI 33%（−0.59，−$3.9）驱动了汇总结果，但各自只是弱（留出 n 只有 6–8）。非农和零售销售是抛硬币（50%，+0.15 / +0.11）。FOMC 14:00 窗口是一个独立的、不稳定的格（n=14，已死）——它的大中位数不可引用。而且这里没有任何一句说反打可以交易：冲高极值只有在第一根K线收盘后才可知。
Release (08:30 ET)发布（08:30 ET）
Spike followed at 2h2 小时后冲高被跟随
Median signed 2h move2 小时签署运动中位数
Median spike冲高中位数
CPI
39%
−0.35 ATR · −$3.4
$11.7 · 1.23 ATR
PPI
33%
−0.59 ATR · −$3.9
$3.8 · 0.46 ATR
NFP
50%
+0.15 ATR · +$0.6
$16.3 · 1.92 ATR
Retail Sales
50%
+0.11 ATR · +$0.7
$5.7 · 0.87 ATR
FOMC (14:00 ET)
57%
unstable, n=14 — DEAD, do not quote
不稳定，n=14——已死，不可引用
—
The release minute is loud but the loudness is scheduled: M15 range runs at 0.86–0.98× the session baseline in the two hours before a release (the market holds its breath), the release bucket itself runs at 1.47× (CPI 1.94×), and the three hours after stay elevated at 1.12–1.32×. Event days are wide but not directional — range/ATR versus the 0.85 ordinary-day baseline: FOMC 1.15, NFP 1.05, CPI 0.96, while median |return|/ATR sits at 0.40 on event days versus 0.41–0.45 on baseline days, and FOMC is actually below baseline (0.28). The widest days carry the smallest net moves: two-sided chop, not journeys.
M15 range around 08:30 ET releases as a multiple of session baseline: quiet before, 1.47× at the release, elevated for hours after.
08:30 ET 发布前后的 M15 幅度（时段基线的倍数）：之前安静，发布时 1.47 倍，其后数小时保持高位。
Watch a CPI release看一次 CPI 发布
A real release, replayed bar by bar from the census corpus: CPI, 12 Aug 2026, 12:30 UTC. Watch what the first bar does — then watch what the next two hours do to it.
Frozen at the close of the release bar: CPI, 13 Jan 2026. The first bar spiked up. The census base rate says a CPI spike is still being followed two hours later only 39% of the time. Your call — then reveal.
Faded, like the majority: the signed 2h move was −0.37 ATR (−$3.1) — two hours after an upward spike, price sat below where the release found it. This instance was chosen because it is the closest to the CPI census median (−0.35 ATR): not a horror story, not a miracle — the typical one.
Now zoom out to the daily scale (2016–2026, all five event types). After up event-days, gold's raw median five days later is +0.22 … +0.52 ATR (event-direction signed). After down event-days, it is +0.47 … +0.56 raw — a bounce. Put together: in this decade, gold drifted up in the week after scheduled red-letter days on both sides of the print. The signed-relative verdicts vary by type (NFP CONFIRMED with a mixture caveat; FOMC, CPI, Retail WEAK; PPI DEAD) because the null also drifts — but the raw medians are uniform, and they are why 'short the bad news' kept failing its census cells (see the honesty box).
Stacked daily paths after NFP event-days, event-direction-signed, ATR units. Event day at 0. GC=F daily, n=115. Note the mixture: the mean runs ahead of the median — a minority of large continuations pulls it.
Descriptive anatomy, daily scale: on days when the US 10-year yield moved at least 5bp, gold closed opposite the yield's direction 73.5% of the time (n=211; on CPI days, 78%, n=50) — against a 56.5% opposite-rate on quiet no-event days. The buckets are wide: 10Y up ≥5bp → gold's median day −0.36% (−$5.4); 10Y flat → +0.08%; 10Y down ≥5bp → +0.57% (+$8.2). The broad-dollar buckets are wider still (−0.73% / +0.08% / +0.56%). If you watch one number on a data day, the census says watch the 10-year's close, not the headline print.
But the intraday version of the same link is invisible in this dataset: conditioning the 2h post-release move on the dollar's 2h direction gives an inverse rate of 48.7% (n=76) and median gold moves of ≈ $0 on both sides (dollar up: +$0.76; dollar down: −$0.85). H1 closes are coarse, and 2h may simply be too short for the conditioning to bind. Both facts are printed because both are true: the yield-conditioning is a daily-scale anatomy fact, not a 2-hour trading signal. Nothing in this section is a verdict — it is labelled descriptive, and the confirmatory version needs a real intraday yields feed this desk does not have.
Gold's event-day outcome conditioned on the 10-year yield's event-day move (daily scale, descriptive).
以 10 年期收益率事件日运动为条件的黄金事件日结果（日线尺度，描述性）。
The hot/cold honesty box冷热诚实框
CRUDE PROXY — read before quoting粗糙代理——引用前必读
This desk could not obtain a verifiable free historical consensus series, so 'hot' and 'cold' here mean: the released month-over-month change versus the prior month's change, computed on revised FRED series. This is NOT a surprise measure — the consensus may already have expected the change. Every hot/cold number below carries that defect. One cell of this family (the 'cold' side) produced a letter-confirmed result whose discovery half was ≈ 0 in ATR and negative in dollars; it is withheld from this book entirely rather than printed with an asterisk.
本交易台无法获得可验证的免费历史共识序列，所以这里的「热」和「冷」指的是：发布的环比变化相对前一个月的变化，用修订后的 FRED 序列计算。这不是意外度量——共识可能早已预期了这个变化。下面每一个冷热数字都带着这个缺陷。这个家族有一格（「冷」的一侧）产生了字面确认的结果，但其发现半区在 ATR 上 ≈ 0、美元上为负；它被从本书中整体扣留，而不是加个星号印出来。
Shorting 'hot' prints loses做空「热」数据会亏
CONFIRMED — NEGATIVE已确认 — 负向
Under the crude proxy, shorting gold after prints hotter than the prior month (the naive 'strong data, sell gold' reflex) lost in both halves versus the signed null (−0.18 / −0.10 ATR; dollar medians −$2.05 / −$6.40), CONFIRMED negative. Gold's raw median after hot prints is up — the same post-event upward drift measured in the previous section. The CPI-only version is WEAK (−0.01 / −0.31).
Frequency: ≈ 4.2 'hot' prints per month (proxy) · GC=F daily + FRED · n = 508频率：每月约 4.2 次「热」数据（代理）· GC=F 日线 + FRED · n = 508
Stacked daily paths after 'hot' prints (crude proxy — see honesty box), fade-signed.
「热」数据（粗糙代理——见诚实框）之后的叠加日线路径，按反打签署。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
Archetypes live in the census file for this behaviour.
原型实例存放在该行为的普查文件中。
Honest limits诚实边界
Event-day returns themselves barely separate hot from cold (median +$0.9 vs +$1.7) — the proxy's information, such as it is, lives in the following days' drift, and the proxy may simply be re-measuring 'gold rose this decade'. A real consensus-surprise series is the single highest-value data upgrade named in How This Book Was Made.
Family G · GC=F daily + FRED yields & broad dollar + ICE DXY
The shortest chapter in the book, and deliberately so. The lore says gold is a puppet of the dollar and real yields — spike the dollar, gold falls; spike yields, gold falls. The census tested the state version of that claim: after days when the dollar or the 10-year yield spiked by ≥1.5× its own ATR, what does gold do next?
Dollar-spike days: DEAD, both measures. Broad-dollar version: −0.49 / −0.02 (collapses to nothing). True ICE DXY on its available 2-year window: −0.09 / +0.37 — a sign flip. There is no confirmed 'fade gold after a dollar spike' behaviour.
Yield-spike days: WEAK, negative (−0.14 / −0.06) — leaning toward 'gold keeps softening after yield spikes', far from proven.
The honest anatomy underneath: after ±spike days in either asset, gold's raw next-5d median is up +0.2 … +0.3 ATR on both sides — spike days in gold's neighbours mark local exhaustion in this sample, but no side of it survived as a behaviour.
Stacked paths after 10-year yield spike days (WEAK, negative). FRED DGS10 changes vs GC=F.
10 年期收益率冲高日之后的叠加路径（弱，负向）。FRED DGS10 变化对 GC=F。
Two labelling notes that matter. First, the 'dollar' in the long daily runs is FRED's broad trade-weighted dollar index (DTWEXBGS), not the ICE DXY of trading lore — it is labelled that way everywhere in the census, and the true-DXY check above is why. Second, the quiet control: on quiet cross-asset days (neither asset spiked), gold's 5-day median is +0.19 ATR versus +0.28 for random days — quiet neighbours mean a quieter gold in every sense, drift included. That baseline, not a spike rule, is what this chapter actually establishes.
Cross-learning addendum — what EUR/USD taught gold (POSEIDON Vol. II, 2026-10-15)跨市场学习附记——欧元/美元教给黄金的事（POSEIDON 第二卷，2026-10-15）
CROSS-LEARNING跨市场
Three measured results from the EUR/USD census (Volume II) belong in this chapter. (1) The dollar-direction filter is dead. Gold's accepted prior-day breaks (n=323) were split by EUR/USD's state at the event: the apparent "tailwind" advantage flips sign between discovery (+0.25 ATR) and holdout (−0.17 ATR), and gold's mid-range engulfing (n=1,997) performs identically with the dollar aligned or opposed (+0.028 vs +0.025 ATR). Filtering gold's confirmed behaviours by the concurrent EUR/USD direction is not supported — the desk's old DXY-filter question is closed, negatively. (2) The two markets share the dollar impulse simultaneously: same-session returns correlate most in the Overlap (r = 0.50) and New York (0.48), least late (0.21); no lead-lag pair reaches |r| ≥ 0.08, so neither market warns the other session-to-session. (3) At 08:30 ET releases the first-spike signs agree 82% of the time (n=79) — and both markets then fade the spike over 2h. "Never chase the first spike" is now confirmed from both sides of the pair. Full log: POSEIDON Vol. II cross-learning file (X/), 2026-10-15.
Cross-learning addendum — what USD/JPY taught gold (POSEIDON Vol. III, 2026-10-16)跨市场学习附记——美元/日元教给黄金的事（POSEIDON 第三卷，2026-10-16）
CROSS-LEARNING跨市场
The yen-state filter is dead too (XJ1). Gold's accepted prior-day breaks (n=323) split by USD/JPY's prior-2h state: the 2h continuation is present in BOTH yen states in the holdout — +0.40 ATR when the yen was bid, +0.42 when it was offered. The split separates nothing, and gold's mid-range engulfing splits flip between the halves. With Volume II's EUR/USD-state filter also dead, conditioning gold's confirmed behaviours on the concurrent FX majors is closed negatively, twice measured.
The haven mirror (XJ2). Same-session gold vs USD/JPY returns correlate negatively in every session — Overlap −0.49, New York −0.48, London −0.30 (n ≈ 485 days) — the mirror image of gold × EUR/USD (+0.50 Overlap). No lead-lag pair reaches |r| ≥ 0.10.
Event twins (XJ4). On 79 matched 08:30 ET releases, gold and USD/JPY first spikes are dollar-consistent (opposite raw signs) 81% of the time, and both markets then fade the spike over two hours. Gold's event-chapter rule — never chase the first spike — is now confirmed from both sides of the dollar.
事件双胞胎（XJ4）。在 79 个匹配的 08:30 ET 发布中，黄金与美元/日元的第一波有 81% 的时间是美元一致的（原始符号相反），且两个市场随后两小时都回吐第一波。黄金事件章节的规则——永不追第一波——现已从美元的两侧都得到确认。
Cross-learning addendum — what NAS100 taught gold (POSEIDON Vol. IV, 2026-10-16)跨市场学习附记——纳斯达克100教给黄金的事（POSEIDON 第四卷，2026-10-16）
CROSS-LEARNING跨市场
The NAS-state filter is dead too (XN1). Gold's accepted prior-day breaks split by NAS100's prior cash-day direction continue in BOTH states in the holdout (+0.16 ATR after NAS up-days, +0.08 after down-days), and gold's mid-range engulfing is near-identical across NAS states (+0.08 vs +0.16). The concurrent-2h NAS states are too rare during gold events (n = 7–10) to condition on at all. The state-filter question is now closed three times over: EUR/USD state (Vol. II), yen state (Vol. III), NAS state (Vol. IV) — none of them conditions gold's confirmed behaviours.
Risk co-movement (XN3). Gold's return over the exact NAS cash window (DST-aware per date) correlates +0.23 with the index's cash return (n=484 days) — a mild risk-on/off co-movement, far weaker than the dollar channels measured in Vols. II–III (|r| ≈ 0.5). NAS cash → gold next-2h: r = +0.04 — no lead.
The mirror (XN2). The index's own accepted breaks, split by gold's state, flip between discovery and holdout: gold's state does not condition the index's breaks either.
NAS 状态过滤器也已否定（XN1）。黄金被接受的前日突破按纳斯达克100 此前现金日方向拆分，在留出期里于两种状态下都延续（NAS 上涨日后 +0.16 ATR，下跌日后 +0.08），黄金的中位吞没在两种 NAS 状态下几乎相同（+0.08 对 +0.16）。黄金事件期间同时出现的 2 小时 NAS 状态太稀少（n=7–10），根本无法据此条件化。状态过滤问题现已三次关闭：欧元/美元状态（第二卷）、日元状态（第三卷）、NAS 状态（第四卷）——没有一个能条件化黄金的已确认行为。
风险共同运动（XN3）。黄金在 NAS 现金窗口（按日期感知夏令时）内的收益与指数的现金收益相关性为 +0.23（n=484 天）——温和的风险开/关共同运动，远弱于第二、三卷测得的美元通道（|r|≈0.5）。NAS 现金 → 黄金随后 2 小时：r=+0.04——没有领先。
This chapter is the census keeping its promise. A behaviour is DEAD when its two chronological halves disagree in sign, or its effect is indistinguishable from the random baseline. Dead is not a failure of the research — dead is the research working. Every entry below was a story a trader could tell at a bar with total conviction. Two of them were confirmed in batch 1 and killed by their own re-examination — they are listed first, because they are the most important corpses in the book.
NFP and Retail are coin flips; FOMC is an unstable n=14 cell. Only the pooled 08:30 fade confirmed.
非农和零售是抛硬币；FOMC 是不稳定的 n=14 格。只有汇总的 08:30 反打得到确认。
F6_cold_cpi
'Cold' CPI prints (crude proxy)
「冷」CPI 数据（粗糙代理）
Dead under the proxy. The pooled cold cell is withheld from this book (see Macro honesty box).
在代理下已死。汇总冷格被从本书扣留（见宏观诚实框）。
G1_dxy_spike
Fade gold after dollar spikes
美元冲高后反打黄金
Dead on the broad index, sign-flipped on true DXY. The puppet-string story has no census support.
在广义指数上已死，在真实 DXY 上符号翻转。提线木偶的故事没有普查支持。
Session Sb/Sd corpses
Overnight first-hour rejections; Asia range as overnight levels; London range as overnight levels; NY first-hour breakouts; Overlap first-hour breakouts (weak-to-dead)
Each died on sign flip or ≈ null on at least one corpus — details in the Sessions chapter and census files.
每一个都在至少一个语料上死于符号翻转或 ≈ 基线——细节见时段章与普查文件。
The pattern across the graveyard is worth more than any single stone: location stories died, mid-range stories lived; first-move stories died, drift stories lived; and every 'obvious' level trick that depended on one bull year was repossessed by the second corpus. If you take one methodological instinct from this book, take that one.
Methods, feeds, caveats — the fine print, in large print
The protocol, in order: (1) every behaviour's definition was written down before it was measured, in a definitions file that still sits beside the code; (2) the census took every instance — no examples were chosen by a human; (3) instances were stacked and compared against random-timestamp nulls matched on session and direction; (4) verdicts used a rule fixed in advance (same sign in both chronological halves; holdout ≥ 0.05 ATR and ≥ 50% of discovery; holdout n ≥ 30); (5) anything calling itself an effect is a median difference versus the null, in ATR units, gross of costs; (6) costs enter only at the strategy stage, which nothing in this book has reached.
Feeds. Intraday: Dukascopy XAU/USD M15 — a trusted 1-year file (24,277 bars, 2025-10-01 → 2026-10-12, 265 days) and an extended corpus (64,224 bars, 2025-01-02 → 2026-10-12, 441 valid days; overlap-verified against the trusted file: 24,266 of 24,277 bars within $0.05). Daily: GC=F futures, 2,514 trading days, Oct 2016 → Oct 2026. Yields: FRED DGS10/DGS2. Dollar: FRED broad index DTWEXBGS (not ICE DXY — see Cross-Asset), with a true-DXY intraday check on 2 years of H1.
Data hygiene. The extended corpus contained 169 flat weekend 'days' of stale Dukascopy quotes; they were detected by a range audit and excluded (441 valid days remain). Four of 560 calendar events had no bar at the exact release instant on the M15 file and are counted as drops. Event windows are computed strictly after the event bar — the chronological-integrity rule this desk adopted after catching a look-ahead error in its own earlier strategy work.
Multiple testing. Batches 1–3 measured ~70 behaviour units; batch 4 added 112 rows. At that scale, a 0.05-ATR threshold will confirm some noise by construction. The defence used throughout: trust cross-sample survival (the extended corpus, the strict re-definitions) more than any single verdict — it is why B2's three confirmations outrank prettier single results.
Known blind spots. Effects are medians of path distributions that are often mixtures (a still median can hide violent offsetting tails — the stack charts show the bands for this reason). The intraday samples are one to two years of a strong bull market; the daily sample is a decade of one. The hot/cold macro split uses a crude revised-series proxy, not consensus surprises. The Overlap→NY data-day split uses a volatility proxy, not the true calendar. Intraday yields are absent; the 2h yield-conditioning test used H1 dollar closes instead.
What would upgrade this book, in order of value: a real consensus-surprise history for the macro chapter; a second intraday feed (or a genuinely bearish gold year) to re-run the session census against; an intraday yields feed for the F7 question; and time — the cheapest validator there is. Until then, everything here is labelled exactly as strongly as it has earned.
Eight possible strategies, drawn. None of them validated.八个可能的策略，画出来了。没有一个被验证过。
8
visual concepts in this tab本页签有 8 个视觉概念
0
validated — every one is unproven已验证 0 个——每一个都未经证实
≥50
trades + PF ≥ 1.3 out-of-sample: the bar to promotion交易 ≥50 笔 + 样本外 PF ≥ 1.3：晋升的门槛
2
market periods required before anything counts任何东西算数之前需要 2 个市场时期
Read the label on every card:读每张卡上的标签：“Possible strategy — visual concept, unvalidated.” Promotion is earned in the census and the backtest, never on this page.「可能的策略——视觉概念，未经验证。」晋升要在普查和回测中挣得，永远不在这一页。
Approaches打法
Visual concepts drawn from Tab 2 behaviours — nothing here has been backtested as a strategy
This tab is a sketchbook, not a system. Each entry takes one behaviour the census confirmed and draws what a trade built on it would look like: what to watch, what confirmation looks like, where the idea is visibly wrong, and — from the census's own dead and opposite cells — how it fails. There are deliberately no calculations here: no entries, stops, targets or profit claims. Anything that ever leaves this tab must first survive the desk's full readiness process (mechanical definition, costed backtest, ≥50 trades, PF ≥ 1.3 out-of-sample, two market periods). Until then, every entry wears the same label, and the label is the truth:
The release bar prints a violent extreme in its first 15 minutes. The concept is patience as a picture: mark the pre-release price, mark the spike extreme, and watch whether the bars that follow hold the spike's side — or leak back through the pre-release price. The census shape (Tab 2) is a median path that ends on the other side.
CPI, 11 Mar 2026 (extended corpus): the release bar spiked DOWN to $5,172.78 and closed at $5,180.07. The concept watches whether the bars after hold the spike's side. Real bars; annotation is the concept's own drawing.
The spike bar closes, and the next bar fails to extend beyond the spike extreme.
Price trades back through the pre-release price within the first hour.
The release is CPI or PPI — the two types whose spikes the census found faded (follow-rates 39% and 33%).
冲高K线收盘，下一根K线无法扩展到冲高极值之外。
价格在一小时内交易回发布前价格的另一侧。
发布是 CPI 或 PPI——普查发现其冲高被淡的两种类型（跟随率 39% 和 33%）。
Invalidation — where the idea is wrong失效——想法错在哪里
A second bar closes beyond the spike extreme and the following bars hold that side — the spike has become a breakout, and the fade picture is wrong. Standing aside is also a complete version of this concept: it forbids chasing, it does not command fading.
08:30 ET releases, which land inside the Overlap (12:30 UTC winter / summer shifts apply). The pooled census result covers CPI, PPI, NFP and Retail Sales together; the per-type picture differs sharply — see the failures.
08:30 ET 发布，落在重叠时段内（12:30 UTC，冬夏季有偏移）。汇总普查结果涵盖 CPI、PPI、非农和零售销售；按类型的图景差异很大——见失败部分。
How it fails — from the census itself它如何失败——来自普查本身
NFP and Retail Sales spikes are coin flips in the census (50% follow-rate) — the pooled fade does not extend to them with the same confidence. The FOMC window is an unstable n=14 cell (DEAD). And the spike extreme is only knowable once the first bar closes: acting during the spike bar is a different act that the census never measured.
London draws a range in its first hour and then, characteristically, pokes out of it. The concept treats that first-hour range as the session's reference box: a later break that closes outside is a candidate to drift back, because London's own shape — the flattest volatility curve of any session, extremes early and late — is a range's shape, not a launchpad's.
Stacked paths: accepted breaks of London's first-hour range, extended corpus — the median fades back. Census chart; the concept drawn from it is the range box above.
叠加路径：伦敦第一小时区间的接受突破，扩展语料——中位数淡回。普查图；概念的区间框画法源于此。
Confirmation signs确认迹象
The break closes outside the first-hour range but the next bars stall — no extension, shrinking ranges.
Price re-enters the first-hour range within the session.
The session is otherwise quiet: no 08:30-class catalyst scheduled inside London hours.
突破收在第一小时区间之外，但接下来的K线停滞——没有扩展，幅度收缩。
价格在时段内重新进入第一小时区间。
时段内其余时间安静：伦敦时段内没有 08:30 级别的催化剂安排。
Invalidation — where the idea is wrong失效——想法错在哪里
Consecutive closes outside the range with expanding bars — London has actually left, and the Overlap's inheritance (below) can run London's direction hard. The fade picture is wrong the moment the break starts travelling.
London session, 07:00–12:30 UTC. Honest context: this is the book's most corpus-split behaviour — the fade confirmed on the extended corpus and the batch-1 trusted run, while the tradable version scored negative on the trusted corpus in batch 4. The concept stands on the confirmed session-shape facts, and it is the weakest-evidenced entry in this tab.
How it fails — from the census itself它如何失败——来自普查本身
London's accepted direction is continued into the Overlap (+0.20 / +0.77) and into New York (+0.16 / +0.16) — the book's most reliable handoff runs with London, so a fade held into 12:30 UTC fights the strongest current in the census. London's rejections also leaned continuation on the extended corpus — the fakeout half of the folklore did not survive either.
伦敦被接受的方向被延续进重叠（+0.20 / +0.77）和纽约（+0.16 / +0.16）——全书最可靠的交接是顺着伦敦的，所以把反打持到 12:30 UTC 是在对抗普查中最强的水流。伦敦的拒绝在扩展语料上也倾向延续——民俗中假突破的那一半同样没有存活。
3Overlap snapback below the London low伦敦低点下方的重叠弹回
Possible strategy — visual concept, unvalidated可能的策略 — 视觉概念，未经验证
Draw the completed London session's low as a line. Inside the Overlap, watch for a bar that closes below it — an accepted downside break, in the day's loudest window. The concept watches for the break to fail: bars that stop making new lows under maximum volume, then closes back above the line.
22 Jun 2026 (extended corpus): the event bar closed at $4,186.78, below the London low of $4,188.40 (red line; London high $4,215.01 above). Real bars; lines drawn from the census's own London range.
A close below the London low inside 12:30–16:30 UTC (a wick is not enough — see failures).
Following bars fail to extend lower; a close back above the London low (the reclaim).
The move down into the break looked climactic — the Overlap's volatility peak doing the pushing.
在 12:30–16:30 UTC 内收在伦敦低点下方（影线不够——见失败部分）。
随后的K线无法向下扩展；一根收回伦敦低点之上的K线（收复）。
跌入突破的走势看起来像高潮——是重叠时段的波动峰值在推。
Invalidation — where the idea is wrong失效——想法错在哪里
Closes keep stacking below the line into the New York window. The census cell is 0.00 by NY — if the snapback has not come inside the Overlap, the concept's window has shut, whatever price does later.
Overlap window only (12:30–16:30 UTC), downside only. The pooled behaviour is ⚠ fragile: its discovery half is ≈ +0.08 and nearly all the strength is recent. The tradable cell held ≈ 63 instances in the census year — this is a rare picture, not a daily one.
How it fails — from the census itself它如何失败——来自普查本身
The mirror images fail, both of them measured: upside accepted breaks of the London high in US hours continued (−0.31 reversion-signed) — never mirror this concept upward. And downside rejections (wick below, close back inside) kept falling (−0.73 reversion-signed): the wick version of this picture is a trap, which is why the definition demands a close.
An engulfing bar in the middle of nowhere: not at yesterday's high or low, not at last week's extremes, not at a completed session's range edge — just a bar whose body swallows the previous bar's, in open field. The concept treats that bar as a small vote for its own direction over the next couple of hours, because that is the exact configuration the census confirmed three times.
1 Oct 2025, 17:30 UTC (trusted corpus): a bearish engulfing bar in mid-range — one of the census's archetype instances for its most-replicated behaviour. Real bars.
The engulfing body fully covers the prior bar's body, in the engulfing's direction.
Location check passes: the bar is clear of every listed level (the level-proximity check that killed the twin).
The next bar holds the engulfing direction rather than immediately retracing the whole body.
吞没实体完全覆盖前一根K线的实体，方向与吞没一致。
位置检查通过：该K线远离每一个所列价位（正是杀死其孪生体的价位邻近检查）。
下一根K线守住吞没方向，而不是立即回吐整个实体。
Invalidation — where the idea is wrong失效——想法错在哪里
Price closes back through the origin of the move — through the swallowed bar's body — the vote has been retaken by the other side. On the chart below, that is a close back above the engulfing bar's top region for this bearish instance.
Any session, M15 — the census ran it pooled across the day, and it is the most frequent confirmed behaviour in the book (≈ 80/month). Its median effect is a tilt (≈ +0.2 ATR), visible only across many instances: the Tab 2 predict card shows a genuine median instance that went the other way.
How it fails — from the census itself它如何失败——来自普查本身
The same shape at a level is DEAD — twice, under both the broad and strict level definitions; textbook location lore is this concept's failure mode. And per-instance, nearly half of these finish the next two hours against the engulfing direction: drawn as a single-trade promise, it fails about as often as it works. The census claim lives at population scale only.
Yesterday's high is a line the whole market can see. The concept watches the first close beyond it — and, per the census's distance gradient, prefers the close that lands just beyond the line (+0.24 raw median for near closes) over the dramatic overshoot that has already travelled (−0.15: spent).
17 Dec 2025, 13:30 UTC (trusted corpus): the event bar closed at $4,338.18, just beyond the prior day's high of $4,334.94 (line) — a 'near' close, the tercile the census found continues best. Real bars.
First close beyond the prior day's high, by a small margin.
The next bars hold outside yesterday's range — no close back inside.
The break arrives with the day's flow, not against a session that historically reverses it (see Sessions).
第一根收在前一日高点之外，幅度小。
接下来的K线守在昨日区间之外——没有收回内侧。
突破顺着当日的流，而不是逆着一个历史上会反转它的时段（见时段章）。
Invalidation — where the idea is wrong失效——想法错在哪里
A close back inside yesterday's range: acceptance has been withdrawn. The picture below shows the drawn line and the holding pattern it demands.
一根收回昨日区间内侧的K线：接受已被撤回。下图画出了那条线和它所要求的守住形态。
Session & context时段与语境
Up-breaks only. The census asymmetry is the concept: up-breaks' raw medians were +0.56 / +0.43 ATR on two corpora; down-breaks were −0.42 / −0.39 — in a sample where gold rose from $3,865 to $4,200. This concept is explicitly regime-coloured until a non-bull sample re-confirms it.
How it fails — from the census itself它如何失败——来自普查本身
Second breaks of the same level showed nothing (medians ≈ 0 to −0.23) — the concept is a first-close picture only. And do not promote it to week levels: at prior-week extremes the census found a magnet, not a launchpad (Approach 8). Down-breaks are the documented failure — drawn the other way, this exact picture lost to baseline on both corpora.
At 12:30 UTC, London hands the day to the Overlap. The concept is a relay picture: if London closed having actually gone somewhere (an accepted direction, not a range day), the Overlap's job, on the census's most reliable handoff, is to keep walking that way — raw median +0.42 ATR, the same number on both corpora.
London's close sits decisively outside its own session range's middle — a direction was accepted.
The Overlap's opening bars hold London's side rather than snapping back inside London's range.
A data release inside the Overlap is treated as fuel-or-fakeout per Approach 1 — the handoff does not license chasing the spike.
伦敦收盘果断地位于其时段区间中部之外——一个方向被接受了。
重叠的开局K线守住伦敦一侧，而不是弹回伦敦区间内。
重叠内的发布按打法 1 处理为燃料或假象——交接不授权追冲高。
Invalidation — where the idea is wrong失效——想法错在哪里
The Overlap closes back through London's range — the inheritance has been refused. Note the built-in tension the census also confirmed: the Overlap's first half leans against itself by its own close (−0.15 / −0.13 raw), so this relay is a picture of the session as a whole, not of its first hour.
The 12:30 UTC handoff, on days London moved. On the 36% of Londons that are RANGE days there is nothing to inherit — the concept is silent, and silence is the correct position. Continuation onward into the New York close is a separate, data-day-dependent question (Sessions chapter) — this picture ends at the Overlap's close.
12:30 UTC 交接，在伦敦有动作的日子。在 36% 属区间日的伦敦，没有什么可继承的——这个概念保持沉默，而沉默是正确的位置。继续延续到纽约收盘是一个独立的、依赖数据日的问题（时段章）——这幅图在重叠收盘时结束。
How it fails — from the census itself它如何失败——来自普查本身
The sibling handoffs are the failure gallery: Asia → London reversal died on the second corpus; Overnight's and Asia's directions get reversed by New York; the Off-Hours seam continues by an economically meaningless +0.01–0.11. This relay is the exception among handoffs, not the rule — which is exactly why it must not be generalised to the other five seams.
A monster day is unmissable: a range beyond 2× ATR, or a close more than 1.5 ATR from the open. The concept reframes the following days: expect the echo in size (the next five days run hot, |return| +0.19 / +0.18 vs baseline) but lean against the direction — the census's monster-day stack bends back, it does not extend.
The day qualifies mechanically: range > 2× ATR(14) or |close − open| > 1.5 ATR.
Follow-through attempts in the monster's direction stall within the next sessions.
For downside monsters: the reversal side is the documented one (raw −0.24 continuation-signed at 5 days).
这一天机械上合格：幅度 > 2× ATR(14) 或 |收盘 − 开盘| > 1.5 ATR。
顺巨兽方向的跟进尝试在接下来的时段内停滞。
对下行巨兽：反转一侧是有据可查的（5 天延续签署原始 −0.24）。
Invalidation — where the idea is wrong失效——想法错在哪里
A second monster day in the same direction: size persistence means clusters happen, and a fresh monster re-arms the move the concept was leaning against.
Daily scale, roughly ten qualifying days a year — a posture for the days after event-driven explosions (the Macro chapter's FOMC days are the widest and least directional of all), not a daily routine.
How it fails — from the census itself它如何失败——来自普查本身
Read the upside cell before leaning: upside monsters still drifted up in raw terms (+0.33 ATR) — merely less than their baseline. 'Lean against' an up-monster means expecting less drift, fading it outright is fighting a positive raw median. The cell is also small (n=95 across ten years): the size-persistence half of this picture is sturdier than the direction half.
This concept is defensive — a picture of what not to do. Price travels to last week's high and prints a rejection: a wick beyond, a close back inside. The folklore move is to short the rejection. The census move is to walk away: week highs are magnets, and price's documented habit is to come back up through them (fading the rejection faced a raw median of −0.16 ATR).
Jan 2024 (GC=F daily): the event day (2024-01-30) wicked to $2,048.50 — above the prior week's high of $2,039.30 (top line; prior week low $2,004.00 below) — and closed back inside at $2,031.50. A textbook magnet touch. Real bars.
First touch of the prior week's high or low on the daily chart, with a rejection close back inside the week's range.
Specifically at the week HIGH: the trap cell — the concept's whole content is refusing the fade-short there.
In the days after, price revisiting or exceeding the level is the expected picture, not a surprise.
日线图上首次触及上周高点或低点，拒绝收盘回到周区间内。
特别是在周高点：陷阱格——这个概念的全部内容就是拒绝在那里反打做空。
其后几天，价格重访或超过该价位是预期的图景，不是意外。
Invalidation — where the idea is wrong失效——想法错在哪里
The magnet framing itself would be wrong if an accepted daily close beyond the week level held and built — but note the census measured that branch too: accepted week breaks were WEAK-negative, and accepted breaks below the week low reversed. Both branches refuse the launch story; the magnet survives them. This concept's invalidation is therefore slow by design: it is a stance, and stances are revised by re-census, not by one day's close.
Daily chart, first touches of prior-week extremes — about twenty rejection instances a year. The sample is a decade-long bull market: 'up moves persist, down moves bounce' is partly the drift, and the honest reading is that week levels failed as launchpads in a bull decade. A bear decade gets its own census before this stance hardens.
How it fails — from the census itself它如何失败——来自普查本身
The failure mode of a defensive concept is opportunity cost, and the census prices it: the fade-short it forbids lost money (raw −0.16), but so did trading week breaks in either direction versus baseline — the magnet chapter's two cells are both negative, which is the rare case where 'do nothing at this level' is the measured answer rather than a shrug.
POSEIDON · Strategy families — the classics, gradedPOSEIDON · 策略家族——经典打法，被评级
Five classic ways to trade. One census to grade them.五种经典的交易方式。一本普查给它们打分。
5
strategy families, drawn step by step个策略家族，逐步画出来
4
markets graded against each family个市场对每个家族被评级
523
census behaviours behind the grades个普查行为支撑这些评级
0
approved strategies — concepts only, as everywhere in this book个已批准策略——只有概念，和本书其他地方一样
The label on this whole tab:整个页签的标签：“Possible strategy — visual concept, unvalidated.” These sections teach the classic families and grade their fit against the POSEIDON census. Where the census has no evidence, the card says so plainly: textbook concept — not census-tested.「可能的策略——视觉概念，未经验证。」这些部分教授经典家族，并对照 POSEIDON 普查评它们的适配度。普查没有证据的地方，卡片会直说：教科书概念——未经普查检验。
Strategy Families策略家族
The fit matrix first — then each family, drawn and graded先看适配矩阵——然后每个家族，画出来并被评级
The fit matrix — 5 families × 4 markets适配矩阵——5 个家族 × 4 个市场
Family家族
GOLD
EUR/USD
USD/JPY
NAS100
Reversals反转Trade the FAILURE of a move — at a level, after it over-extends.交易一次走势的失败——在价位处、在它过度延伸之后。
WORKABLE可用
STRONG FIT适配强
WORKABLE可用
STRONG FIT适配强
Breakouts突破Trade acceptance beyond a range — closes that hold outside.交易区间之外的被接受——收盘能守在外面。
STRONG FIT适配强
POOR FIT适配差
WORKABLE可用
STRONG FIT适配强
Trends趋势Ride structure at the DAILY horizon. Intraday pullback-buying is where this family goes to die.在日线尺度上骑乘结构。日内买回调是这个家族在本普查中死去的地方。
WORKABLE可用
WORKABLE可用
WORKABLE可用
STRONG FIT适配强
Price Action价格行为The reading layer itself: levels first — candles only count AT levels.阅读层本身：先看价位——K 线只在价位处才算数。
STRONG FIT适配强
WORKABLE可用
WORKABLE可用
WORKABLE可用
Order Flow订单流Who is forced to trade, and when — read through retail proxies, not a depth-of-market feed.谁被迫交易、何时被迫——用零售代理指标读，不用深度行情。
STRONG FIT适配强
WORKABLE可用
WORKABLE可用
WORKABLE可用
Grades are the census talking, not taste: STRONG FIT = the family’s core behaviour is CONFIRMED in that market; WORKABLE = real but small, horizon-bound, or spot-specific; POOR FIT = the census measured the family’s core move and it loses. Each family section below shows its receipts.评级是普查在说话，不是口味：适配强 = 该家族的核心行为在该市场被确认；可用 = 真实但小、受尺度限制、或只在特定位置成立；适配差 = 普查测过该家族的核心走势，而它亏钱。下面每个家族部分都会出示它的凭据。
1 · Reversals1 · 反转
Trading the failure of a move — at a level, after the stretch交易一次走势的失败——在价位处、在拉伸之后
1 · What it is1 · 它是什么
A reversal trade bets that a move has gone too far and will turn. You are not predicting the future — you are standing at a level where the move has a reason to fail: yesterday’s extreme, last week’s edge, the end of a three-day streak — and waiting for the failure to show itself.反转交易赌的是一次走势已经走得太远、将会掉头。你不是在预测未来——你站在一个走势有理由失败的价位旁：昨日的极值、上周的边缘、三天连涨连跌的尽头——然后等待失败自己显形。
The shape has three beats. Exhaustion: the move arrives stretched — fast, late, into a level everyone can see. The sweep: price pokes beyond the level, and the wick does the damage, collecting the stops parked there. The verdict: price closes back inside. Only the third beat is information. The first two are just weather.这个形态有三个节拍。耗尽：走势拉伸着抵达——快速、迟到、冲进一个人人都看得见的价位。扫盘：价格戳出价位之外，影线造成杀伤，收走停在那里的止损。裁决：价格收回内侧。只有第三个节拍是信息。前两个只是天气。
This is the hardest family to trade honestly, because the identical picture — price at an extreme, a wick beyond it — also appears at the start of genuine breakouts. The census exists to tell you in WHICH market the failure is the norm and in which it is the exception. The answer is not the same four times.这是最难诚实交易的家族，因为完全相同的画面——价格在极值处、一根影线戳出去——也出现在真正突破的起点。普查存在的意义，就是告诉你在哪个市场「失败」是常态、在哪个市场它是例外。这个答案四次都不一样。
Judged on closes, always. A reversal entered on a wick is a guess wearing a costume.永远用收盘判断。在影线上入场的反转，是穿了戏服的猜测。
2 · The anatomy — sweep, close back, retest2 · 解剖——扫盘、收回、复测
1 · The run-in1 · 冲向价位
The move arrives late and stretched, into a level the whole market can see. Nothing to do yet.走势迟到且拉伸着抵达，冲进一个人人都看得见的价位。此时无事可做。
2 · The sweep2 · 扫盘
The wick pokes beyond and collects the stops parked there. Then the candle CLOSES back inside — the first real information.影线戳出去，收走停在那里的止损。然后 K 线收回内侧——这是第一个真实信息。
3 · The retest entry3 · 复测入场
Price retests the level from inside and fails again. Enter on the retest close. SL beyond the sweep’s extreme; TP at the range mid, then the far side.价格从内侧复测价位，再次失败。在复测收盘入场。止损放在扫盘极值之外；止盈放在区间中部，然后是远侧。
4 · Invalidation4 · 失效
A close OUTSIDE the level that holds. Acceptance has replaced rejection — the reversal idea is dead, whatever the wicks do next.一根收在价位之外并守住的 K 线。接受取代了拒绝——反转想法已死，不管影线接下来做什么。
3 · How to set it up3 · 如何布置
The level is drawn BEFORE price arrives: yesterday’s high/low, the prior week’s edge, a session extreme.在价格抵达之前就把价位画好：昨日高低点、前周边缘、时段极值。
The move into it is visibly stretched — a streak, a spike, a run at twice the normal pace.冲向它的走势明显被拉伸——连涨连跌、冲高、或以两倍于平常的速度奔跑。
Wait for the sweep AND a close back inside. No close, no trade.等待扫盘，并且收回内侧。没有收盘，就没有交易。
Enter on the retest of the level from inside — not on the first close back.在从内侧复测价位时入场——不是在第一根收回的 K 线上。
SL beyond the sweep’s wick extreme. TP1 at the range middle, TP2 at the far side.止损放在扫盘影线极值之外。TP1 在区间中部，TP2 在远侧。
Stops here are wide by design — so the size comes DOWN. The stop decides the size.这里的止损生来就宽——所以仓位要减小。由止损决定仓位。
4 · Best in which market4 · 在哪个市场最好
GOLD黄金WORKABLE可用
Real in specific spots: London’s first-hour breakouts are traps, and failed acceptances cascade back through the range. But never fade strength at week levels — prior-week extremes are magnets, and fading week-high rejections is a measured loser.在特定位置是真的：伦敦第一小时的突破是陷阱，被拒绝的接受会瀑布般穿回区间。但永不要在周线价位反打强势——前周极值是磁铁，反打周高拒绝是被测量过的输家。
CENSUS · London 1st-hour breaks fade · week-high fade loses普查 · 伦敦第一小时突破被反打 · 反打周高会亏
EUR/USD欧元/美元STRONG FIT适配强
The reversal market. Accepted breaks of yesterday’s range reverse, Asia’s direction is handed to London backwards, stretched H1 moves snap back. If a reversal idea works anywhere in this book, it works here first.反转的市场。被接受的昨日区间突破会反转，亚洲的方向被反着交给伦敦，拉伸的 H1 走势弹回。如果反转想法在本书任何地方有效，它先在这里有效。
The reversal lives at the session seam: the Overlap’s direction is reversed into New York — the strongest handoff in the volume. Three-day streaks also snap back hard. But plain break-fading fights this market’s continuation: pick the seam or the streak, not every level.反转活在时段接缝处：重叠时段的方向被反转进纽约——本卷最强的交接。三天连涨连跌也会狠狠弹回。但普通的反打突破是在对抗这个市场的延续性：选接缝或选连涨连跌，而不是每个价位都反打。
The biggest fades in the whole census live here — at the WEEK’s horizon and after streaks. Accepted prior-week breaks collapse; three up-days snap back ≈130 points over five. The daily break, though, continues: reverse the week, never the day.整个普查中最大的反打都在这里——在周线尺度和连涨连跌之后。被接受的前周突破会崩塌；连涨三天会在五天内弹回约 130 点。但日线突破会延续：反打周线，永不反打日线。
You fade a move that is being ACCEPTED — consecutive closes beyond the level with follow-through. In gold, USD/JPY and NAS100 that is the norm, not the exception, and your reversal becomes its fuel.你反打的是一次正在被接受的走势——连续收盘在价位之外并有跟随。在黄金、美元/日元和纳斯达克100 中那是常态而非例外，你的反转会变成它的燃料。
You trade the wick instead of the close. A sweep that never closes back inside is not a rejection — it is a breakout clearing its throat.你交易影线而不是收盘。从未收回内侧的扫盘不是拒绝——那是突破在清嗓子。
The level is a news event, not structure. At 08:30 ET the first move is a liquidity event (see Order Flow): fading it without the close-back rule is how the spike’s 43% follow-rate eats beginners.价位是新闻事件，不是结构。在 08:30 ET，第一波走势是流动性事件（见订单流）：不用「收回」规则去反打它，就是冲高的 43% 跟随率吃掉新手的方式。
2 · Breakouts2 · 突破
Trading acceptance — closes that hold outside the range交易接受——守在区间之外的收盘
1 · What it is1 · 它是什么
A breakout trade bets the opposite of a reversal: that a range has ended and price will travel. Ranges store energy — a market that has spent hours inside a box has orders stacked on both sides of it, and when one side gives way, the losing side’s stops become fuel for the move.突破交易赌的是反转的反面：区间已经结束，价格将要旅行。区间储存能量——一个在箱体里待了几个小时的市场，在箱体两侧都堆满了订单，当一侧让步时，输家的止损就成了走势的燃料。
The census word for a real breakout is ACCEPTANCE: not a wick through the line, but closes beyond it that hold. One close outside is a knock on the door; closes that stay outside and build are the door opening. Every breakout definition in this book is written in closes, for that reason.普查给真正突破的词是「接受」：不是影线穿过线，而是收盘越过并守住。一根收在外面是敲门；收盘留在外面并筑起，才是门开了。本书的每个突破定义都用收盘写成，就是这个原因。
The beginner’s version — “price crossed the line, buy” — skips the only step that matters. By the time acceptance is visible the move has started, and the honest entry is the retest: price returns to the line it broke, and the line either flips to support or it doesn’t.新手的版本——「价格穿线了，买」——跳过了唯一重要的步骤。等接受可见时走势已经启动，诚实的入场是复测：价格回到它突破的那条线，线要么翻成支撑，要么不翻。
Then there is the centrepiece of this whole tab: acceptance means opposite things in different markets. Three of the four POSEIDON markets reward the accepted break. One of them punishes it — reliably, at n=540. A breakout trader who doesn’t know which market he is in is donating.然后是整个页签的中心展品：接受在不同市场里意味着相反的东西。四个 POSEIDON 市场里有三个奖赏被接受的突破。其中一个惩罚它——可靠地，在 n=540 上。不知道自己身在哪个市场的突破交易者，是在捐款。
Hours inside a box. The tighter and longer the base, the more orders are stacked on both sides of it.在箱体里待几个小时。筑底越紧、越久，两侧堆积的订单越多。
2 · Acceptance2 · 接受
Closes beyond the line — and they HOLD outside. A wick through is not acceptance; a close that stays, is.收盘越过线——并守在外面。影线穿过不是接受；留在外面的收盘才是。
3 · The retest entry3 · 复测入场
Price returns to the broken line. If the line flips — old ceiling, new floor — the retest close is the entry. A close back inside the box kills the idea.价格回到被突破的线。如果线翻转——旧天花板变新地板——复测收盘就是入场。收回箱内，想法死亡。
4 · The fakeout4 · 假突破
The close outside that doesn’t hold: two bars later price is back inside, travelling the other way. In EUR/USD this panel is not the failure case — it is the base case.收在外面却守不住：两根 K 线后价格回到内侧，朝反方向走。在欧元/美元中，这一格不是失败案例——它是基本案例。
The centrepiece — the same accepted break, four markets中心展品——同一次被接受的突破，四个市场
GOLD — continues黄金——延续
Accepted breaks of yesterday’s range keep going, the up side carrying it.被接受的昨日区间突破会继续，上行扛起。
Cash breaks continue — and so do premarket breaks. Even rejected breaks lean INTO the break (−0.185, n=316): a first close back inside is a pause, not a rejection.现金时段突破会延续——盘前突破也是。连被拒绝的突破都倾向顺着突破方向（−0.185，n=316）：第一根收回内侧是暂停，不是拒绝。
The range is drawn before the break: yesterday’s range, the premarket range, a session box.在突破之前就把区间画好：昨日区间、盘前区间、时段箱体。
Acceptance = a CLOSE beyond the line, then another close that stays outside. Wicks don’t count.接受 = 一根收在线外的 K 线，然后再一根留在外面的收盘。影线不算。
Prefer the retest entry — the first return to the line, judged on its close.优先复测入场——第一次回到线边，用它的收盘判断。
SL inside the broken range. The invalidation is a close back inside, so the stop lives just beyond it.止损放在被突破的区间之内。失效是收回内侧，所以止损就放在它之外一点。
Check your market’s verdict BEFORE the trade (fit cards below). In EUR/USD this family’s default is OFF.交易前先查你的市场的裁决（见下方适配卡）。在欧元/美元中，这个家族的默认状态是关。
Skip the second break of the same level. In gold the census found second breaks show nothing — the first close is the picture.跳过同一价位的第二次突破。在黄金中普查发现第二次突破什么都不显示——第一根收盘才是画面。
4 · Best in which market4 · 在哪个市场最好
GOLD黄金STRONG FIT适配强
Accepted breaks of yesterday’s range continue, the up side carrying it — and the desk’s whole level culture (draw yesterday, trade the close) is built on this family.被接受的昨日区间突破会延续，上行扛起——交易台的整个价位文化（画昨日、交易收盘）就建立在这个家族之上。
The lone fader. Accepted breaks of yesterday’s range REVERSE over the next two hours — breakout buying here is the census’s clearest donation. The census-backed trade on a EUR/USD break is the fade of it (see Reversals).唯一的反打者。被接受的昨日区间突破会在接下来两小时反转——在这里买突破是普查中最清晰的捐款。普查支持的欧元/美元突破交易，是反打它（见「反转」）。
Breaks continue here too — gold’s sign — but the effect is small (+0.11 ATR) against this market’s 0.12× friction tax. The continuation is real; the margin for sloppy entries is not.这里的突破也会延续——黄金的方向符号——但效应小（+0.11 ATR），而这个市场的摩擦税是 0.12×。延续是真的；给潦草入场留的余地不是。
Cash-session breaks of the prior day continue, the premarket range is a real range whose breaks continue too, and even its rejections lean into the break. This is the index’s native language.现金时段突破前日区间会延续，盘前区间是真实区间、其突破也会延续，连它的拒绝都倾向顺着突破。这是这个指数的母语。
You chase: entering on the spike candle itself, mid-wick, because the line was crossed. Acceptance hasn’t happened yet — you have bought the most expensive second of the move.你追：在影线中段的冲高 K 线本身入场，只因为线被穿过了。接受还没发生——你买下了这次走势中最贵的一秒。
The crowded line meets a news day. At 08:30 ET the first “break” is usually a stop run wearing a breakout’s clothes (43% follow-rate). The spike rule outranks the breakout rule.拥挤的线遇上新闻日。在 08:30 ET，第一波「突破」通常是穿着突破外衣的扫止损（43% 跟随率）。冲高规则压过突破规则。
You carry the family across markets unchanged. The identical picture that continues in gold, USD/JPY and NAS100 reverses in EUR/USD. The family doesn’t travel — the market’s verdict does.你把家族原封不动地搬过市场。那个在黄金、美元/日元和纳斯达克100 中延续的相同画面，在欧元/美元中反转。家族不会旅行——旅行的是市场的裁决。
3 · Trends3 · 趋势
A daily-horizon family — the intraday version is the census’s favourite corpse一个日线尺度的家族——日内版本是普查最喜欢的尸体
1 · What it is1 · 它是什么
Trend trading is the oldest family: identify the direction of the bigger move and trade only that way — buying strength’s pauses instead of fighting them. Its unit of truth is structure: a staircase of higher highs and higher lows on the chart that matters.趋势交易是最古老的家族：认清大走势的方向，只朝那个方向交易——买强势的停顿，而不是与之对抗。它的真理单位是结构：在重要的图表上一道由更高高点和更高低点组成的阶梯。
Here is the honest core this tab owes you. The beginner’s trend tool — buy the pullback to the moving average — was tested in all four POSEIDON markets at intraday scale, and it rescued trend-following NOWHERE. EMA pullbacks are dead in EUR/USD and USD/JPY, tiny and fragile in gold. The census is not anti-trend. It is anti-THAT entry.这是本页签欠你的诚实核心。新手的趋势工具——买回到移动平均线的回调——在四个 POSEIDON 市场的日内尺度上都被测过，它在任何地方都没有挽救趋势跟踪。EMA 回调在欧元/美元和美元/日元中已死，在黄金中微小而脆弱。普查不是反趋势。它是反那个入场。
Where trend evidence actually lives in this book is the DAILY horizon: daily swing breaks that keep travelling for five days, an index whose trend days lean up two-to-one, a currency pair that follows the bond yield on the days yields move. Trend is a property of days — not of the next fifteen minutes.趋势证据在本书记中真正活着的地方是日线尺度：会继续走五天的日线摆动突破、趋势日以二比一向上倾斜的指数、在收益率波动的日子里跟随债券收益率的货币对。趋势是「天」的属性——不是接下来十五分钟的属性。
So the tradeable version: let the daily chart choose the side, let structure — the last swing, the proven shelf — be the place a pullback must hold, and accept that on range days this family has no trade at all. Standing down is one of its positions.所以可交易的版本是：让日线图选边，让结构——上一个摆动、已被证明的架子价位——成为回调必须守住的地方，并接受在震荡日这个家族根本没有交易。站在一边也是它的仓位之一。
2 · The anatomy — staircase, shelf, and the trap2 · 解剖——阶梯、架子与陷阱
1 · The staircase1 · 阶梯
On the DAILY chart: higher highs, higher lows. This staircase — not an indicator — picks the side. No staircase, no trend trade.在日线图上：更高的高点、更高的低点。这道阶梯——不是指标——决定方向。没有阶梯，就没有趋势交易。
2 · Pullback to STRUCTURE2 · 回调到结构
The pullback falls to the last proven shelf — the prior swing high. Closes that hold above it keep the staircase intact; the holding close is the entry.回调跌向上一个已被证明的架子——前一个摆动高点。守在它上方的收盘保持阶梯完好；守住的收盘就是入场。
3 · The version that died3 · 死去的那个版本
Buy every dip to the moving-average line. Tested at M15 scale in all four markets, it beat nothing anywhere. The line is smooth; the losses were real.买每一个回到移动平均线的回调。在四个市场以 M15 尺度测试，它在任何地方都没赢过任何东西。线条很平滑；亏损很真实。
4 · The chop4 · 震荡
31% of EUR/USD days are RANGE days (and 30% more are quiet). Trend tools pointed at chop just pay spread and friction, slowly, with confidence.欧元/美元有 31% 的日子是震荡日（另有 30% 是安静日）。把趋势工具对准震荡，只会慢慢地、自信地支付点差和摩擦。
3 · How to set it up3 · 如何布置
The side is chosen on the DAILY chart only: staircase up, staircase down — or no staircase, which means no trade.方向只在日线图上选：阶梯向上、阶梯向下——或者没有阶梯，那就意味着没有交易。
The pullback target is a STRUCTURE level — prior swing high/low, a proven shelf — drawn in advance.回调目标是结构价位——前摆动高低点、已被证明的架子——提前画好。
Enter on the close that HOLDS the structure, not on the touch.在守住结构的收盘上入场，不是在触碰时。
SL beyond the structure: a close through it breaks the staircase, and the idea with it.止损放在结构之外：收盘穿过它，阶梯断裂，想法也随之而死。
No EMA-dip entries at intraday scale — census-dead in every market tested.日内尺度上不用 EMA 回调入场——在每个被测市场中都是普查判死。
Check the day type first. If the market is ranging, this family stands down.先检查日类型。如果市场在震荡，这个家族就站在一边。
4 · Best in which market4 · 在哪个市场最好
GOLD黄金WORKABLE可用
Gold trends inside the day, and the desk’s regime work is trend-aware (its current read: TREND_UP, marginal — desk practice, not a census verdict). But intraday EMA pullbacks are tiny and fragile here, and streaks fade: gold’s trend is a same-day animal, ridden on closes — not a dip-buying machine.黄金在日内走趋势，交易台的市况工作也对趋势敏感（它目前的解读：趋势向上、勉强——这是交易台实践，不是普查裁决）。但日内 EMA 回调在这里微小而脆弱，连涨连跌会反打：黄金的趋势是当日动物，用收盘骑乘——不是买回调的机器。
CENSUS · EMA pullbacks tiny & fragile · streaks fade普查 · EMA 回调微小而脆弱 · 连涨连跌反打
EUR/USD欧元/美元WORKABLE可用
A split personality, decided by timeframe: intraday it is the fade market — but at the daily horizon its swing breaks CONTINUE, about +17 pips over five days. Trend-trade the days; fade the hours; never confuse the two.由时间框架决定的双重人格：日内它是反打市场——但在日线尺度上，它的摆动突破会延续，五天约 +17 点。用趋势交易「天」，用反打交易「小时」；永不要混淆两者。
Its trend engine is the bond market: on big-yield-move days the pair followed the 10-year yield 84% of the time — the closest thing this volume has to a trend law (a descriptive fact). EMA pullbacks are dead, and three-day streaks snap back against trend-chasers.它的趋势引擎是债券市场：在收益率大幅波动的日子里，这对货币有 84% 的时间跟随 10 年期收益率——这是本卷最接近趋势规律的东西（一个描述性事实）。EMA 回调已死，三天连涨连跌会反咬追趋势的人。
DESCRIPTIVE · follows the 10Y 84.2% of big-yield days · n=76描述性 · 大收益率日 84.2% 跟随 10 年期 · n=76
NAS100纳斯达克100STRONG FIT适配强
The trend market of the four — by construction. Trend days lean up nearly 2:1 (structural drift), accepted daily breaks continue, premarket breaks continue into the cash session. Its own streaks still snap back: ride the drift, respect the streak.四者中的趋势市场——生来如此。趋势日以近 2:1 向上倾斜（结构性漂移），被接受的日线突破会延续，盘前突破延续进现金时段。它自己的连涨连跌仍会弹回：骑乘漂移，尊重连涨连跌。
CONFIRMED · breaks continue +0.11 · trend days up ≈ 2:1已确认 · 突破延续 +0.11 · 趋势日向上 ≈ 2:1
5 · It fails when5 · 它在何时失败
You import the intraday dip-buy. EMA pullbacks at M15 scale are this census’s most replicated corpse — four markets, no rescue. If your trend entry is an average, this book has already run your backtest.你进口了日内买回调。M15 尺度的 EMA 回调是本普查复制次数最多的尸体——四个市场，无一获救。如果你的趋势入场是一条平均线，本书已经替你跑过回测了。
You trend-trade a range day. On EUR/USD, 31% of days are RANGE and another 30% quiet-other — structure tools in chop produce small losses with great confidence.你在震荡日做趋势。欧元/美元 31% 的日子是震荡日，另 30% 是安静其他类——结构工具在震荡中自信地产出小额亏损。
You chase the streak. Three same-direction days lean the OTHER way over the next five in all four markets — the trend family’s fuel gauge reads empty exactly when it feels fullest.你追连涨连跌。连续三天同方向在四个市场中都会在接下来五天倾向反方向——趋势家族的油量表，恰恰在感觉最满的时候见底。
4 · Price Action4 · 价格行为
The reading layer under every other family — levels first, closes always所有其他家族之下的阅读层——先价位，永远看收盘
1 · What it is1 · 它是什么
Price action is not really a strategy family — it is the reading layer underneath all the others: the skill of looking at raw candles and levels and saying what just happened, without an indicator translating for you. Every other family in this tab is written in its language.价格行为其实不是一个策略家族——它是其他所有家族之下的阅读层：看着原始 K 线和价位、说出刚刚发生了什么的技能，不需要指标替你翻译。本页签的每个其他家族，都是用它的语言写成的。
Its first law in this book: LEVELS FIRST, CANDLES SECOND. A candle shape means almost nothing in the middle of nowhere and a great deal at a level the market has already proved. The census tested the same shapes in both places — and location, not shape, was usually the difference between a finding and a corpse.它在本书的第一定律：先价位，后 K 线。K 线形态在无名之处几乎什么都不代表，在市场已证明的价位处却意义重大。普查在两个地方测过相同的形态——位置，而不是形态，通常就是结论与尸体之间的区别。
Its second law: the CLOSE is the verdict, the wick is just the visit. A wick tells you price went somewhere; a close tells you whether it was accepted or rejected there. That is why the census defines everything — breaks, rejections, engulfings — on bodies and closes.第二定律：收盘是裁决，影线只是到访。影线告诉你价格去过某处；收盘告诉你它在那里被接受还是被拒绝。这就是为什么普查用实体和收盘定义一切——突破、拒绝、吞没。
And its honest warning: single-candle signals are weak tea. The strongest candle behaviour in the entire census — gold’s mid-range engulfing, confirmed three times — is a +0.23 ATR tilt: a lean you can only see across hundreds of instances, not a promise about the next candle. Read honestly, price action is a probability language, not a signal machine.还有它的诚实警告：单根 K 线信号是淡茶。整个普查中最强的 K 线行为——黄金的区间中部吞没，被确认三次——是 +0.23 ATR 的倾斜：一种只能在数百个实例中看到的倾向，不是对下一根 K 线的承诺。诚实地读，价格行为是一种概率语言，不是信号机器。
2 · The anatomy — how to read, in order2 · 解剖——阅读的顺序
1 · One candle, honestly1 · 一根 K 线，诚实地看
The wick is a visit; the body is the argument; the close is the verdict. Every definition in this book starts here.影线是到访；实体是论证；收盘是裁决。本书的每个定义都从这里开始。
2 · Levels first2 · 先画价位
Draw the levels first. Inside the box, individual candles are mostly static — the mid-range versions of famous patterns keep dying in there.先画价位。在箱体内部，单根 K 线大多是静电噪声——著名形态的区间中部版本不断死在里面。
3 · The one pattern that lives3 · 唯一活着的形态
A body that swallows the prior body, AWAY from any level, leans continuation. Gold: +0.23 ATR, n=1,788, confirmed three times.一根吞没前一根实体的 K 线，远离任何价位时，倾向延续。黄金：+0.23 ATR，n=1,788，被确认三次。
4 · The pin that lies4 · 说谎的针
The textbook’s favourite reversal candle. In gold it looked alive for a year, then died on the bigger corpus (+0.011 ≈ nothing, n=3,847). Do not trade the pin.教科书最喜欢的反转 K 线。在黄金中它活了一年，然后死在更大的语料上（+0.011 ≈ 零，n=3,847）。不要交易针形线。
3 · How to set it up3 · 如何布置
Draw before you read: yesterday’s high/low, the prior week’s edges, session boxes — then look at candles.读之前先画：昨日高低点、前周边缘、时段框——然后再看 K 线。
Classify every candle by location first: at a level, or mid-range. The same shape is a different animal in each place.先按位置给每根 K 线分类：在价位处，还是在区间中部。同一个形态在两个地方是不同的动物。
Read closes, not wicks: what was accepted? What was rejected?读收盘，不读影线：什么被接受了？什么被拒绝了？
Treat any single-candle signal as a lean, not a trigger — require the next close to agree.把任何单根 K 线信号当作倾向，不是触发——要求下一根收盘同意。
Skip pins at levels entirely: dead or negative in all four markets.完全跳过价位处的针形线：在四个市场中已死或为负。
In NAS100, treat an inside bar as a volatility warning, not a direction: the next 2 hours expand (+0.36 ATR). Size down, or stand aside.在纳斯达克100 中，把内包线当作波动警告，不是方向：接下来 2 小时扩张（+0.36 ATR）。减小仓位，或站在一边。
4 · Best in which market4 · 在哪个市场最好
GOLD黄金STRONG FIT适配强
The home market of this book’s reading method — and the one market where a candle pattern earned real confirmation: mid-range engulfing continuation, confirmed three times. Pins at levels: dead on the full corpus.本书阅读方法的主场——也是唯一有 K 线形态赢得真实确认的市场：区间中部吞没延续，被确认三次。价位处的针形线：在完整语料上已死。
Levels matter enormously here — its whole volume is level behaviour — but the famous candle signals betray the lore: engulfing is CONFIRMED NEGATIVE (the pattern leans the wrong way) and pins at levels are negative or dead. Read EUR/USD’s levels; do not import gold’s candle lore.价位在这里极其重要——它整卷都是价位行为——但著名的 K 线信号背叛了传说：吞没是确认的负值（形态倾向错误方向），价位处的针形线为负或已死。读欧元/美元的价位；不要进口黄金的 K 线传说。
Engulfing mid-range is confirmed but tiny (+0.057 ATR ≈ half a pip — under its 0.12× friction, so a fact, not a trade). Pins at levels are confirmed NEGATIVE. The reading layer informs here; it does not pay.区间中部吞没被确认但微小（+0.057 ATR ≈ 半个点——低于它 0.12× 的摩擦，所以是事实，不是交易）。价位处的针形线是确认的负值。阅读层在这里提供信息；它不付钱。
One genuine price-action fact: inside bars precede 2-hour volatility expansion (+0.36 ATR, n=2,403) — the only inside-bar effect in the book. Engulfing leans negative; pins are dead. Read this market for volatility, not direction.一个真正的价格行为事实：内包线之后是 2 小时波动扩张（+0.36 ATR，n=2,403）——本书唯一的内包线效应。吞没倾向负值；针形线已死。读这个市场是为了波动，不是方向。
You pattern-spot in the static: naming shapes between levels, where the census’s null models eat them. Most famous patterns, pooled everywhere, are indistinguishable from noise.你在静电噪声中找形态：在价位之间命名形状，普查的零模型会在那里吃掉它们。大多数著名形态，在任何地方汇总，都是与噪声无法区分的。
You trade the single candle. Even the best-confirmed candle effect is a population tilt — the NEXT instance still goes the other way nearly half the time. One candle is a hint; a stack of closes is an argument.你交易单根 K 线。即使是确认得最好的 K 线效应也是总体倾斜——下一个实例仍有近一半时间反着走。一根 K 线是暗示；一叠收盘才是论证。
You import lore across markets. The engulfing that continues in gold leans the wrong way in EUR/USD and NAS100. A pattern is not a law of nature — it is a local habit, and it expires at the border.你把传说搬过市场。在黄金中延续的吞没，在欧元/美元和纳斯达克100 中倾向错误方向。形态不是自然法则——它是本地习惯，到边境就过期。
5 · Order Flow — the retail edition5 · 订单流——零售版
Who is forced to trade, and when — read through proxies, honestly谁被迫交易、何时被迫——用代理指标读，诚实地读
1 · What it is1 · 它是什么
Order flow starts from a different question. Not “what pattern is that?” but “who is being FORCED to trade right now?” Prices move hardest when someone has no choice: stops triggering, positions being liquidated, a session’s worth of orders arriving in a market too thin to hold them.订单流从一个不同的问题开始。不是「那是什么形态？」而是「谁现在正被迫交易？」当有人别无选择时，价格动得最猛：止损被触发、仓位被强平、一个时段的订单涌进一个承接不住它们的薄市场。
The honest retail framing, up front: on a retail MT5 spot feed there is NO depth-of-market, no footprint chart, no real volume. Anyone selling you those pictures for spot gold is selling you a drawing. What a retail trader actually has is proxies — and proxies, used honestly, are enough to read the flow events this book keeps meeting.先说诚实的零售框架：在零售 MT5 现货数据上，没有市场深度、没有足迹图、没有真实成交量。任何向你出售现货黄金那些图的人，卖给你的是一幅画。零售交易者真正拥有的是代理指标——而代理指标，诚实地用，足以读懂本书不断遇到的流动事件。
Proxy one: tick volume — the count of price changes, not contracts. It measures activity, not size, and it still tells you when a move is crowded. Proxy two: the clock — liquidity arrives in relays (Asia hands to London, London to the Overlap), and forced flow concentrates at the seams and at scheduled releases. Proxy three: the level itself — stops cluster at obvious prices, which is why obvious prices get raided.代理一：跳动量——价格变化的计数，不是合约数。它测量活动，不是规模，它仍能告诉你一次走势何时拥挤。代理二：时钟——流动性以接力方式抵达（亚洲交给伦敦，伦敦交给重叠），被迫的流动集中在接缝和定时发布上。代理三：价位本身——止损聚集在显眼的价格旁，这就是为什么显眼的价格会被突袭。
The census’s most famous number is an order-flow fact wearing a pattern’s clothes: on 08:30 ET data days the first spike is followed only 43% of the time. The spike is not information — it is the sound of stops and market orders colliding in a thin book. The fade is the flow trade: let the forced traders finish, then trade against the exhaustion.普查最著名的数字，是一件穿着形态外衣的订单流事实：在 08:30 ET 数据日，第一波冲高只有 43% 的时间被跟随。冲高不是信息——它是止损和市价单在薄订单簿中相撞的声音。反打就是流动交易：让被迫的交易者先做完，然后逆着耗尽交易。
Read this way, “liquidity grab” and “stop hunt” stop being conspiracy words and become bookkeeping: someone needed size, the nearest size was parked beyond an obvious level, and price went to collect it. Your job is not to be the size.这样读，「流动性抓取」和「猎杀止损」不再是阴谋词，而是记账：有人需要规模，最近的规模停在一个显眼价位之外，价格就去收取它。你的工作不是成为那个规模。
Obvious levels collect stops the way gutters collect rain: everyone’s protection parks in the same few dollars below the shelf — and above the ceiling.显眼价位像排水沟收集雨水一样收集止损：每个人的保护都停在架子下方同样的几美元里——天花板上方也是。
2 · The cascade2 · 瀑布
One push through the shelf fires the clustered stops; their market orders become the waterfall. When the forced flow is spent, price often reclaims the level — the cascade was the point, not the start of a trend.一次穿过架子的推送点燃聚集的止损；它们的市价单变成瀑布。当被迫的流动耗尽，价格常常收回价位——瀑布本身就是目的，不是趋势的开始。
3 · The relay3 · 接力
Liquidity arrives in shifts. London hands its book to the Overlap at 12:30 UTC — in gold the direction is inherited (+0.42 ATR); in EUR/USD the same seam runs backwards. Trade the seam, not the silence between seams.流动性分班次抵达。伦敦在 12:30 UTC 把它的订单簿交给重叠——在黄金中方向被继承（+0.42 ATR）；在欧元/美元中同一接缝反向运行。交易接缝，不是接缝之间的沉默。
4 · The crowded candle4 · 拥挤的 K 线
Tick volume counts price changes, not contracts. It can tell you a candle was crowded; it cannot tell you who was big. Reading more into it is reading tea leaves.跳动量数的是价格变化，不是合约。它能告诉你一根 K 线拥挤；它不能告诉你谁大。读得更多，就是看茶叶占卜。
3 · How to set it up3 · 如何布置
Name the forced trader BEFORE the trade: whose stops, whose liquidation, whose session open is about to fire?交易前先说出被迫交易者的名字：谁的止损、谁的强平、谁的时段开盘即将点火？
Trade the seams: session handoffs and the 08:30 ET window are where forced flow concentrates.交易接缝：时段交接和 08:30 ET 窗口是被迫流动集中的地方。
On release days, let the first spike finish. The fade is the flow trade; chasing the spike is BEING the flow.在发布日，让第一波冲高先走完。反打是流动交易；追冲高就是成为流动本身。
Treat every obvious level as a stop magnet: expect the raid through it before you trust the rejection of it.把每个显眼价位都当作止损磁铁：在信任它被拒绝之前，先预期它会被突袭穿过。
Use tick volume only as an activity gauge — crowded vs quiet. Never as “smart money”.跳动量只用作活动仪表——拥挤还是安静。永远不要当作「聪明钱」。
If you cannot say who is forced, there is no flow trade. It is just a pattern trade wearing flow’s clothes.如果你说不出谁被迫，就没有流动交易。那只是一笔穿着流动外衣的形态交易。
4 · Best in which market4 · 在哪个市场最好
GOLD黄金STRONG FIT适配强
Gold is this book’s flow textbook: the 08:30 spike fade (43% follow, CPI 39%), the London→Overlap relay (+0.42 ATR), and raid-and-reclaim anatomy at crowded shelves are one lesson — the forced traders move first, and their exhaustion is the signal.黄金是本书的流动教科书：08:30 冲高反打（跟随 43%，CPI 39%）、伦敦→重叠接力（+0.42 ATR）、拥挤架子处的突袭与收回解剖，是同一课——被迫的交易者先动，他们的耗尽就是信号。
The flow trade here is the same spike, bigger: the pooled 08:30 first spike fades −0.92 ATR. And its session relay runs backwards (Asia→London −0.41) — the flow is real; the directions are local.这里的流动交易是同一波冲高，更大：汇总的 08:30 第一波冲高反打 −0.92 ATR。而它的时段接力反向运行（亚洲→伦敦 −0.41）——流动是真的；方向是本地的。
The spike fade is confirmed a third time (−0.65 ATR), and Tokyo is a full session here — real flow, not a warm-up. The famous flow folklore, though, failed measurement: suspected intervention shocks in Tokyo numbered exactly one in three years.冲高反打第三次被确认（−0.65 ATR），而东京在这里是完整的时段——真实的流动，不是热身。但著名的流动传说没能通过测量：东京疑似干预冲击在三年里恰好只有一次。
The premarket range (median 124 pts) is a real liquidity map — its breaks continue (+0.22). But the overnight gap, the loudest “flow” story in index folklore, carries zero directional information: it fills 57.6% of the time and predicts nothing either way.盘前区间（中位数 124 点）是真实的流动性地图——它的突破会延续（+0.22）。但隔夜跳空，指数传说中最响亮的「流动」故事，不携带任何方向信息：它有 57.6% 的时间被填补，两个方向都什么也预测不了。
You pretend tick volume is real volume. It counts updates, not size — the moment a “flow” read depends on knowing WHO traded, a retail feed cannot pay for it.你假装跳动量是真实成交量。它数的是更新次数，不是规模——一旦某个「流动」解读依赖于知道是谁在交易，零售数据就付不起这个钱。
You see conspiracies in noise. Most wicks through levels are thin books and ordinary stops, not a hunt staged for you personally. The census version is duller and truer: extremes form at session edges because that is where the orders are.你在噪声中看到阴谋。大多数穿过价位的影线是薄订单簿和普通止损，不是专为你个人上演的猎杀。普查的版本更无聊也更真实：极值在时段边缘形成，因为订单就在那里。
You fade the first move of a REAL repricing. 43% of first spikes ARE followed — which is why the fade waits for exhaustion (the stall, the close back) instead of shorting altitude alone.你反打一次真实重新定价的第一波。43% 的第一波冲高确实被跟随——这就是为什么反打要等耗尽（停滞、收回），而不是只做空高度。
POSEIDON · Volume II · EUR/USD — the picturePOSEIDON · 第二卷 · 欧元/美元——图像版
Where gold continues, EUR/USD fades.黄金延续的地方，欧元/美元反打。
48CONFIRMED — survived unseen data已确认——通过未见数据检验26WEAK — a flag, not a finding弱——只是标记，不是结论52DEAD — killed by the holdout已死亡——被留出数据杀掉15DESCRIPTIVE — measured, no direction claim描述性——只测量，不下方向结论
How wide each session runs每个时段跑多宽Median session range in pips · 788 days. Six days in ten are range or quiet.时段区间中位数（点）· 788 天。十天里有六天是震荡或安静。
Off-hours非交易时段
9p
Overnight隔夜
14p
Asia亚洲
18p
London伦敦
29p
New York纽约
29p
Overlap重叠
32p
The friction tax — cost of one round trip摩擦税——一次往返的成本Round-trip cost as a fraction of one median M15 candle (ATR). Smaller = cheaper to trade.往返成本占一根典型 M15 K 线（ATR）的比例。越小 = 交易成本越低。
Gold黄金
0.07×
NAS100纳斯达克100
0.096×
USD/JPY美元/日元
0.12×
EUR/USD欧元/美元
0.20×
One day on the clock时钟上的一天Session blocks on the desk clock (UTC). Edge segments — first and last 30 minutes — are where the highs and lows form.交易台时钟（UTC）的时段区块。边缘时段——最初和最后 30 分钟——是高低点形成的地方。
Off-hrs非交易时段
9p
20–23 UTC
Overnight隔夜
14p
23–02
Asia亚洲
18p
02–07
London伦敦
29p
07–12
Overlap重叠
32p
12–15
New York纽约
29p
15–20
Laws that survived all four markets在四个市场都活下来的规律
📰
The first spike lies第一波冲高会说谎
On 08:30 ET data days the first spike is faded in gold, EUR/USD and USD/JPY. Never chase it.在 08:30 ET 数据日，第一波冲高在黄金、欧元/美元和美元/日元中都会被反打。永远不要追它。
📉
Streaks snap back连涨连跌会弹回
Three same-direction days lean the other way over the next five — in all four markets.连续三天同方向，接下来五天会倾向反方向——四个市场全部如此。
🧭
Breaks continue — except EUR/USD突破会延续——欧元/美元除外
An accepted break of yesterday’s range keeps going in gold, USD/JPY and NAS100. In EUR/USD it reverses.被接受的昨日区间突破在黄金、美元/日元和纳斯达克100中会继续。在欧元/美元中它会反转。
⏱
Edges make the extremes边缘产生极值
Session highs and lows form in the first or last 30 minutes — everywhere, in every market.时段高低点在最初或最后 30 分钟形成——每个市场、每个地方都如此。
The headline behaviours — look first, read later头条行为——先看图，后读字
The break that FAILS会失败的突破
An accepted close beyond yesterday’s range reverses over the next 2 hours. The mirror image of gold — this is the volume’s most important number.被接受的昨日区间外收盘，在接下来 2 小时会反转。黄金的镜像——这是本卷最重要的数字。
The pooled 08:30 ET first spike is faded over the next 2 hours — the same behaviour as gold, and one of the few effects big enough to clear this market’s costs.汇总的 08:30 ET 第一波冲高在接下来 2 小时被反打——与黄金相同的行为，也是少数大到能覆盖本市场成本的效应之一。
Asia’s direction is reversed across the London session. Handoffs in this market are fade-the-prior-session trades, not continuation trades.亚洲的方向在伦敦时段被反转。这个市场的时段接力是「反打上一时段」，不是延续。
The direction of London’s first half hour tends to survive to the London close. (Gold leans the other way — the two markets open differently.)伦敦最初半小时的方向往往会活到伦敦收盘。（黄金倾向相反——两个市场的开盘方式不同。）
An H1 move stretched at least 2× ATR from its average snaps back over the next 2 hours. One of the few effects that clears the cost bar.H1 走势从均值拉伸至少 2 倍 ATR 后，会在接下来 2 小时弹回。少数能越过成本线的效应之一。
Daily swing breaks continue — about +17 pips over 5 days. Intraday fades, daily trends: the timeframe decides which market you are looking at.日线摆动突破会延续——5 天约 +17 点。日内反打、日线趋势：时间框架决定你看到的是哪个市场。
The full chapters follow below — every number above is drawn from them, unchanged.完整章节在下方——上面的每一个数字都来自它们，未作任何改动。
Start Here从这里开始
POSEIDON — Volume II: EUR/USD. The same observation-first census that produced Volume I (gold), now run on the world's most traded instrument. Feed: Dukascopy EUR/USD M15 (aggregated from M1 BID), 2023-10-01 → 2026-10-12 — 106,368 bars, 788 valid trading days, one uniform corpus; daily and hourly bars are derived from the same feed, never mixed with another. 141 behaviours were defined mechanically before measurement, extracted completely, stacked against session- and direction-matched random nulls, and judged on a chronological 60/40 discovery/holdout split: 48 CONFIRMED, 26 WEAK, 52 DEAD, 15 DESCRIPTIVE. Verdicts use the charter rule (same sign in both halves; CONFIRMED needs |holdout| ≥ 0.05 ATR, ≥ 50% of the discovery effect, holdout n ≥ 30). All effects are gross of trading costs; the cost audit is its own chapter.
The one-sentence character of EUR/USD: where gold continues, EUR/USD fades. Accepted range breaks reverse, release spikes are given back, streaks stall — intraday. On the daily chart the same market trends. The timeframe is the regime.
Accepted breaks of yesterday's range — they fade昨日区间被接受突破——随后回吐
CONFIRMED — NEGATIVE已确认 — 负向
The volume's signature result, and the mirror image of gold: when EUR/USD closes beyond yesterday's high or low, the next two hours lean AGAINST the break. A closing break of the prior-day range is an exhaustion print in this market, not a starting gun. n = 540 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.360 ATR, holdout -0.359 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
本卷的标志性结果，与黄金恰好相反：欧元/美元收盘突破昨日高点或低点后，随后两小时倾向于逆突破方向运行。对这个市场而言，收盘突破前日区间是衰竭信号，而不是发令枪。样本 n = 540。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.360 ATR，留出期 -0.359 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 14.9 per month (≈ 179 per year) · Dukascopy EUR/USD M15 · n = 540频率：每月约 14.9 次（每年约 179 次） · n = 540
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=540.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=540。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-09-11T13:30:00+00:00', 'bar_index': 48638, 'dir': 1.0, 'primary_ret_ATR': 0.021735242447466253, 'primary_ret_usd': 2.999999999997449e-05, 'path_rmse_vs_median': 0.5104468209626478, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-10-02T14:15:00+00:00', 'bar_index': 25121, 'dir': -1.0, 'primary_ret_ATR': -0.7113296548946472, 'primary_ret_usd': -0.0005500000000000504, 'path_rmse_vs_median': 0.5753575032399895, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-06T15:15:00+00:00', 'bar_index': 71109, 'dir': -1.0, 'primary_ret_ATR': 0.6516636834920756, 'primary_ret_usd': 0.000360000000000138, 'path_rmse_vs_median': 0.6356619908710451, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-09-11T13:30:00+00:00', 'bar_index': 48638, 'dir': 1.0, 'primary_ret_ATR': 0.021735242447466253, 'primary_ret_usd': 2.999999999997449e-05, 'path_rmse_vs_median': 0.5104468209626478, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-10-02T14:15:00+00:00', 'bar_index': 25121, 'dir': -1.0, 'primary_ret_ATR': -0.7113296548946472, 'primary_ret_usd': -0.0005500000000000504, 'path_rmse_vs_median': 0.5753575032399895, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-06T15:15:00+00:00', 'bar_index': 71109, 'dir': -1.0, 'primary_ret_ATR': 0.6516636834920756, 'primary_ret_usd': 0.000360000000000138, 'path_rmse_vs_median': 0.6356619908710451, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Rejected breaks of yesterday's range昨日区间被拒绝突破
CONFIRMED已确认
A wick beyond yesterday's high/low that closes back inside is followed by continued movement in the rejection direction — the level holds and price leaves it (+2.2 pips holdout median effect). n = 422 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.028 ATR, holdout +0.407 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
刺破昨日高/低但收盘回到区间内的形态，随后价格继续朝拒绝方向运行——价位守住，价格离开（留出期中位效应 +2.2 点）。样本 n = 422。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.028 ATR，留出期 +0.407 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 11.6 per month (≈ 140 per year) · Dukascopy EUR/USD M15 · n = 422频率：每月约 11.6 次（每年约 140 次） · n = 422
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=422.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=422。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-12-25T00:45:00+00:00', 'bar_index': 30827, 'dir': -1.0, 'primary_ret_ATR': -0.1904940116547172, 'primary_ret_usd': -9.999999999998899e-05, 'path_rmse_vs_median': 0.48047685464112977, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-31T20:00:00+00:00', 'bar_index': 8344, 'dir': 1.0, 'primary_ret_ATR': None, 'primary_ret_usd': None, 'path_rmse_vs_median': 0.6118723656364616, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-01-07T22:00:00+00:00', 'bar_index': 56640, 'dir': 1.0, 'primary_ret_ATR': 0.6519620339688286, 'primary_ret_usd': 0.000280000000000058, 'path_rmse_vs_median': 0.7134491944700581, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-12-25T00:45:00+00:00', 'bar_index': 30827, 'dir': -1.0, 'primary_ret_ATR': -0.1904940116547172, 'primary_ret_usd': -9.999999999998899e-05, 'path_rmse_vs_median': 0.48047685464112977, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-31T20:00:00+00:00', 'bar_index': 8344, 'dir': 1.0, 'primary_ret_ATR': None, 'primary_ret_usd': None, 'path_rmse_vs_median': 0.6118723656364616, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-01-07T22:00:00+00:00', 'bar_index': 56640, 'dir': 1.0, 'primary_ret_ATR': 0.6519620339688286, 'primary_ret_usd': 0.000280000000000058, 'path_rmse_vs_median': 0.7134491944700581, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Asia range breaks accepted in London伦敦时段接受亚洲区间突破
DEAD已否定
n = 517 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.023 ATR, holdout -0.108 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 517。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.023 ATR，留出期 -0.108 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 14.2 per month (≈ 171 per year) · Dukascopy EUR/USD M15 · n = 517频率：每月约 14.2 次（每年约 171 次） · n = 517
Asia range sweeps rejected in London伦敦时段拒绝亚洲区间扫破
CONFIRMED已确认
London's first touch of the Asia range that fails (wick beyond, close back) continues in the rejection direction. Asia's edges are London's fade levels in EUR/USD. n = 466 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.318 ATR, holdout +0.274 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
伦敦首次触及亚洲区间失败（刺破后收回）时，价格继续朝拒绝方向运行。在欧元/美元里，亚洲区间的边缘是伦敦的反打价位。样本 n = 466。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.318 ATR，留出期 +0.274 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 12.9 per month (≈ 155 per year) · Dukascopy EUR/USD M15 · n = 466频率：每月约 12.9 次（每年约 155 次） · n = 466
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=466.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=466。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2023-12-25T07:45:00+00:00', 'bar_index': 5799, 'dir': -1.0, 'primary_ret_ATR': -0.0, 'primary_ret_usd': -0.0, 'path_rmse_vs_median': 0.4521901273476026, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-28T08:45:00+00:00', 'bar_index': 33067, 'dir': 1.0, 'primary_ret_ATR': -0.05153469913055235, 'primary_ret_usd': -4.0000000000040004e-05, 'path_rmse_vs_median': 0.777581870037938, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-10-27T09:45:00+00:00', 'bar_index': 51695, 'dir': -1.0, 'primary_ret_ATR': -1.5592239423195022, 'primary_ret_usd': -0.0008099999999999774, 'path_rmse_vs_median': 0.9663826016275048, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2023-12-25T07:45:00+00:00', 'bar_index': 5799, 'dir': -1.0, 'primary_ret_ATR': -0.0, 'primary_ret_usd': -0.0, 'path_rmse_vs_median': 0.4521901273476026, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-28T08:45:00+00:00', 'bar_index': 33067, 'dir': 1.0, 'primary_ret_ATR': -0.05153469913055235, 'primary_ret_usd': -4.0000000000040004e-05, 'path_rmse_vs_median': 0.777581870037938, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-10-27T09:45:00+00:00', 'bar_index': 51695, 'dir': -1.0, 'primary_ret_ATR': -1.5592239423195022, 'primary_ret_usd': -0.0008099999999999774, 'path_rmse_vs_median': 0.9663826016275048, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
London range breaks accepted in US hours美国时段接受伦敦区间突破
CONFIRMED — NEGATIVE已确认 — 负向
n = 480 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.183 ATR, holdout -0.137 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 480。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.183 ATR，留出期 -0.137 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 13.3 per month (≈ 160 per year) · Dukascopy EUR/USD M15 · n = 480频率：每月约 13.3 次（每年约 160 次） · n = 480
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-07-14T16:15:00+00:00', 'bar_index': 44521, 'dir': -1.0, 'primary_ret_ATR': -0.33225642000821054, 'primary_ret_usd': -0.000280000000000058, 'path_rmse_vs_median': 0.506481902722717, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-04T16:00:00+00:00', 'bar_index': 64584, 'dir': -1.0, 'primary_ret_ATR': -0.9675101043947395, 'primary_ret_usd': -0.0008699999999999264, 'path_rmse_vs_median': 0.6212088669580378, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-06T15:00:00+00:00', 'bar_index': 29636, 'dir': -1.0, 'primary_ret_ATR': 0.06562468873728133, 'primary_ret_usd': 9.000000000014552e-05, 'path_rmse_vs_median': 0.7756836595413862, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-07-14T16:15:00+00:00', 'bar_index': 44521, 'dir': -1.0, 'primary_ret_ATR': -0.33225642000821054, 'primary_ret_usd': -0.000280000000000058, 'path_rmse_vs_median': 0.506481902722717, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-04T16:00:00+00:00', 'bar_index': 64584, 'dir': -1.0, 'primary_ret_ATR': -0.9675101043947395, 'primary_ret_usd': -0.0008699999999999264, 'path_rmse_vs_median': 0.6212088669580378, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-06T15:00:00+00:00', 'bar_index': 29636, 'dir': -1.0, 'primary_ret_ATR': 0.06562468873728133, 'primary_ret_usd': 9.000000000014552e-05, 'path_rmse_vs_median': 0.7756836595413862, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
London range sweeps rejected in US hours美国时段拒绝伦敦区间扫破
CONFIRMED已确认
n = 438 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.001 ATR, holdout +0.162 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 438。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.001 ATR，留出期 +0.162 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 12.1 per month (≈ 145 per year) · Dukascopy EUR/USD M15 · n = 438频率：每月约 12.1 次（每年约 145 次） · n = 438
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-06-24T15:15:00+00:00', 'bar_index': 68133, 'dir': -1.0, 'primary_ret_ATR': 0.42689584065702874, 'primary_ret_usd': 0.00036999999999998145, 'path_rmse_vs_median': 0.46985120032636346, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-03-19T17:00:00+00:00', 'bar_index': 11596, 'dir': -1.0, 'primary_ret_ATR': 0.18235067596720556, 'primary_ret_usd': 8.000000000008001e-05, 'path_rmse_vs_median': 0.6088187205612269, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-10-13T12:30:00+00:00', 'bar_index': 50746, 'dir': 1.0, 'primary_ret_ATR': -0.6514592837785953, 'primary_ret_usd': -0.0005599999999998939, 'path_rmse_vs_median': 0.6347482825232297, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-06-24T15:15:00+00:00', 'bar_index': 68133, 'dir': -1.0, 'primary_ret_ATR': 0.42689584065702874, 'primary_ret_usd': 0.00036999999999998145, 'path_rmse_vs_median': 0.46985120032636346, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-03-19T17:00:00+00:00', 'bar_index': 11596, 'dir': -1.0, 'primary_ret_ATR': 0.18235067596720556, 'primary_ret_usd': 8.000000000008001e-05, 'path_rmse_vs_median': 0.6088187205612269, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-10-13T12:30:00+00:00', 'bar_index': 50746, 'dir': 1.0, 'primary_ret_ATR': -0.6514592837785953, 'primary_ret_usd': -0.0005599999999998939, 'path_rmse_vs_median': 0.6347482825232297, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Fractal swing breaks (M15)分形摆动突破（M15）
CONFIRMED — NEGATIVE已确认 — 负向
n = 6804 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.086 ATR, holdout -0.059 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 6804。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.086 ATR，留出期 -0.059 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 187.1 per month (≈ 2246 per year) · Dukascopy EUR/USD M15 · n = 6804频率：每月约 187.1 次（每年约 2246 次） · n = 6804
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-06-04T14:45:00+00:00', 'bar_index': 16867, 'dir': 1.0, 'primary_ret_ATR': 0.014864922460198564, 'primary_ret_usd': 1.0000000000065512e-05, 'path_rmse_vs_median': 0.4509909189328722, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-04-14T15:30:00+00:00', 'bar_index': 63238, 'dir': -1.0, 'primary_ret_ATR': -0.8492678486081543, 'primary_ret_usd': -0.0006699999999999484, 'path_rmse_vs_median': 0.49266457896874405, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-03-27T15:15:00+00:00', 'bar_index': 37125, 'dir': 1.0, 'primary_ret_ATR': 0.029647278703674652, 'primary_ret_usd': 2.999999999997449e-05, 'path_rmse_vs_median': 0.5265436957323172, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-06-04T14:45:00+00:00', 'bar_index': 16867, 'dir': 1.0, 'primary_ret_ATR': 0.014864922460198564, 'primary_ret_usd': 1.0000000000065512e-05, 'path_rmse_vs_median': 0.4509909189328722, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-04-14T15:30:00+00:00', 'bar_index': 63238, 'dir': -1.0, 'primary_ret_ATR': -0.8492678486081543, 'primary_ret_usd': -0.0006699999999999484, 'path_rmse_vs_median': 0.49266457896874405, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-03-27T15:15:00+00:00', 'bar_index': 37125, 'dir': 1.0, 'primary_ret_ATR': 0.029647278703674652, 'primary_ret_usd': 2.999999999997449e-05, 'path_rmse_vs_median': 0.5265436957323172, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Fractal swing breaks (daily) — these continue分形摆动突破（日线）——延续型
CONFIRMED已确认
On the daily chart the story flips: confirmed swing breaks continue over five days (≈ +17 pips holdout). Intraday EUR/USD fades; daily EUR/USD trends. The timeframe is the regime. n = 78 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.285 ATR, holdout +0.283 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
在日线图上故事反转：被确认的摆动突破在五天内延续（留出期约 +17 点）。日内欧元/美元回吐，日线欧元/美元趋势。时间框架决定市场状态。样本 n = 78。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.285 ATR，留出期 +0.283 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.2 per month (≈ 27 per year) · Dukascopy EUR/USD M15→D1 derived · n = 78频率：每月约 2.2 次（每年约 27 次） · n = 78
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=78.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=78。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-10-30T00:00:00+00:00', 'bar_index': 541, 'dir': -1.0, 'primary_ret_ATR': 0.3331534803433213, 'primary_ret_usd': 0.0021199999999998997, 'path_rmse_vs_median': 0.5951522759676103, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-20T00:00:00+00:00', 'bar_index': 338, 'dir': 1.0, 'primary_ret_ATR': 0.9046061335283152, 'primary_ret_usd': 0.007560000000000011, 'path_rmse_vs_median': 0.6651649015063612, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-31T00:00:00+00:00', 'bar_index': 86, 'dir': -1.0, 'primary_ret_ATR': 0.7182315429160238, 'primary_ret_usd': 0.0047600000000000975, 'path_rmse_vs_median': 0.6951504937603773, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-10-30T00:00:00+00:00', 'bar_index': 541, 'dir': -1.0, 'primary_ret_ATR': 0.3331534803433213, 'primary_ret_usd': 0.0021199999999998997, 'path_rmse_vs_median': 0.5951522759676103, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-20T00:00:00+00:00', 'bar_index': 338, 'dir': 1.0, 'primary_ret_ATR': 0.9046061335283152, 'primary_ret_usd': 0.007560000000000011, 'path_rmse_vs_median': 0.6651649015063612, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-31T00:00:00+00:00', 'bar_index': 86, 'dir': -1.0, 'primary_ret_ATR': 0.7182315429160238, 'primary_ret_usd': 0.0047600000000000975, 'path_rmse_vs_median': 0.6951504937603773, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Prior-week high/low breaks — accepted前周高/低突破——被接受
DEAD已否定
n = 73 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.147 ATR, holdout -0.632 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 73。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.147 ATR，留出期 -0.632 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.1 per month (≈ 25 per year) · Dukascopy EUR/USD M15→D1 derived · n = 73频率：每月约 2.1 次（每年约 25 次） · n = 73
Prior-week high/low breaks — rejected前周高/低突破——被拒绝
WEAK弱
n = 63 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.163 ATR, holdout -0.103 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 63。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.163 ATR，留出期 -0.103 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.8 per month (≈ 22 per year) · Dukascopy EUR/USD M15→D1 derived · n = 63频率：每月约 1.8 次（每年约 22 次） · n = 63
Candlesticks in ContextK线形态
Pin bars at a level价位处的针形线
DEAD已否定
n = 7583 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.072 ATR, holdout -0.059 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 7583。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.072 ATR，留出期 -0.059 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 208.6 per month (≈ 2503 per year) · Dukascopy EUR/USD M15 · n = 7583频率：每月约 208.6 次（每年约 2503 次） · n = 7583
Pin bars mid-range区间中部的针形线
DEAD已否定
n = 3956 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.084 ATR, holdout -0.019 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 3956。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.084 ATR，留出期 -0.019 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 108.8 per month (≈ 1306 per year) · Dukascopy EUR/USD M15 · n = 3956频率：每月约 108.8 次（每年约 1306 次） · n = 3956
Engulfing candles at a level价位处的吞没形态
CONFIRMED — NEGATIVE已确认 — 负向
n = 8353 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.032 ATR, holdout -0.054 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 8353。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.032 ATR，留出期 -0.054 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 229.8 per month (≈ 2757 per year) · Dukascopy EUR/USD M15 · n = 8353频率：每月约 229.8 次（每年约 2757 次） · n = 8353
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=8353.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=8353。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-05-24T20:00:00+00:00', 'bar_index': 16216, 'dir': -1.0, 'primary_ret_ATR': None, 'primary_ret_usd': None, 'path_rmse_vs_median': 0.5259783629478024, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-03-04T15:15:00+00:00', 'bar_index': 60453, 'dir': 1.0, 'primary_ret_ATR': -0.7955489583160672, 'primary_ret_usd': -0.0009999999999998899, 'path_rmse_vs_median': 0.5359161701792351, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-17T23:00:00+00:00', 'bar_index': 38596, 'dir': 1.0, 'primary_ret_ATR': 0.24433723127564047, 'primary_ret_usd': 0.00018999999999991246, 'path_rmse_vs_median': 0.5572801816510768, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-05-24T20:00:00+00:00', 'bar_index': 16216, 'dir': -1.0, 'primary_ret_ATR': None, 'primary_ret_usd': None, 'path_rmse_vs_median': 0.5259783629478024, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-03-04T15:15:00+00:00', 'bar_index': 60453, 'dir': 1.0, 'primary_ret_ATR': -0.7955489583160672, 'primary_ret_usd': -0.0009999999999998899, 'path_rmse_vs_median': 0.5359161701792351, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-17T23:00:00+00:00', 'bar_index': 38596, 'dir': 1.0, 'primary_ret_ATR': 0.24433723127564047, 'primary_ret_usd': 0.00018999999999991246, 'path_rmse_vs_median': 0.5572801816510768, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Engulfing candles mid-range区间中部的吞没形态
DEAD已否定
n = 3488 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.015 ATR, holdout -0.019 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 3488。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.015 ATR，留出期 -0.019 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 96.0 per month (≈ 1152 per year) · Dukascopy EUR/USD M15 · n = 3488频率：每月约 96.0 次（每年约 1152 次） · n = 3488
Inside bars at a level价位处的内包线
DEAD已否定
n = 6468 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.076 ATR, holdout +0.003 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 6468。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.076 ATR，留出期 +0.003 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 178.0 per month (≈ 2135 per year) · Dukascopy EUR/USD M15 · n = 6468频率：每月约 178.0 次（每年约 2135 次） · n = 6468
Inside bars mid-range区间中部的内包线
WEAK弱
n = 4554 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.028 ATR, holdout -0.022 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 4554。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.028 ATR，留出期 -0.022 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 125.3 per month (≈ 1504 per year) · Dukascopy EUR/USD M15 · n = 4554频率：每月约 125.3 次（每年约 1504 次） · n = 4554
Pin bars at strict levels only仅严格价位处的针形线
CONFIRMED — NEGATIVE已确认 — 负向
n = 4419 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.028 ATR, holdout -0.050 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 4419。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.028 ATR，留出期 -0.050 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 121.6 per month (≈ 1459 per year) · Dukascopy EUR/USD M15 · n = 4419频率：每月约 121.6 次（每年约 1459 次） · n = 4419
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-03-15T15:45:00+00:00', 'bar_index': 11399, 'dir': 1.0, 'primary_ret_ATR': -0.7440371137008831, 'primary_ret_usd': -0.00038999999999989043, 'path_rmse_vs_median': 0.4882396368904419, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T07:15:00+00:00', 'bar_index': 30853, 'dir': -1.0, 'primary_ret_ATR': 0.3105828425914497, 'primary_ret_usd': 0.00010999999999983245, 'path_rmse_vs_median': 0.5293766368093231, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-18T15:00:00+00:00', 'bar_index': 71876, 'dir': -1.0, 'primary_ret_ATR': 0.16485648470475425, 'primary_ret_usd': 8.999999999992347e-05, 'path_rmse_vs_median': 0.5378575288869594, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-03-15T15:45:00+00:00', 'bar_index': 11399, 'dir': 1.0, 'primary_ret_ATR': -0.7440371137008831, 'primary_ret_usd': -0.00038999999999989043, 'path_rmse_vs_median': 0.4882396368904419, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T07:15:00+00:00', 'bar_index': 30853, 'dir': -1.0, 'primary_ret_ATR': 0.3105828425914497, 'primary_ret_usd': 0.00010999999999983245, 'path_rmse_vs_median': 0.5293766368093231, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-18T15:00:00+00:00', 'bar_index': 71876, 'dir': -1.0, 'primary_ret_ATR': 0.16485648470475425, 'primary_ret_usd': 8.999999999992347e-05, 'path_rmse_vs_median': 0.5378575288869594, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Pin bars away from strict levels远离严格价位的针形线
DEAD已否定
n = 7120 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.031 ATR, holdout -0.087 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 7120。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.031 ATR，留出期 -0.087 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 195.9 per month (≈ 2350 per year) · Dukascopy EUR/USD M15 · n = 7120频率：每月约 195.9 次（每年约 2350 次） · n = 7120
Engulfing at strict levels only仅严格价位处的吞没形态
WEAK弱
n = 5340 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.038 ATR, holdout -0.044 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 5340。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.038 ATR，留出期 -0.044 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 146.9 per month (≈ 1763 per year) · Dukascopy EUR/USD M15 · n = 5340频率：每月约 146.9 次（每年约 1763 次） · n = 5340
Engulfing away from strict levels远离严格价位的吞没形态
CONFIRMED — NEGATIVE已确认 — 负向
Even held to gold's strict-level standard, EUR/USD engulfing leans slightly AGAINST its own direction — the opposite of gold's sturdiest behaviour, and at ≈0.3 pips far too small to trade. n = 6501 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.019 ATR, holdout -0.062 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
即使用黄金的严格价位标准衡量，欧元/美元的吞没形态也略微逆自身方向运行——与黄金最稳固的行为恰好相反，且幅度约 0.3 点，小到不可交易。样本 n = 6501。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.019 ATR，留出期 -0.062 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 178.8 per month (≈ 2146 per year) · Dukascopy EUR/USD M15 · n = 6501频率：每月约 178.8 次（每年约 2146 次） · n = 6501
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-08-06T22:00:00+00:00', 'bar_index': 71136, 'dir': 1.0, 'primary_ret_ATR': -0.0849123720205196, 'primary_ret_usd': -2.999999999997449e-05, 'path_rmse_vs_median': 0.44719196345937445, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-17T23:30:00+00:00', 'bar_index': 38598, 'dir': 1.0, 'primary_ret_ATR': 0.14082813967573357, 'primary_ret_usd': 9.999999999998899e-05, 'path_rmse_vs_median': 0.48187865164633215, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-06-21T14:00:00+00:00', 'bar_index': 18112, 'dir': 1.0, 'primary_ret_ATR': -0.20706892221601217, 'primary_ret_usd': -0.00015999999999993797, 'path_rmse_vs_median': 0.5041297710519687, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-08-06T22:00:00+00:00', 'bar_index': 71136, 'dir': 1.0, 'primary_ret_ATR': -0.0849123720205196, 'primary_ret_usd': -2.999999999997449e-05, 'path_rmse_vs_median': 0.44719196345937445, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-17T23:30:00+00:00', 'bar_index': 38598, 'dir': 1.0, 'primary_ret_ATR': 0.14082813967573357, 'primary_ret_usd': 9.999999999998899e-05, 'path_rmse_vs_median': 0.48187865164633215, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-06-21T14:00:00+00:00', 'bar_index': 18112, 'dir': 1.0, 'primary_ret_ATR': -0.20706892221601217, 'primary_ret_usd': -0.00015999999999993797, 'path_rmse_vs_median': 0.5041297710519687, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Trend & Momentum趋势与动量
3-bar close streaks (M15)3 根收盘连串（M15）
WEAK弱
n = 8889 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.065 ATR, holdout -0.045 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 8889。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.065 ATR，留出期 -0.045 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 244.5 per month (≈ 2934 per year) · Dukascopy EUR/USD M15 · n = 8889频率：每月约 244.5 次（每年约 2934 次） · n = 8889
4-bar close streaks (M15)4 根收盘连串（M15）
CONFIRMED — NEGATIVE已确认 — 负向
n = 4202 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.023 ATR, holdout -0.067 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 4202。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.023 ATR，留出期 -0.067 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 115.6 per month (≈ 1387 per year) · Dukascopy EUR/USD M15 · n = 4202频率：每月约 115.6 次（每年约 1387 次） · n = 4202
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-03-17T17:45:00+00:00', 'bar_index': 61327, 'dir': 1.0, 'primary_ret_ATR': -0.474409357930404, 'primary_ret_usd': -0.000400000000000178, 'path_rmse_vs_median': 0.4177897360437875, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T05:45:00+00:00', 'bar_index': 30847, 'dir': -1.0, 'primary_ret_ATR': 0.19609220836914687, 'primary_ret_usd': 4.999999999988347e-05, 'path_rmse_vs_median': 0.4453095567706815, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-12-11T18:45:00+00:00', 'bar_index': 4883, 'dir': 1.0, 'primary_ret_ATR': 0.16395323697285197, 'primary_ret_usd': 0.00011999999999989797, 'path_rmse_vs_median': 0.5629239756607765, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-03-17T17:45:00+00:00', 'bar_index': 61327, 'dir': 1.0, 'primary_ret_ATR': -0.474409357930404, 'primary_ret_usd': -0.000400000000000178, 'path_rmse_vs_median': 0.4177897360437875, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T05:45:00+00:00', 'bar_index': 30847, 'dir': -1.0, 'primary_ret_ATR': 0.19609220836914687, 'primary_ret_usd': 4.999999999988347e-05, 'path_rmse_vs_median': 0.4453095567706815, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-12-11T18:45:00+00:00', 'bar_index': 4883, 'dir': 1.0, 'primary_ret_ATR': 0.16395323697285197, 'primary_ret_usd': 0.00011999999999989797, 'path_rmse_vs_median': 0.5629239756607765, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
5-bar close streaks (M15)5 根收盘连串（M15）
DEAD已否定
n = 1945 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.078 ATR, holdout -0.003 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 1945。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.078 ATR，留出期 -0.003 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 53.5 per month (≈ 643 per year) · Dukascopy EUR/USD M15 · n = 1945频率：每月约 53.5 次（每年约 643 次） · n = 1945
3-day close streaks — they fade3 日收盘连串——随后回吐
CONFIRMED — NEGATIVE已确认 — 负向
Three same-direction daily closes are followed by five days that lean the other way (≈ −10 pips holdout). In EUR/USD a streak is a crowd to fade, not a train to board. n = 96 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.107 ATR, holdout -0.123 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
连续三天同向收盘后，随后五天倾向于反向运行（留出期约 −10 点）。在欧元/美元里，连串是可反打的人群，而不是可搭乘的列车。样本 n = 96。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.107 ATR，留出期 -0.123 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.8 per month (≈ 33 per year) · Dukascopy EUR/USD M15→D1 derived · n = 96频率：每月约 2.8 次（每年约 33 次） · n = 96
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=96.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=96。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-04-23T00:00:00+00:00', 'bar_index': 665, 'dir': -1.0, 'primary_ret_ATR': -0.6289334147277875, 'primary_ret_usd': -0.004630000000000134, 'path_rmse_vs_median': 0.5566031008990269, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-08-07T00:00:00+00:00', 'bar_index': 481, 'dir': 1.0, 'primary_ret_ATR': -0.15560751716789487, 'primary_ret_usd': -0.0014400000000001079, 'path_rmse_vs_median': 0.6929218848124064, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-02T00:00:00+00:00', 'bar_index': 65, 'dir': -1.0, 'primary_ret_ATR': 0.12376192812636316, 'primary_ret_usd': 0.0008900000000000574, 'path_rmse_vs_median': 0.8529964520569935, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-04-23T00:00:00+00:00', 'bar_index': 665, 'dir': -1.0, 'primary_ret_ATR': -0.6289334147277875, 'primary_ret_usd': -0.004630000000000134, 'path_rmse_vs_median': 0.5566031008990269, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-08-07T00:00:00+00:00', 'bar_index': 481, 'dir': 1.0, 'primary_ret_ATR': -0.15560751716789487, 'primary_ret_usd': -0.0014400000000001079, 'path_rmse_vs_median': 0.6929218848124064, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-02T00:00:00+00:00', 'bar_index': 65, 'dir': -1.0, 'primary_ret_ATR': 0.12376192812636316, 'primary_ret_usd': 0.0008900000000000574, 'path_rmse_vs_median': 0.8529964520569935, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
4-day close streaks4 日收盘连串
DEAD已否定
n = 49 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.277 ATR, holdout +0.005 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 49。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.277 ATR，留出期 +0.005 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.4 per month (≈ 17 per year) · Dukascopy EUR/USD M15→D1 derived · n = 49频率：每月约 1.4 次（每年约 17 次） · n = 49
5-day close streaks5 日收盘连串
WEAK弱
n = 21 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.479 ATR, holdout +0.356 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 21。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.479 ATR，留出期 +0.356 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.6 per month (≈ 8 per year) · Dukascopy EUR/USD M15→D1 derived · n = 21频率：每月约 0.6 次（每年约 8 次） · n = 21
EMA20 pullbacks (H1)EMA20 回调（H1）
WEAK弱
n = 266 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.129 ATR, holdout -0.023 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 266。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.129 ATR，留出期 -0.023 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 7.4 per month (≈ 89 per year) · Dukascopy EUR/USD M15→H1 derived · n = 266频率：每月约 7.4 次（每年约 89 次） · n = 266
EMA50 pullbacks (H1)EMA50 回调（H1）
CONFIRMED — NEGATIVE已确认 — 负向
n = 307 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.076 ATR, holdout -0.100 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 307。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.076 ATR，留出期 -0.100 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 8.6 per month (≈ 103 per year) · Dukascopy EUR/USD M15→H1 derived · n = 307频率：每月约 8.6 次（每年约 103 次） · n = 307
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-05-14T16:00:00+00:00', 'bar_index': 10098, 'dir': 1.0, 'primary_ret_ATR': -0.5118500379514523, 'primary_ret_usd': -0.0010900000000000354, 'path_rmse_vs_median': 0.7080584198974079, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-25T07:00:00+00:00', 'bar_index': 5097, 'dir': -1.0, 'primary_ret_ATR': -0.3332916556395501, 'primary_ret_usd': -0.000280000000000058, 'path_rmse_vs_median': 0.8331250008096756, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-24T14:00:00+00:00', 'bar_index': 13936, 'dir': 1.0, 'primary_ret_ATR': -0.1239717561339533, 'primary_ret_usd': -0.00011999999999989797, 'path_rmse_vs_median': 0.849317863827401, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-05-14T16:00:00+00:00', 'bar_index': 10098, 'dir': 1.0, 'primary_ret_ATR': -0.5118500379514523, 'primary_ret_usd': -0.0010900000000000354, 'path_rmse_vs_median': 0.7080584198974079, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-25T07:00:00+00:00', 'bar_index': 5097, 'dir': -1.0, 'primary_ret_ATR': -0.3332916556395501, 'primary_ret_usd': -0.000280000000000058, 'path_rmse_vs_median': 0.8331250008096756, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-24T14:00:00+00:00', 'bar_index': 13936, 'dir': 1.0, 'primary_ret_ATR': -0.1239717561339533, 'primary_ret_usd': -0.00011999999999989797, 'path_rmse_vs_median': 0.849317863827401, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
EMA20 pullbacks (daily)EMA20 回调（日线）
DEAD已否定
n = 9 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.322 ATR, holdout -0.601 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 9。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.322 ATR，留出期 -0.601 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.3 per month (≈ 4 per year) · Dukascopy EUR/USD M15→D1 derived · n = 9频率：每月约 0.3 次（每年约 4 次） · n = 9
EMA50 pullbacks (daily)EMA50 回调（日线）
DEAD已否定
n = 16 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.379 ATR, holdout +0.566 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 16。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.379 ATR，留出期 +0.566 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.6 per month (≈ 7 per year) · Dukascopy EUR/USD M15→D1 derived · n = 16频率：每月约 0.6 次（每年约 7 次） · n = 16
After a +0.5% day+0.5% 大涨日之后
DEAD已否定
n = 75 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.142 ATR, holdout -0.054 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 75。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.142 ATR，留出期 -0.054 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.3 per month (≈ 27 per year) · Dukascopy EUR/USD M15→D1 derived · n = 75频率：每月约 2.3 次（每年约 27 次） · n = 75
After a −0.5% day−0.5% 大跌日之后
DEAD已否定
n = 62 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.111 ATR, holdout -0.506 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 62。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.111 ATR，留出期 -0.506 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.8 per month (≈ 22 per year) · Dukascopy EUR/USD M15→D1 derived · n = 62频率：每月约 1.8 次（每年约 22 次） · n = 62
After a +1% day+1% 大涨日之后
DEAD已否定
n = 15 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +1.096 ATR, holdout -0.531 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 15。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +1.096 ATR，留出期 -0.531 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy EUR/USD M15→D1 derived · n = 15频率：每月约 0.5 次（每年约 6 次） · n = 15
After a −1% day−1% 大跌日之后
DEAD已否定
n = 9 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.170 ATR, holdout +0.332 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 9。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.170 ATR，留出期 +0.332 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.4 per month (≈ 5 per year) · Dukascopy EUR/USD M15→D1 derived · n = 9频率：每月约 0.4 次（每年约 5 次） · n = 9
Day Types日类型
Mechanical day-type census on the 788 derived daily bars: RANGE 31.1%, QUIET OTHER 30.1%, TREND UP 18.3%, TREND DOWN 17.0%, EXPANSION OTHER 3.0%, SPIKE REVERSE UP 0.4%, SPIKE REVERSE DOWN 0.1%. Six days in ten are range or quiet — EUR/USD's default day is a sideways day, which is the soil the fade behaviours grow in. Trend days are the minority and, per the streak results, they cluster rather than persist.
在 788 根派生日线上的机械日类型普查：RANGE 31.1%, QUIET OTHER 30.1%, TREND UP 18.3%, TREND DOWN 17.0%, EXPANSION OTHER 3.0%, SPIKE REVERSE UP 0.4%, SPIKE REVERSE DOWN 0.1%。十天里有六天是区间或安静日——欧元/美元的默认日是横盘日，这正是回吐型行为生长的土壤。趋势日是少数，且按连串结果看，它们扎堆出现而非持续。
London's first hour → the rest of the day伦敦第一小时 → 当日其余时间
WEAK弱
n = 770 instances. Median forward effect vs the matched random null at the primary horizon (rest-of-day): discovery +0.016 ATR, holdout +0.044 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 770。在主要周期（rest-of-day）相对匹配随机基线的中位前向效应：发现期 +0.016 ATR，留出期 +0.044 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.2 per month (≈ 254 per year) · Dukascopy EUR/USD M15 · n = 770频率：每月约 21.2 次（每年约 254 次） · n = 770
Time & Volatility时间与波动
Hour of day (D1, descriptive). The day's range is manufactured in the London and Overlap hours; bar ranges shrink through the US afternoon and bottom in the late evening. Asia is the quietest full session but not a dead one (median session range 18 pips).
Asia width (D2, descriptive). Days are split by Asia-range tercile. Narrow-Asia days (median width 0.18 prior-day ATR) still finish with a median full-day range of 0.85 ATR — 4.7× the Asia width — and close beyond the Asia range 89.9% of the time. A narrow Asia is a coiled day, not a small day.
Weekday (D3) and ATR regime (E1) are descriptive conditioning studies in the census files: no weekday carries a tradeable directional claim, and forward volatility rises with the trailing ATR regime while forward direction does not.
n = 58 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.220 ATR, holdout -0.564 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 58。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.220 ATR，留出期 -0.564 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.7 per month (≈ 20 per year) · Dukascopy EUR/USD M15→D1 derived · n = 58频率：每月约 1.7 次（每年约 20 次） · n = 58
Compression episodes (daily)压缩区间（日线）
DEAD已否定
n = 27 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.021 ATR, holdout +0.654 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 27。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.021 ATR，留出期 +0.654 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.9 per month (≈ 11 per year) · Dukascopy EUR/USD M15→D1 derived · n = 27频率：每月约 0.9 次（每年约 11 次） · n = 27
Bollinger-width squeezes (daily)布林带宽挤压（日线）
WEAK弱
n = 12 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.188 ATR, holdout +0.189 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 12。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.188 ATR，留出期 +0.189 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.4 per month (≈ 5 per year) · Dukascopy EUR/USD M15→D1 derived · n = 12频率：每月约 0.4 次（每年约 5 次） · n = 12
After >2×ATR days — continuation-signed>2×ATR 大日之后——延续方向
WEAK弱
n = 29 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.359 ATR, holdout +0.204 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 29。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.359 ATR，留出期 +0.204 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.8 per month (≈ 10 per year) · Dukascopy EUR/USD M15→D1 derived · n = 29频率：每月约 0.8 次（每年约 10 次） · n = 29
After >2×ATR days — |return|>2×ATR 大日之后——波动幅度
DEAD已否定
n = 29 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.145 ATR, holdout -0.294 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 29。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.145 ATR，留出期 -0.294 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.8 per month (≈ 10 per year) · Dukascopy EUR/USD M15→D1 derived · n = 29频率：每月约 0.8 次（每年约 10 次） · n = 29
Mean Reversion均值回归
Stretched H1 moves snap backH1 过度延伸后的回拉
CONFIRMED已确认
When an H1 move stretches ≥ 2× ATR away from its EMA50, the next two hours snap back toward it (+0.21 ATR holdout, ≈ +2.4 pips, n=1,088). One of the few effects in this volume that clears the cost bar with room. n = 1088 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.140 ATR, holdout +0.209 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 30.0 per month (≈ 360 per year) · Dukascopy EUR/USD M15→H1 derived · n = 1088频率：每月约 30.0 次（每年约 360 次） · n = 1088
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=1088.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=1088。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-11-17T14:00:00+00:00', 'bar_index': 13288, 'dir': 1.0, 'primary_ret_ATR': 1.1287935062293695, 'primary_ret_usd': 0.0011599999999998278, 'path_rmse_vs_median': 0.6193719448911236, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-03-04T14:00:00+00:00', 'bar_index': 2632, 'dir': -1.0, 'primary_ret_ATR': -0.1687804268598045, 'primary_ret_usd': -0.00017000000000000348, 'path_rmse_vs_median': 0.6391049557308275, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-26T16:00:00+00:00', 'bar_index': 18114, 'dir': 1.0, 'primary_ret_ATR': 0.20633541523651244, 'primary_ret_usd': 0.00016000000000016001, 'path_rmse_vs_median': 0.6617305285987426, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-11-17T14:00:00+00:00', 'bar_index': 13288, 'dir': 1.0, 'primary_ret_ATR': 1.1287935062293695, 'primary_ret_usd': 0.0011599999999998278, 'path_rmse_vs_median': 0.6193719448911236, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-03-04T14:00:00+00:00', 'bar_index': 2632, 'dir': -1.0, 'primary_ret_ATR': -0.1687804268598045, 'primary_ret_usd': -0.00017000000000000348, 'path_rmse_vs_median': 0.6391049557308275, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-26T16:00:00+00:00', 'bar_index': 18114, 'dir': 1.0, 'primary_ret_ATR': 0.20633541523651244, 'primary_ret_usd': 0.00016000000000016001, 'path_rmse_vs_median': 0.6617305285987426, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
London range reversion in US hours美国时段对伦敦区间的回归
CONFIRMED已确认
n = 745 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.004 ATR, holdout +0.084 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 745。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.004 ATR，留出期 +0.084 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 20.5 per month (≈ 246 per year) · Dukascopy EUR/USD M15 · n = 745频率：每月约 20.5 次（每年约 246 次） · n = 745
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-08-29T14:00:00+00:00', 'bar_index': 22816, 'dir': 1.0, 'primary_ret_ATR': 0.8976202078292742, 'primary_ret_usd': 0.0008099999999999774, 'path_rmse_vs_median': 0.5533155081182837, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-10-13T12:30:00+00:00', 'bar_index': 50746, 'dir': 1.0, 'primary_ret_ATR': -0.6514592837785953, 'primary_ret_usd': -0.0005599999999998939, 'path_rmse_vs_median': 0.5816900140561575, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-16T15:45:00+00:00', 'bar_index': 67559, 'dir': -1.0, 'primary_ret_ATR': 0.38455617804416237, 'primary_ret_usd': 0.00026999999999999247, 'path_rmse_vs_median': 0.6114552399544325, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-08-29T14:00:00+00:00', 'bar_index': 22816, 'dir': 1.0, 'primary_ret_ATR': 0.8976202078292742, 'primary_ret_usd': 0.0008099999999999774, 'path_rmse_vs_median': 0.5533155081182837, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-10-13T12:30:00+00:00', 'bar_index': 50746, 'dir': 1.0, 'primary_ret_ATR': -0.6514592837785953, 'primary_ret_usd': -0.0005599999999998939, 'path_rmse_vs_median': 0.5816900140561575, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-16T15:45:00+00:00', 'bar_index': 67559, 'dir': -1.0, 'primary_ret_ATR': 0.38455617804416237, 'primary_ret_usd': 0.00026999999999999247, 'path_rmse_vs_median': 0.6114552399544325, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Daily RSI(2) extremes, faded日线 RSI(2）极端——反打
DEAD已否定
n = 58 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.321 ATR, holdout +0.254 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 58。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.321 ATR，留出期 +0.254 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.7 per month (≈ 20 per year) · Dukascopy EUR/USD M15→D1 derived · n = 58频率：每月约 1.7 次（每年约 20 次） · n = 58
Daily RSI(14) extremes, faded日线 RSI(14）极端——反打
DEAD已否定
n = 26 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.055 ATR, holdout +0.527 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 26。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.055 ATR，留出期 +0.527 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.8 per month (≈ 9 per year) · Dukascopy EUR/USD M15→D1 derived · n = 26频率：每月约 0.8 次（每年约 9 次） · n = 26
Macro Events — the US side宏观事件——美国一侧
Coverage limit, stated plainly: the desk's release calendar holds US events only (CPI, PPI, NFP, Retail Sales at 08:30 ET; FOMC at 14:00 ET). ECB and euro-area releases are not in this volume — half of this pair's event life is unmeasured here, and the chapters below are the US half.
Range anatomy. At the release minute the median M15 bar range is 20.4 pips vs 7.9 on matched non-event clocks (2.6×), and is still elevated thirty minutes later (12.6 vs 8.3). The event is a volatility fact before it is a direction fact.
Yield conditioning (F7, descriptive). On event days when the US 10Y moved ≥ 5bp, EUR/USD closed opposite the yield move 77.3% of the time (n=75) vs a 54.9% quiet-day baseline: on release days the yield → dollar → EUR chain actually engages. Intraday, EUR/USD's 2h return after 08:30 releases opposes the ICE DXY's move over the same window with the mirror sign, as it should if both are the same dollar trade.
The F6 warning. 'Hot' and 'cold' below use a crude proxy — the released month-on-month change vs the prior month's change, NOT a consensus surprise. Under that proxy every hot/cold trade is DEAD or fragile in this market, exactly as in gold: do not build a 'short the hot print' rule on it.
All 08:30 releases — the first spike is faded全部 08:30 发布——第一波被回吐
CONFIRMED — NEGATIVE已确认 — 负向
Pooled across CPI/PPI/NFP/Retail: the release spike's direction is systematically overdone — the next two hours give back −0.92 ATR against it in the holdout (≈ −4.3 pips, n=142). The same headline behaviour gold shows; here it is measured on the other side of the pair. n = 142 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.826 ATR, holdout -0.920 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 4.0 per month (≈ 47 per year) · Dukascopy EUR/USD M15 · n = 142频率：每月约 4.0 次（每年约 47 次） · n = 142
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=142.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=142。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-05-08T12:30:00+00:00', 'bar_index': 64954, 'dir': 1.0, 'primary_ret_ATR': -0.5524696757861631, 'primary_ret_usd': -0.0003500000000000725, 'path_rmse_vs_median': 0.4177361043241976, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-05T12:30:00+00:00', 'bar_index': 19066, 'dir': 1.0, 'primary_ret_ATR': -1.1788260288778867, 'primary_ret_usd': -0.0007899999999998464, 'path_rmse_vs_median': 0.7102045077511613, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-11-15T13:30:00+00:00', 'bar_index': 3134, 'dir': -1.0, 'primary_ret_ATR': -0.8648431043554519, 'primary_ret_usd': -0.0008099999999999774, 'path_rmse_vs_median': 0.8775883514592431, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-05-08T12:30:00+00:00', 'bar_index': 64954, 'dir': 1.0, 'primary_ret_ATR': -0.5524696757861631, 'primary_ret_usd': -0.0003500000000000725, 'path_rmse_vs_median': 0.4177361043241976, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-05T12:30:00+00:00', 'bar_index': 19066, 'dir': 1.0, 'primary_ret_ATR': -1.1788260288778867, 'primary_ret_usd': -0.0007899999999998464, 'path_rmse_vs_median': 0.7102045077511613, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-11-15T13:30:00+00:00', 'bar_index': 3134, 'dir': -1.0, 'primary_ret_ATR': -0.8648431043554519, 'primary_ret_usd': -0.0008099999999999774, 'path_rmse_vs_median': 0.8775883514592431, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
CPI releases — the first spikeCPI 发布——第一波
WEAK弱
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.848 ATR, holdout -1.067 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.848 ATR，留出期 -1.067 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15 · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
PPI releases — the first spikePPI 发布——第一波
WEAK弱
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -1.534 ATR, holdout -1.473 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -1.534 ATR，留出期 -1.473 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15 · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
NFP releases — the first spike非农发布——第一波
WEAK弱
n = 36 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -1.172 ATR, holdout -0.545 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 36。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -1.172 ATR，留出期 -0.545 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15 · n = 36频率：每月约 1.0 次（每年约 12 次） · n = 36
Retail Sales releases — the first spike零售销售发布——第一波
DEAD已否定
n = 36 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.451 ATR, holdout -0.043 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 36。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.451 ATR，留出期 -0.043 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15 · n = 36频率：每月约 1.0 次（每年约 12 次） · n = 36
FOMC releases — the first spikeFOMC 发布——第一波
WEAK弱
n = 24 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.032 ATR, holdout +2.968 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 24。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.032 ATR，留出期 +2.968 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy EUR/USD M15 · n = 24频率：每月约 0.7 次（每年约 8 次） · n = 24
CPI days — direction persistenceCPI 日——方向持续性
WEAK弱
n = 33 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.424 ATR, holdout +0.047 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 33。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.424 ATR，留出期 +0.047 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15→D1 derived · n = 33频率：每月约 1.0 次（每年约 12 次） · n = 33
PPI days — direction persistencePPI 日——方向持续性
WEAK弱
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.603 ATR, holdout +0.397 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.603 ATR，留出期 +0.397 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15→D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
NFP days — direction persistence非农日——方向持续性
DEAD已否定
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.113 ATR, holdout -0.454 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.113 ATR，留出期 -0.454 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15→D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
Retail Sales days — direction persistence零售销售日——方向持续性
DEAD已否定
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.524 ATR, holdout +0.767 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.524 ATR，留出期 +0.767 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy EUR/USD M15→D1 derived · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
FOMC days — direction persistenceFOMC 日——方向持续性
WEAK弱
n = 23 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.582 ATR, holdout +1.247 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 23。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.582 ATR，留出期 +1.247 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy EUR/USD M15→D1 derived · n = 23频率：每月约 0.7 次（每年约 8 次） · n = 23
After 'hot' prints (crude proxy)'热'数据之后（粗略代理）
DEAD已否定
n = 69 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.052 ATR, holdout +0.329 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 69。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.052 ATR，留出期 +0.329 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.0 per month (≈ 24 per year) · Dukascopy EUR/USD M15→D1 derived · n = 69频率：每月约 2.0 次（每年约 24 次） · n = 69
After 'cold' prints (crude proxy)'冷'数据之后（粗略代理）
DEAD已否定
n = 66 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.154 ATR, holdout -0.556 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 66。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.154 ATR，留出期 -0.556 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.9 per month (≈ 23 per year) · Dukascopy EUR/USD M15→D1 derived · n = 66频率：每月约 1.9 次（每年约 23 次） · n = 66
After 'hot' CPI (crude proxy)'热'CPI 之后（粗略代理）
WEAK弱
n = 17 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.126 ATR, holdout -0.681 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 17。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.126 ATR，留出期 -0.681 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy EUR/USD M15→D1 derived · n = 17频率：每月约 0.5 次（每年约 6 次） · n = 17
After 'cold' CPI (crude proxy)'冷'CPI 之后（粗略代理）
DEAD已否定
n = 15 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.212 ATR, holdout -0.852 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 15。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.212 ATR，留出期 -0.852 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy EUR/USD M15→D1 derived · n = 15频率：每月约 0.5 次（每年约 6 次） · n = 15
Cross-Asset States跨资产状态
Broad-dollar spike days广义美元冲高日
WEAK弱
n = 27 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.229 ATR, holdout +0.194 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 27。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.229 ATR，留出期 +0.194 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.8 per month (≈ 10 per year) · Dukascopy EUR/USD M15→D1 derived · n = 27频率：每月约 0.8 次（每年约 10 次） · n = 27
ICE DXY spike days (2y window)ICE 美元指数冲高日（2 年窗口）
WEAK弱
n = 49 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.247 ATR, holdout -0.506 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 49。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.247 ATR，留出期 -0.506 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.3 per month (≈ 27 per year) · Dukascopy EUR/USD M15→D1 derived · n = 49频率：每月约 2.3 次（每年约 27 次） · n = 49
10Y yield spike days10 年期收益率冲高日
DEAD已否定
n = 41 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.002 ATR, holdout +0.679 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 41。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.002 ATR，留出期 +0.679 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.2 per month (≈ 14 per year) · Dukascopy EUR/USD M15→D1 derived · n = 41频率：每月约 1.2 次（每年约 14 次） · n = 41
Quiet cross-asset days (baseline control)跨资产安静日（基线对照）
DESCRIPTIVE描述性
n = 157 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.017 ATR, holdout -0.211 ATR. Verdict: DESCRIPTIVE under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 157。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.017 ATR，留出期 -0.211 ATR。按章程规则判定为：DESCRIPTIVE（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 4.5 per month (≈ 54 per year) · Dukascopy EUR/USD M15→D1 derived · n = 157频率：每月约 4.5 次（每年约 54 次） · n = 157
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-10-21T00:00:00+00:00', 'bar_index': 534, 'dir': 1.0, 'primary_ret_ATR': 0.7147459931814056, 'primary_ret_usd': 0.0048399999999999554, 'path_rmse_vs_median': 0.6062914502638638, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-05T00:00:00+00:00', 'bar_index': 673, 'dir': 1.0, 'primary_ret_ATR': 0.6674907367951001, 'primary_ret_usd': 0.004450000000000065, 'path_rmse_vs_median': 0.635971366973733, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-30T00:00:00+00:00', 'bar_index': 85, 'dir': 1.0, 'primary_ret_ATR': -1.4115303616156323, 'primary_ret_usd': -0.009070000000000134, 'path_rmse_vs_median': 0.772234601671093, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-10-21T00:00:00+00:00', 'bar_index': 534, 'dir': 1.0, 'primary_ret_ATR': 0.7147459931814056, 'primary_ret_usd': 0.0048399999999999554, 'path_rmse_vs_median': 0.6062914502638638, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-05T00:00:00+00:00', 'bar_index': 673, 'dir': 1.0, 'primary_ret_ATR': 0.6674907367951001, 'primary_ret_usd': 0.004450000000000065, 'path_rmse_vs_median': 0.635971366973733, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-01-30T00:00:00+00:00', 'bar_index': 85, 'dir': 1.0, 'primary_ret_ATR': -1.4115303616156323, 'primary_ret_usd': -0.009070000000000134, 'path_rmse_vs_median': 0.772234601671093, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Sessions in Their Own Right时段本身
Session anatomy, 788 trading days (medians). The Overlap is the widest session (32 pips), then London and New York (29 each); even Asia carries 18. In every session the high and the low most often form in the first or last 30 minutes — extremes live at the edges, the same shape gold's census found.
Session
Median range (pips)
Median net (pips)
Share closing up
Median CLV
Overnight (23:00-02:00 UTC)
14
-0
0.49
0.48
Asia (02:00-07:00 UTC)
18
-0
0.48
0.49
London (07:00-12:00 UTC)
29
-0
0.49
0.48
Overlap (12:00-15:00 UTC)
32
+1
0.53
0.53
NewYork (15:00-20:00 UTC)
29
-1
0.48
0.48
OffHours (20:00-23:00 UTC)
9
+0
0.51
0.56
First-hour breaks. An accepted break of a session's first-hour range is CONFIRMED negative in five of six sessions (holdout −0.13…−0.69 ATR; New York's is DEAD): the opening break mean-reverts into the session close. Rejections of the first-hour range are mostly DEAD — only one session's rejection survives (CONFIRMED +0.18).
Prior-session extremes (Sd). The strongest single effect in this family is the rejection of the Overlap range during the Overnight session (+0.88 ATR holdout, n=144); the same rejection in Asia and London also confirms (+0.34/+0.38). New York's range, by contrast, is a level whose rejection helps in one touching session (+0.57) and hurts in another (−0.61) — which session does the touching decides the sign. Full per-session tables: census files S/<session>/.
The direction of London's first half hour tends to survive to the London close (+0.44 ATR holdout, n=785) — the opposite lean from gold's first-30 mark in Volume I. n = 785 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery +0.170 ATR, holdout +0.441 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 21.6 per month (≈ 259 per year) · Dukascopy EUR/USD M15 · n = 785频率：每月约 21.6 次（每年约 259 次） · n = 785
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=785.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=785。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-12-25T07:15:00+00:00', 'bar_index': 30853, 'dir': -1.0, 'primary_ret_ATR': 0.3105828425914497, 'primary_ret_usd': 0.00010999999999983245, 'path_rmse_vs_median': 0.38693161794062625, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-18T07:15:00+00:00', 'bar_index': 71845, 'dir': 1.0, 'primary_ret_ATR': 0.5236108455335926, 'primary_ret_usd': 0.00019999999999997797, 'path_rmse_vs_median': 0.9212796924442965, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-11-05T07:15:00+00:00', 'bar_index': 52357, 'dir': -1.0, 'primary_ret_ATR': -0.6370921263709797, 'primary_ret_usd': -0.0002300000000001745, 'path_rmse_vs_median': 1.1450358958756193, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-12-25T07:15:00+00:00', 'bar_index': 30853, 'dir': -1.0, 'primary_ret_ATR': 0.3105828425914497, 'primary_ret_usd': 0.00010999999999983245, 'path_rmse_vs_median': 0.38693161794062625, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-18T07:15:00+00:00', 'bar_index': 71845, 'dir': 1.0, 'primary_ret_ATR': 0.5236108455335926, 'primary_ret_usd': 0.00019999999999997797, 'path_rmse_vs_median': 0.9212796924442965, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-11-05T07:15:00+00:00', 'bar_index': 52357, 'dir': -1.0, 'primary_ret_ATR': -0.6370921263709797, 'primary_ret_usd': -0.0002300000000001745, 'path_rmse_vs_median': 1.1450358958756193, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Asia → London handoff — reversed亚洲 → 伦敦交接——反转
CONFIRMED — NEGATIVE已确认 — 负向
Asia's session direction is, on median, given back across London (−0.41 ATR holdout). In EUR/USD the handoff is a fade, not a baton pass. n = 783 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery -0.247 ATR, holdout -0.410 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
亚洲时段的方向在中位数上会被伦敦时段回吐（留出期 −0.41 ATR）。在欧元/美元里，交接是反打，而不是接力。样本 n = 783。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 -0.247 ATR，留出期 -0.410 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 259 per year) · Dukascopy EUR/USD M15 · n = 783频率：每月约 21.5 次（每年约 259 次） · n = 783
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-12-25T06:45:00+00:00', 'bar_index': 55811, 'dir': 1.0, 'primary_ret_ATR': -0.30400471757082337, 'primary_ret_usd': -6.000000000017103e-05, 'path_rmse_vs_median': 0.7660848119219538, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T06:45:00+00:00', 'bar_index': 30851, 'dir': -1.0, 'primary_ret_ATR': 0.7120605707481368, 'primary_ret_usd': 0.00019999999999997797, 'path_rmse_vs_median': 0.8932931645514597, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-11T06:45:00+00:00', 'bar_index': 71363, 'dir': -1.0, 'primary_ret_ATR': 0.18972674222272912, 'primary_ret_usd': 6.000000000017103e-05, 'path_rmse_vs_median': 0.9067283063895901, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-12-25T06:45:00+00:00', 'bar_index': 55811, 'dir': 1.0, 'primary_ret_ATR': -0.30400471757082337, 'primary_ret_usd': -6.000000000017103e-05, 'path_rmse_vs_median': 0.7660848119219538, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T06:45:00+00:00', 'bar_index': 30851, 'dir': -1.0, 'primary_ret_ATR': 0.7120605707481368, 'primary_ret_usd': 0.00019999999999997797, 'path_rmse_vs_median': 0.8932931645514597, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-11T06:45:00+00:00', 'bar_index': 71363, 'dir': -1.0, 'primary_ret_ATR': 0.18972674222272912, 'primary_ret_usd': 6.000000000017103e-05, 'path_rmse_vs_median': 0.9067283063895901, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Off-hours → Overnight handoff盘后 → 隔夜交接
CONFIRMED — NEGATIVE已确认 — 负向
n = 784 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery -0.127 ATR, holdout -0.372 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 784。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 -0.127 ATR，留出期 -0.372 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.6 per month (≈ 260 per year) · Dukascopy EUR/USD M15 · n = 784频率：每月约 21.6 次（每年约 260 次） · n = 784
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2023-10-27T21:45:00+00:00', 'bar_index': 1919, 'dir': -1.0, 'primary_ret_ATR': -0.6045637047678858, 'primary_ret_usd': -0.000320000000000098, 'path_rmse_vs_median': 0.6589115958750079, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-06T22:45:00+00:00', 'bar_index': 71139, 'dir': -1.0, 'primary_ret_ATR': 0.03316404942700111, 'primary_ret_usd': 9.999999999843467e-06, 'path_rmse_vs_median': 0.6639962935030698, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-29T22:45:00+00:00', 'bar_index': 33219, 'dir': 1.0, 'primary_ret_ATR': -0.10941906114682236, 'primary_ret_usd': -9.000000000014552e-05, 'path_rmse_vs_median': 0.6749856502924385, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2023-10-27T21:45:00+00:00', 'bar_index': 1919, 'dir': -1.0, 'primary_ret_ATR': -0.6045637047678858, 'primary_ret_usd': -0.000320000000000098, 'path_rmse_vs_median': 0.6589115958750079, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-06T22:45:00+00:00', 'bar_index': 71139, 'dir': -1.0, 'primary_ret_ATR': 0.03316404942700111, 'primary_ret_usd': 9.999999999843467e-06, 'path_rmse_vs_median': 0.6639962935030698, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-29T22:45:00+00:00', 'bar_index': 33219, 'dir': 1.0, 'primary_ret_ATR': -0.10941906114682236, 'primary_ret_usd': -9.000000000014552e-05, 'path_rmse_vs_median': 0.6749856502924385, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
New York → Off-hours handoff纽约 → 盘后交接
CONFIRMED已确认
n = 784 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery +0.073 ATR, holdout +0.088 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 784。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 +0.073 ATR，留出期 +0.088 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.6 per month (≈ 259 per year) · Dukascopy EUR/USD M15 · n = 784频率：每月约 21.6 次（每年约 259 次） · n = 784
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-03-14T19:45:00+00:00', 'bar_index': 36279, 'dir': 1.0, 'primary_ret_ATR': -0.8848148231395312, 'primary_ret_usd': -0.0006399999999999739, 'path_rmse_vs_median': 0.5461944761986663, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-26T19:45:00+00:00', 'bar_index': 68343, 'dir': -1.0, 'primary_ret_ATR': 0.6568556686246191, 'primary_ret_usd': 0.00041000000000002146, 'path_rmse_vs_median': 0.5482174363921927, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-22T19:45:00+00:00', 'bar_index': 9879, 'dir': 1.0, 'primary_ret_ATR': -0.24875846617670386, 'primary_ret_usd': -0.000140000000000029, 'path_rmse_vs_median': 0.580610456148189, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-03-14T19:45:00+00:00', 'bar_index': 36279, 'dir': 1.0, 'primary_ret_ATR': -0.8848148231395312, 'primary_ret_usd': -0.0006399999999999739, 'path_rmse_vs_median': 0.5461944761986663, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-26T19:45:00+00:00', 'bar_index': 68343, 'dir': -1.0, 'primary_ret_ATR': 0.6568556686246191, 'primary_ret_usd': 0.00041000000000002146, 'path_rmse_vs_median': 0.5482174363921927, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-22T19:45:00+00:00', 'bar_index': 9879, 'dir': 1.0, 'primary_ret_ATR': -0.24875846617670386, 'primary_ret_usd': -0.000140000000000029, 'path_rmse_vs_median': 0.580610456148189, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Asia midpoint mark亚洲中点标记
CONFIRMED — NEGATIVE已确认 — 负向
n = 785 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery -0.182 ATR, holdout -0.111 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 785。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 -0.182 ATR，留出期 -0.111 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.6 per month (≈ 259 per year) · Dukascopy EUR/USD M15 · n = 785频率：每月约 21.6 次（每年约 259 次） · n = 785
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-01-28T04:15:00+00:00', 'bar_index': 33049, 'dir': -1.0, 'primary_ret_ATR': -1.084946688472574, 'primary_ret_usd': -0.0009299999999998754, 'path_rmse_vs_median': 0.7910857061545957, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-25T04:15:00+00:00', 'bar_index': 55801, 'dir': 1.0, 'primary_ret_ATR': 0.299661378809415, 'primary_ret_usd': 7.00000000000145e-05, 'path_rmse_vs_median': 0.8360085918828694, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-23T04:15:00+00:00', 'bar_index': 9913, 'dir': 1.0, 'primary_ret_ATR': -0.7925287862511803, 'primary_ret_usd': -0.00025999999999992696, 'path_rmse_vs_median': 1.2398996722393079, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-01-28T04:15:00+00:00', 'bar_index': 33049, 'dir': -1.0, 'primary_ret_ATR': -1.084946688472574, 'primary_ret_usd': -0.0009299999999998754, 'path_rmse_vs_median': 0.7910857061545957, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-25T04:15:00+00:00', 'bar_index': 55801, 'dir': 1.0, 'primary_ret_ATR': 0.299661378809415, 'primary_ret_usd': 7.00000000000145e-05, 'path_rmse_vs_median': 0.8360085918828694, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-23T04:15:00+00:00', 'bar_index': 9913, 'dir': 1.0, 'primary_ret_ATR': -0.7925287862511803, 'primary_ret_usd': -0.00025999999999992696, 'path_rmse_vs_median': 1.2398996722393079, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Overnight first-hour mark隔夜第一小时标记
CONFIRMED — NEGATIVE已确认 — 负向
n = 777 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery -0.062 ATR, holdout -0.061 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 777。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 -0.062 ATR，留出期 -0.061 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.4 per month (≈ 257 per year) · Dukascopy EUR/USD M15 · n = 777频率：每月约 21.4 次（每年约 257 次） · n = 777
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-04-17T23:45:00+00:00', 'bar_index': 38599, 'dir': 1.0, 'primary_ret_ATR': -0.08935115435703798, 'primary_ret_usd': -6.000000000017103e-05, 'path_rmse_vs_median': 0.5133913728519557, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-03-10T23:45:00+00:00', 'bar_index': 10951, 'dir': 1.0, 'primary_ret_ATR': -0.523629090635273, 'primary_ret_usd': -0.00021000000000004349, 'path_rmse_vs_median': 0.6219106437834421, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-02-15T23:45:00+00:00', 'bar_index': 59239, 'dir': -1.0, 'primary_ret_ATR': 0.3009092824631624, 'primary_ret_usd': 0.00014999999999987246, 'path_rmse_vs_median': 0.6259781070584424, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-04-17T23:45:00+00:00', 'bar_index': 38599, 'dir': 1.0, 'primary_ret_ATR': -0.08935115435703798, 'primary_ret_usd': -6.000000000017103e-05, 'path_rmse_vs_median': 0.5133913728519557, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-03-10T23:45:00+00:00', 'bar_index': 10951, 'dir': 1.0, 'primary_ret_ATR': -0.523629090635273, 'primary_ret_usd': -0.00021000000000004349, 'path_rmse_vs_median': 0.6219106437834421, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-02-15T23:45:00+00:00', 'bar_index': 59239, 'dir': -1.0, 'primary_ret_ATR': 0.3009092824631624, 'primary_ret_usd': 0.00014999999999987246, 'path_rmse_vs_median': 0.6259781070584424, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
What EUR/USD Teaches Gold欧元/美元对黄金的启示
The filter question is settled — negatively. Gold's confirmed behaviours were re-measured split by EUR/USD's state at the event (this is the tradable form of the desk's old DXY-filter question; EUR/USD is the largest DXY component). Gold's accepted breaks (n=323): the 'tailwind' advantage flips between discovery (+0.25 ATR) and holdout (−0.17 ATR). Gold's mid-range engulfing (n=1,997): identical in aligned and opposed states (+0.028 vs +0.025 ATR). The filter does not survive its holdout and is recorded as dead in the gold chapters (logged write-back, 2026-10-15).
Co-movement. Same-session returns (own-ATR units) correlate most where the dollar is busiest: Overlap r = 0.50, New York 0.48, London 0.36, Asia 0.33. No lead-lag pair reaches |r| ≥ 0.08 — neither market warns the other; they share one impulse at the same time.
Event twins. On 79 matched 08:30 releases the first-spike signs agree 82% of the time, and both markets fade the spike over the next two hours. After EUR/USD trend-down days, gold's next day leans up (+0.13 ATR median, n=82, descriptive) — a haven echo, noted, not promoted.
The central contrast. Same definitions, opposite characters: gold's accepted prior-day breaks continue (Volume I); EUR/USD's reverse (−0.36 ATR). Any desk rule that assumes 'markets break out the same way' is wrong in at least one of these two markets.
Cross-learning addendum — the yield chain from the other side (POSEIDON Vol. III, 2026-10-16)跨市场学习附记——从另一侧看收益率链条（POSEIDON 第三卷，2026-10-16）
CROSS-LEARNING跨市场
Volume II measured EUR/USD moving OPPOSITE the US 10Y yield on 77.3% of big-yield event days (F7). Volume III measured the same chain from the other side of the dollar: on US event days with a ≥ 5bp 10Y move, USD/JPY moved WITH the yield 84.2% of the time (n=76) and with the broad dollar 87.0% (n=46); yields up ≥ 5bp → the pair's median day was +72.8 pips, yields down → −106.3 pips. The two measurements are consistent — one dollar chain, opposite pair signs — and the transmission engages more cleanly in USD/JPY than in EUR/USD. No Volume II finding is changed; the F7 chapter's chain now has both ends measured.
Cost audit. Measured on this corpus: median M15 ATR 5.5 pips, median daily range 62 pips (788 days). At a typical retail spread of 0.6–1.0 pips plus 0.3 slippage, a round trip costs ≈ 1.1 pips = 0.20× the median M15 ATR — about three times gold's per-ATR friction (0.07×, re-measured in this volume). In R terms: an 8-pip stop pays 0.138R per trade, a 12-pip stop 0.092R, a 20-pip stop 0.055R. Consequence: this volume's CONFIRMED effects below ~1 pip (all of the candle family, M15 swing breaks) are measurements, not trades. The effects that clear friction with room are the fades — the 08:30 spike (−4.3 pips), Overlap-range rejections (+4.3), rejected prior-day breaks (+2.2), H1 extensions (+2.4) — and the daily swing break (+17 pips over 5 days).
Methods. Definitions were fixed in writing before measurement (Volume II DEFINITIONS). Sessions are fixed UTC windows — real London/New York clock times shift ±1h with DST, and session-edge results inherit that approximation, identically to Volume I. Post-midnight bars belong to the prior evening's trading day (22:00 UTC roll); no behaviour is grouped by calendar date against a same-date range — the look-ahead error that invalidated earlier desk backtests is excluded by construction and by assertion in code.
Limits. (1) The daily sample is three years, not gold's ten — daily verdicts carry less weight, and much of the window was an unusual Fed/ECB-divergence regime. (2) ECB/euro-area events are unmeasured (US calendar only). (3) With 141 behaviours, a few CONFIRMED labels are expected to be false positives by chance alone; instance counts and effect sizes, not labels, are the evidence. (4) Census numbers are gross of costs and are NOT approved strategies — anything trade-shaped must still pass the desk's readiness bar (mechanical definition, costed backtest, ≥50 trades, PF ≥ 1.3 out-of-sample, ≥2 market periods).
方法。定义在测量之前以书面固定（第二卷 DEFINITIONS）。时段为固定 UTC 窗口——伦敦/纽约的真实钟表时间随夏令时 ±1 小时移动，时段边缘结果继承这一近似，与第一卷相同。午夜后的 K 线属于前一晚的交易日（22:00 UTC 滚动）；没有任何行为按日历日期与同日区间分组——曾使本桌早期回测失效的前视错误，在构造与代码断言两个层面被排除。
31CONFIRMED — survived unseen data已确认——通过未见数据检验31WEAK — a flag, not a finding弱——只是标记，不是结论65DEAD — killed by the holdout已死亡——被留出数据杀掉17DESCRIPTIVE — measured, no direction claim描述性——只测量，不下方向结论
How wide each session runs每个时段跑多宽Median session range in pips · 788 days. Asia is a FULL session here — Tokyo is home turf.时段区间中位数（点）· 788 天。亚洲在这里是完整的时段——东京是主场。
Off-hours非交易时段
19p
Asia (Tokyo)亚洲（东京）
41p
New York纽约
42p
Overnight隔夜
42p
London伦敦
46p
Overlap重叠
50p
The friction tax — cost of one round trip摩擦税——一次往返的成本Round-trip cost as a fraction of one median M15 candle (ATR). Smaller = cheaper to trade.往返成本占一根典型 M15 K 线（ATR）的比例。越小 = 交易成本越低。
Gold黄金
0.07×
NAS100纳斯达克100
0.096×
USD/JPY美元/日元
0.12×
EUR/USD欧元/美元
0.20×
One day on the clock时钟上的一天Session blocks on the desk clock (UTC). Edge segments — first and last 30 minutes — are where the highs and lows form.交易台时钟（UTC）的时段区块。边缘时段——最初和最后 30 分钟——是高低点形成的地方。
Off-hrs非交易时段
19p
20–23 UTC
Overnight隔夜
42p
23–02
Asia亚洲
41p
02–07
London伦敦
46p
07–12
Overlap重叠
50p
12–15
New York纽约
42p
15–20
Laws that survived all four markets在四个市场都活下来的规律
📰
The first spike lies第一波冲高会说谎
On 08:30 ET data days the first spike is faded in gold, EUR/USD and USD/JPY. Never chase it.在 08:30 ET 数据日，第一波冲高在黄金、欧元/美元和美元/日元中都会被反打。永远不要追它。
📉
Streaks snap back连涨连跌会弹回
Three same-direction days lean the other way over the next five — in all four markets.连续三天同方向，接下来五天会倾向反方向——四个市场全部如此。
🧭
Breaks continue — except EUR/USD突破会延续——欧元/美元除外
An accepted break of yesterday’s range keeps going in gold, USD/JPY and NAS100. In EUR/USD it reverses.被接受的昨日区间突破在黄金、美元/日元和纳斯达克100中会继续。在欧元/美元中它会反转。
⏱
Edges make the extremes边缘产生极值
Session highs and lows form in the first or last 30 minutes — everywhere, in every market.时段高低点在最初或最后 30 分钟形成——每个市场、每个地方都如此。
The headline behaviours — look first, read later头条行为——先看图，后读字
The break that sticks守得住的突破
A closing break of yesterday’s range keeps travelling for 2 hours. Small — but the sign is gold’s sign, not EUR/USD’s.收盘突破昨日区间后会继续走 2 小时。幅度小——但方向符号是黄金的，不是欧元/美元的。
The first spike lies — third market in a row第一波冲高说谎——连续第三个市场
The pooled 08:30 ET first spike is faded over 2 hours here too. Gold, EUR/USD, now USD/JPY — the spike fade is a POSEIDON law.汇总的 08:30 ET 第一波冲高在这里也会在 2 小时内被反打。黄金、欧元/美元、现在美元/日元——冲高反打是 POSEIDON 的规律。
On big-yield-move event days this pair moved WITH the 10-year yield 84% of the time. Yields up ≥5bp → pair median +72.8 pips that day.在大收益率波动的事件日，这对货币有 84% 的时间与 10 年期收益率同向。收益率上升 ≥5 个基点 → 当日中位数 +72.8 点。
Three same-direction days are followed by five days leaning the other way — the largest tradeable-scale effect in this volume.连续三天同方向后，接下来五天倾向反方向——本卷最大的可交易量级效应。
Rejections of the Asia range during a later session resolve in the rejection direction — the biggest single-session rejection effect measured here.之后时段对亚洲区间的拒绝会沿拒绝方向走完——这里测到的最大单时段拒绝效应。
The full chapters follow below — every number above is drawn from them, unchanged.完整章节在下方——上面的每一个数字都来自它们，未作任何改动。
Start Here从这里开始
POSEIDON — Volume III: USD/JPY. The same observation-first census as Volumes I (gold) and II (EUR/USD), now on the dollar's other great pair — the haven twin. Feed: Dukascopy USD/JPY M15 (aggregated from M1 BID), 2023-10-01 → 2026-10-12 — 106,368 bars, 788 valid trading days, one uniform corpus; the manifest records 0 duplicate timestamps, 0 partial bars, 0 gaps. 144 behaviours were defined mechanically before measurement (round levels at 0.50 steps — the whole and half big figures this pair actually watches), extracted completely, stacked against session- and direction-matched random nulls, and judged on a chronological 60/40 discovery/holdout split: 31 CONFIRMED, 31 WEAK, 65 DEAD, 17 DESCRIPTIVE. All effects are gross of trading costs; the cost audit is its own chapter.
The one-sentence character of USD/JPY: gold's haven twin, the dollar's yield slave. Accepted prior-day breaks continue here as they do in gold — not fade as in EUR/USD — and on US event days the pair moves with the 10-year yield 84% of the time. But inside sessions it is still an FX pair: first-hour breakouts fade, handoffs reverse, and daily streaks snap back hard.
Accepted breaks of yesterday's range — they continue昨日区间被接受突破——随后延续
CONFIRMED已确认
The three-market contrast, measured: where EUR/USD's accepted prior-day breaks reverse (−0.359 ATR holdout), USD/JPY's keep travelling (+0.112), the same sign as gold's. The effect is small (≈ +1.4 pips) but it survives its holdout — this pair resolves prior-day breaks by continuation, not exhaustion. n = 524 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.016 ATR, holdout +0.112 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 14.4 per month (≈ 173 per year) · Dukascopy USD/JPY M15 · n = 524频率：每月约 14.4 次（每年约 173 次） · n = 524
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=524.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=524。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-04-07T14:00:00+00:00', 'bar_index': 37792, 'dir': 1.0, 'primary_ret_ATR': 0.4607895038978186, 'primary_ret_usd': 0.17399999999997817, 'path_rmse_vs_median': 0.6807904712255818, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-18T04:00:00+00:00', 'bar_index': 71832, 'dir': 1.0, 'primary_ret_ATR': 0.35168477831737904, 'primary_ret_usd': 0.02199999999999136, 'path_rmse_vs_median': 0.7699529062244161, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-10-11T14:45:00+00:00', 'bar_index': 739, 'dir': 1.0, 'primary_ret_ATR': 0.14241591365242903, 'primary_ret_usd': 0.014000000000010004, 'path_rmse_vs_median': 0.845769135177345, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-04-07T14:00:00+00:00', 'bar_index': 37792, 'dir': 1.0, 'primary_ret_ATR': 0.4607895038978186, 'primary_ret_usd': 0.17399999999997817, 'path_rmse_vs_median': 0.6807904712255818, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-18T04:00:00+00:00', 'bar_index': 71832, 'dir': 1.0, 'primary_ret_ATR': 0.35168477831737904, 'primary_ret_usd': 0.02199999999999136, 'path_rmse_vs_median': 0.7699529062244161, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-10-11T14:45:00+00:00', 'bar_index': 739, 'dir': 1.0, 'primary_ret_ATR': 0.14241591365242903, 'primary_ret_usd': 0.014000000000010004, 'path_rmse_vs_median': 0.845769135177345, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Rejected breaks of yesterday's range昨日区间被拒绝突破
CONFIRMED已确认
A wick beyond yesterday's high/low that closes back inside keeps travelling in the rejection direction (+0.220 ATR holdout, ≈ +1.8 pips). Prior-day levels resolve cleanly in both directions in this pair. n = 400 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.312 ATR, holdout +0.220 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 11.2 per month (≈ 134 per year) · Dukascopy USD/JPY M15 · n = 400频率：每月约 11.2 次（每年约 134 次） · n = 400
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=400.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=400。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-05-20T15:15:00+00:00', 'bar_index': 65733, 'dir': 1.0, 'primary_ret_ATR': 0.4971536858433875, 'primary_ret_usd': 0.04800000000000182, 'path_rmse_vs_median': 0.539322856329799, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-09T03:45:00+00:00', 'bar_index': 8951, 'dir': -1.0, 'primary_ret_ATR': 0.5361795699422817, 'primary_ret_usd': 0.038000000000010914, 'path_rmse_vs_median': 0.7315236746368593, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-14T09:30:00+00:00', 'bar_index': 32110, 'dir': -1.0, 'primary_ret_ATR': 0.37652467905249404, 'primary_ret_usd': 0.0589999999999975, 'path_rmse_vs_median': 0.8585769251100756, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-05-20T15:15:00+00:00', 'bar_index': 65733, 'dir': 1.0, 'primary_ret_ATR': 0.4971536858433875, 'primary_ret_usd': 0.04800000000000182, 'path_rmse_vs_median': 0.539322856329799, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-09T03:45:00+00:00', 'bar_index': 8951, 'dir': -1.0, 'primary_ret_ATR': 0.5361795699422817, 'primary_ret_usd': 0.038000000000010914, 'path_rmse_vs_median': 0.7315236746368593, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-14T09:30:00+00:00', 'bar_index': 32110, 'dir': -1.0, 'primary_ret_ATR': 0.37652467905249404, 'primary_ret_usd': 0.0589999999999975, 'path_rmse_vs_median': 0.8585769251100756, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Asia range breaks accepted in London伦敦时段接受亚洲区间突破
DEAD已否定
n = 421 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.082 ATR, holdout -0.126 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 421。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.082 ATR，留出期 -0.126 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 11.6 per month (≈ 139 per year) · Dukascopy USD/JPY M15 · n = 421频率：每月约 11.6 次（每年约 139 次） · n = 421
Asia range sweeps rejected in London伦敦时段拒绝亚洲区间扫破
CONFIRMED — NEGATIVE已确认 — 负向
n = 367 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.138 ATR, holdout -0.161 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 367。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.138 ATR，留出期 -0.161 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 10.1 per month (≈ 122 per year) · Dukascopy USD/JPY M15 · n = 367频率：每月约 10.1 次（每年约 122 次） · n = 367
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=367.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=367。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2023-12-25T07:45:00+00:00', 'bar_index': 5799, 'dir': -1.0, 'primary_ret_ATR': -0.0, 'primary_ret_usd': -0.0, 'path_rmse_vs_median': 0.48396746524704054, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-01-26T08:30:00+00:00', 'bar_index': 57834, 'dir': 1.0, 'primary_ret_ATR': -0.7075806428293638, 'primary_ret_usd': -0.18799999999998818, 'path_rmse_vs_median': 0.7305426091759809, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T07:45:00+00:00', 'bar_index': 30855, 'dir': 1.0, 'primary_ret_ATR': 0.0, 'primary_ret_usd': 0.0, 'path_rmse_vs_median': 0.768333894498572, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2023-12-25T07:45:00+00:00', 'bar_index': 5799, 'dir': -1.0, 'primary_ret_ATR': -0.0, 'primary_ret_usd': -0.0, 'path_rmse_vs_median': 0.48396746524704054, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-01-26T08:30:00+00:00', 'bar_index': 57834, 'dir': 1.0, 'primary_ret_ATR': -0.7075806428293638, 'primary_ret_usd': -0.18799999999998818, 'path_rmse_vs_median': 0.7305426091759809, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-12-25T07:45:00+00:00', 'bar_index': 30855, 'dir': 1.0, 'primary_ret_ATR': 0.0, 'primary_ret_usd': 0.0, 'path_rmse_vs_median': 0.768333894498572, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
London range breaks accepted in US hours美国时段接受伦敦区间突破
DEAD已否定
n = 454 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.027 ATR, holdout +0.048 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 454。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.027 ATR，留出期 +0.048 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 12.5 per month (≈ 150 per year) · Dukascopy USD/JPY M15 · n = 454频率：每月约 12.5 次（每年约 150 次） · n = 454
London range sweeps rejected in US hours美国时段拒绝伦敦区间扫破
WEAK弱
n = 422 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.194 ATR, holdout -0.072 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 422。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.194 ATR，留出期 -0.072 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 11.6 per month (≈ 139 per year) · Dukascopy USD/JPY M15 · n = 422频率：每月约 11.6 次（每年约 139 次） · n = 422
Fractal swing breaks (M15)分形摆动突破（M15）
CONFIRMED — NEGATIVE已确认 — 负向
n = 6857 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.025 ATR, holdout -0.057 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 6857。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.025 ATR，留出期 -0.057 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 188.6 per month (≈ 2263 per year) · Dukascopy USD/JPY M15 · n = 6857频率：每月约 188.6 次（每年约 2263 次） · n = 6857
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-04-19T11:00:00+00:00', 'bar_index': 13780, 'dir': 1.0, 'primary_ret_ATR': -0.01148332186773873, 'primary_ret_usd': -0.0010000000000047748, 'path_rmse_vs_median': 0.4294642869537194, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-28T15:15:00+00:00', 'bar_index': 33093, 'dir': 1.0, 'primary_ret_ATR': 0.5334185762026732, 'primary_ret_usd': 0.09799999999998477, 'path_rmse_vs_median': 0.46108013882294185, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-26T16:45:00+00:00', 'bar_index': 68331, 'dir': 1.0, 'primary_ret_ATR': -0.343880508109807, 'primary_ret_usd': -0.018999999999977035, 'path_rmse_vs_median': 0.5385438701717036, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-04-19T11:00:00+00:00', 'bar_index': 13780, 'dir': 1.0, 'primary_ret_ATR': -0.01148332186773873, 'primary_ret_usd': -0.0010000000000047748, 'path_rmse_vs_median': 0.4294642869537194, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-28T15:15:00+00:00', 'bar_index': 33093, 'dir': 1.0, 'primary_ret_ATR': 0.5334185762026732, 'primary_ret_usd': 0.09799999999998477, 'path_rmse_vs_median': 0.46108013882294185, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-26T16:45:00+00:00', 'bar_index': 68331, 'dir': 1.0, 'primary_ret_ATR': -0.343880508109807, 'primary_ret_usd': -0.018999999999977035, 'path_rmse_vs_median': 0.5385438701717036, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Fractal swing breaks (daily) — dead here分形摆动突破（日线）——在此已被否定
DEAD已否定
The daily swing-break continuation that EUR/USD confirmed (+0.283 ATR) flips sign in USD/JPY (discovery +0.440 → holdout −0.294) and is DEAD. Daily continuation is not an FX universal. n = 77 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.440 ATR, holdout -0.294 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 2.2 per month (≈ 27 per year) · Dukascopy USD/JPY M15→D1 derived · n = 77频率：每月约 2.2 次（每年约 27 次） · n = 77
Prior-week high/low breaks — accepted前周高/低突破——被接受
WEAK弱
n = 63 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.081 ATR, holdout -0.408 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 63。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.081 ATR，留出期 -0.408 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.8 per month (≈ 22 per year) · Dukascopy USD/JPY M15→D1 derived · n = 63频率：每月约 1.8 次（每年约 22 次） · n = 63
Prior-week high/low breaks — rejected前周高/低突破——被拒绝
DEAD已否定
n = 84 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.055 ATR, holdout -0.163 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 84。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.055 ATR，留出期 -0.163 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.4 per month (≈ 29 per year) · Dukascopy USD/JPY M15→D1 derived · n = 84频率：每月约 2.4 次（每年约 29 次） · n = 84
Candlesticks in ContextK线形态
Pin bars at a level价位处的针形线
CONFIRMED — NEGATIVE已确认 — 负向
n = 8155 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.083 ATR, holdout -0.069 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 8155。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.083 ATR，留出期 -0.069 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 224.3 per month (≈ 2692 per year) · Dukascopy USD/JPY M15 · n = 8155频率：每月约 224.3 次（每年约 2692 次） · n = 8155
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-08-08T15:15:00+00:00', 'bar_index': 46341, 'dir': 1.0, 'primary_ret_ATR': -0.2474650762137011, 'primary_ret_usd': -0.028999999999996362, 'path_rmse_vs_median': 0.43650131926916674, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-04-16T16:30:00+00:00', 'bar_index': 13514, 'dir': 1.0, 'primary_ret_ATR': -0.08530301566707031, 'primary_ret_usd': -0.009999999999990905, 'path_rmse_vs_median': 0.52294954810607, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-23T16:00:00+00:00', 'bar_index': 68040, 'dir': -1.0, 'primary_ret_ATR': -0.3632949487806072, 'primary_ret_usd': -0.02199999999999136, 'path_rmse_vs_median': 0.5344243584176308, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-08-08T15:15:00+00:00', 'bar_index': 46341, 'dir': 1.0, 'primary_ret_ATR': -0.2474650762137011, 'primary_ret_usd': -0.028999999999996362, 'path_rmse_vs_median': 0.43650131926916674, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-04-16T16:30:00+00:00', 'bar_index': 13514, 'dir': 1.0, 'primary_ret_ATR': -0.08530301566707031, 'primary_ret_usd': -0.009999999999990905, 'path_rmse_vs_median': 0.52294954810607, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-23T16:00:00+00:00', 'bar_index': 68040, 'dir': -1.0, 'primary_ret_ATR': -0.3632949487806072, 'primary_ret_usd': -0.02199999999999136, 'path_rmse_vs_median': 0.5344243584176308, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Pin bars mid-range区间中部的针形线
DEAD已否定
n = 3511 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.036 ATR, holdout -0.015 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 3511。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.036 ATR，留出期 -0.015 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 96.6 per month (≈ 1159 per year) · Dukascopy USD/JPY M15 · n = 3511频率：每月约 96.6 次（每年约 1159 次） · n = 3511
Engulfing candles at a level价位处的吞没形态
DEAD已否定
n = 8674 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.005 ATR, holdout -0.007 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 8674。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.005 ATR，留出期 -0.007 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 238.6 per month (≈ 2864 per year) · Dukascopy USD/JPY M15 · n = 8674频率：每月约 238.6 次（每年约 2864 次） · n = 8674
Mid-range engulfing leans WITH its own direction (+0.057 ATR holdout) — the same sign as gold's sturdiest behaviour and the opposite of EUR/USD's. At ≈ +0.5 pips it is far too small to trade; as a character marker it matters. n = 3031 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.063 ATR, holdout +0.057 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 83.4 per month (≈ 1001 per year) · Dukascopy USD/JPY M15 · n = 3031频率：每月约 83.4 次（每年约 1001 次） · n = 3031
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=3031.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=3031。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-04-18T01:45:00+00:00', 'bar_index': 38607, 'dir': -1.0, 'primary_ret_ATR': 0.3432044954180887, 'primary_ret_usd': 0.03700000000000614, 'path_rmse_vs_median': 0.3354347690184587, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-04-16T18:00:00+00:00', 'bar_index': 13520, 'dir': -1.0, 'primary_ret_ATR': -0.018064620313381404, 'primary_ret_usd': -0.001999999999981128, 'path_rmse_vs_median': 0.5090525454037892, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-23T18:30:00+00:00', 'bar_index': 68050, 'dir': -1.0, 'primary_ret_ATR': -0.4218362642189616, 'primary_ret_usd': -0.01999999999998181, 'path_rmse_vs_median': 0.6388408367681204, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-04-18T01:45:00+00:00', 'bar_index': 38607, 'dir': -1.0, 'primary_ret_ATR': 0.3432044954180887, 'primary_ret_usd': 0.03700000000000614, 'path_rmse_vs_median': 0.3354347690184587, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-04-16T18:00:00+00:00', 'bar_index': 13520, 'dir': -1.0, 'primary_ret_ATR': -0.018064620313381404, 'primary_ret_usd': -0.001999999999981128, 'path_rmse_vs_median': 0.5090525454037892, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-23T18:30:00+00:00', 'bar_index': 68050, 'dir': -1.0, 'primary_ret_ATR': -0.4218362642189616, 'primary_ret_usd': -0.01999999999998181, 'path_rmse_vs_median': 0.6388408367681204, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Inside bars at a level价位处的内包线
WEAK弱
n = 7021 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.004 ATR, holdout -0.045 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 7021。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.004 ATR，留出期 -0.045 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 193.1 per month (≈ 2317 per year) · Dukascopy USD/JPY M15 · n = 7021频率：每月约 193.1 次（每年约 2317 次） · n = 7021
Inside bars mid-range区间中部的内包线
CONFIRMED — NEGATIVE已确认 — 负向
n = 4392 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.042 ATR, holdout -0.136 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 4392。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.042 ATR，留出期 -0.136 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 120.8 per month (≈ 1450 per year) · Dukascopy USD/JPY M15 · n = 4392频率：每月约 120.8 次（每年约 1450 次） · n = 4392
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-04-18T01:30:00+00:00', 'bar_index': 38606, 'dir': -1.0, 'primary_ret_ATR': 0.6547969529378327, 'primary_ret_usd': 0.0730000000000075, 'path_rmse_vs_median': 0.4519990818021987, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-09T15:00:00+00:00', 'bar_index': 8996, 'dir': -1.0, 'primary_ret_ATR': 0.317374767800558, 'primary_ret_usd': 0.04099999999999682, 'path_rmse_vs_median': 0.5090597522317751, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-07-22T14:15:00+00:00', 'bar_index': 70049, 'dir': 1.0, 'primary_ret_ATR': 0.3577215018660149, 'primary_ret_usd': 0.019000000000005457, 'path_rmse_vs_median': 0.5408916237876197, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-04-18T01:30:00+00:00', 'bar_index': 38606, 'dir': -1.0, 'primary_ret_ATR': 0.6547969529378327, 'primary_ret_usd': 0.0730000000000075, 'path_rmse_vs_median': 0.4519990818021987, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-02-09T15:00:00+00:00', 'bar_index': 8996, 'dir': -1.0, 'primary_ret_ATR': 0.317374767800558, 'primary_ret_usd': 0.04099999999999682, 'path_rmse_vs_median': 0.5090597522317751, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-07-22T14:15:00+00:00', 'bar_index': 70049, 'dir': 1.0, 'primary_ret_ATR': 0.3577215018660149, 'primary_ret_usd': 0.019000000000005457, 'path_rmse_vs_median': 0.5408916237876197, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Pin bars at strict levels only仅严格价位处的针形线
DEAD已否定
n = 4195 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.032 ATR, holdout -0.008 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 4195。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.032 ATR，留出期 -0.008 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 115.4 per month (≈ 1385 per year) · Dukascopy USD/JPY M15 · n = 4195频率：每月约 115.4 次（每年约 1385 次） · n = 4195
Pin bars away from strict levels远离严格价位的针形线
CONFIRMED — NEGATIVE已确认 — 负向
n = 7471 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.094 ATR, holdout -0.059 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 7471。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.094 ATR，留出期 -0.059 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 205.5 per month (≈ 2466 per year) · Dukascopy USD/JPY M15 · n = 7471频率：每月约 205.5 次（每年约 2466 次） · n = 7471
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-04-16T14:15:00+00:00', 'bar_index': 13505, 'dir': 1.0, 'primary_ret_ATR': 0.037857901197716044, 'primary_ret_usd': 0.0049999999999954525, 'path_rmse_vs_median': 0.4618833541376979, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-18T03:45:00+00:00', 'bar_index': 38615, 'dir': 1.0, 'primary_ret_ATR': 0.38438540196496696, 'primary_ret_usd': 0.03100000000000591, 'path_rmse_vs_median': 0.4906963228589626, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-03-17T17:30:00+00:00', 'bar_index': 61326, 'dir': 1.0, 'primary_ret_ATR': 0.5813846259094902, 'primary_ret_usd': 0.057999999999992724, 'path_rmse_vs_median': 0.5571006670877464, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-04-16T14:15:00+00:00', 'bar_index': 13505, 'dir': 1.0, 'primary_ret_ATR': 0.037857901197716044, 'primary_ret_usd': 0.0049999999999954525, 'path_rmse_vs_median': 0.4618833541376979, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-18T03:45:00+00:00', 'bar_index': 38615, 'dir': 1.0, 'primary_ret_ATR': 0.38438540196496696, 'primary_ret_usd': 0.03100000000000591, 'path_rmse_vs_median': 0.4906963228589626, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-03-17T17:30:00+00:00', 'bar_index': 61326, 'dir': 1.0, 'primary_ret_ATR': 0.5813846259094902, 'primary_ret_usd': 0.057999999999992724, 'path_rmse_vs_median': 0.5571006670877464, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Engulfing at strict levels only仅严格价位处的吞没形态
CONFIRMED — NEGATIVE已确认 — 负向
n = 5007 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.040 ATR, holdout -0.103 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 5007。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.040 ATR，留出期 -0.103 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 137.8 per month (≈ 1653 per year) · Dukascopy USD/JPY M15 · n = 5007频率：每月约 137.8 次（每年约 1653 次） · n = 5007
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-08-08T15:00:00+00:00', 'bar_index': 46340, 'dir': -1.0, 'primary_ret_ATR': 0.28267456673110203, 'primary_ret_usd': 0.033999999999991815, 'path_rmse_vs_median': 0.44897136056490894, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-12-05T16:15:00+00:00', 'bar_index': 4489, 'dir': 1.0, 'primary_ret_ATR': 0.7486162334977126, 'primary_ret_usd': 0.14300000000000068, 'path_rmse_vs_median': 0.4960451290150998, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-25T15:30:00+00:00', 'bar_index': 72358, 'dir': -1.0, 'primary_ret_ATR': -0.7595566533263063, 'primary_ret_usd': -0.05500000000000682, 'path_rmse_vs_median': 0.5413598090970444, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-08-08T15:00:00+00:00', 'bar_index': 46340, 'dir': -1.0, 'primary_ret_ATR': 0.28267456673110203, 'primary_ret_usd': 0.033999999999991815, 'path_rmse_vs_median': 0.44897136056490894, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-12-05T16:15:00+00:00', 'bar_index': 4489, 'dir': 1.0, 'primary_ret_ATR': 0.7486162334977126, 'primary_ret_usd': 0.14300000000000068, 'path_rmse_vs_median': 0.4960451290150998, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-08-25T15:30:00+00:00', 'bar_index': 72358, 'dir': -1.0, 'primary_ret_ATR': -0.7595566533263063, 'primary_ret_usd': -0.05500000000000682, 'path_rmse_vs_median': 0.5413598090970444, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Engulfing away from strict levels远离严格价位的吞没形态
DEAD已否定
n = 6698 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.008 ATR, holdout +0.039 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 6698。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.008 ATR，留出期 +0.039 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 184.3 per month (≈ 2211 per year) · Dukascopy USD/JPY M15 · n = 6698频率：每月约 184.3 次（每年约 2211 次） · n = 6698
Trend & Momentum趋势与动量
3-bar close streaks (M15)3 根收盘连串（M15）
WEAK弱
n = 9045 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.043 ATR, holdout -0.048 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 9045。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.043 ATR，留出期 -0.048 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 248.8 per month (≈ 2986 per year) · Dukascopy USD/JPY M15 · n = 9045频率：每月约 248.8 次（每年约 2986 次） · n = 9045
4-bar close streaks (M15)4 根收盘连串（M15）
DEAD已否定
n = 4296 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.019 ATR, holdout -0.104 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 4296。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.019 ATR，留出期 -0.104 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 118.2 per month (≈ 1418 per year) · Dukascopy USD/JPY M15 · n = 4296频率：每月约 118.2 次（每年约 1418 次） · n = 4296
5-bar close streaks (M15)5 根收盘连串（M15）
WEAK弱
n = 2032 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.048 ATR, holdout -0.050 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 2032。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.048 ATR，留出期 -0.050 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 55.9 per month (≈ 671 per year) · Dukascopy USD/JPY M15 · n = 2032频率：每月约 55.9 次（每年约 671 次） · n = 2032
3-day close streaks — they fade hard3 日收盘连串——随后大幅回吐
CONFIRMED — NEGATIVE已确认 — 负向
Three same-direction daily closes are followed by five days leaning the other way (−0.255 ATR holdout, ≈ −27.9 pips). The largest tradeable-scale effect in this volume, and the same sign as EUR/USD's streak fade: in both FX majors a daily streak is a crowd to fade. n = 91 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.197 ATR, holdout -0.255 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 2.7 per month (≈ 32 per year) · Dukascopy USD/JPY M15→D1 derived · n = 91频率：每月约 2.7 次（每年约 32 次） · n = 91
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=91.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=91。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-03-26T00:00:00+00:00', 'bar_index': 645, 'dir': 1.0, 'primary_ret_ATR': -0.10253700117966386, 'primary_ret_usd': -0.12000000000000455, 'path_rmse_vs_median': 0.5069599443681979, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-05-21T00:00:00+00:00', 'bar_index': 425, 'dir': -1.0, 'primary_ret_ATR': -0.6831333690353789, 'primary_ret_usd': -1.1409999999999911, 'path_rmse_vs_median': 0.5827482825837663, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-09-05T00:00:00+00:00', 'bar_index': 242, 'dir': -1.0, 'primary_ret_ATR': 0.846904769424761, 'primary_ret_usd': 1.570999999999998, 'path_rmse_vs_median': 0.7281428521134172, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-03-26T00:00:00+00:00', 'bar_index': 645, 'dir': 1.0, 'primary_ret_ATR': -0.10253700117966386, 'primary_ret_usd': -0.12000000000000455, 'path_rmse_vs_median': 0.5069599443681979, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-05-21T00:00:00+00:00', 'bar_index': 425, 'dir': -1.0, 'primary_ret_ATR': -0.6831333690353789, 'primary_ret_usd': -1.1409999999999911, 'path_rmse_vs_median': 0.5827482825837663, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-09-05T00:00:00+00:00', 'bar_index': 242, 'dir': -1.0, 'primary_ret_ATR': 0.846904769424761, 'primary_ret_usd': 1.570999999999998, 'path_rmse_vs_median': 0.7281428521134172, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
4-day close streaks4 日收盘连串
WEAK弱
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.120 ATR, holdout -0.687 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.120 ATR，留出期 -0.687 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
5-day close streaks5 日收盘连串
DEAD已否定
n = 18 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.013 ATR, holdout -0.343 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 18。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.013 ATR，留出期 -0.343 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy USD/JPY M15→D1 derived · n = 18频率：每月约 0.5 次（每年约 6 次） · n = 18
EMA20 pullbacks (H1)EMA20 回调（H1）
DEAD已否定
n = 247 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.080 ATR, holdout +0.039 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 247。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.080 ATR，留出期 +0.039 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 6.8 per month (≈ 82 per year) · Dukascopy USD/JPY M15→H1 derived · n = 247频率：每月约 6.8 次（每年约 82 次） · n = 247
EMA50 pullbacks (H1)EMA50 回调（H1）
DEAD已否定
n = 316 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.120 ATR, holdout -0.021 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 316。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.120 ATR，留出期 -0.021 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 8.7 per month (≈ 105 per year) · Dukascopy USD/JPY M15→H1 derived · n = 316频率：每月约 8.7 次（每年约 105 次） · n = 316
EMA20 pullbacks (daily)EMA20 回调（日线）
DEAD已否定
n = 9 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -1.035 ATR, holdout +0.865 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 9。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -1.035 ATR，留出期 +0.865 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.3 per month (≈ 4 per year) · Dukascopy USD/JPY M15→D1 derived · n = 9频率：每月约 0.3 次（每年约 4 次） · n = 9
EMA50 pullbacks (daily)EMA50 回调（日线）
DEAD已否定
n = 12 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.445 ATR, holdout -0.101 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 12。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.445 ATR，留出期 -0.101 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.4 per month (≈ 5 per year) · Dukascopy USD/JPY M15→D1 derived · n = 12频率：每月约 0.4 次（每年约 5 次） · n = 12
After a +0.5% day+0.5% 大涨日之后
DEAD已否定
n = 120 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.008 ATR, holdout -0.278 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 120。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.008 ATR，留出期 -0.278 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 3.5 per month (≈ 42 per year) · Dukascopy USD/JPY M15→D1 derived · n = 120频率：每月约 3.5 次（每年约 42 次） · n = 120
After a −0.5% day−0.5% 大跌日之后
DEAD已否定
n = 119 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.388 ATR, holdout -0.219 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 119。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.388 ATR，留出期 -0.219 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 3.5 per month (≈ 42 per year) · Dukascopy USD/JPY M15→D1 derived · n = 119频率：每月约 3.5 次（每年约 42 次） · n = 119
After a +1% day+1% 大涨日之后
DEAD已否定
n = 24 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.159 ATR, holdout -0.376 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 24。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.159 ATR，留出期 -0.376 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 24频率：每月约 1.0 次（每年约 12 次） · n = 24
After a −1% day−1% 大跌日之后
WEAK弱
n = 40 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.340 ATR, holdout +0.121 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 40。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.340 ATR，留出期 +0.121 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.2 per month (≈ 15 per year) · Dukascopy USD/JPY M15→D1 derived · n = 40频率：每月约 1.2 次（每年约 15 次） · n = 40
Day Types日类型
Mechanical day-type census on the 788 derived daily bars: RANGE 32.6%, QUIET OTHER 28.6%, TREND UP 20.6%, TREND DOWN 14.1%, EXPANSION OTHER 3.4%, SPIKE REVERSE ~0.9%. Trend-up days are the single largest directional class — this pair spent more of the window trending up than down, the carry-era signature — yet six days in ten are still range or quiet.
在 788 根派生日线上的机械日类型普查：RANGE 32.6%, QUIET OTHER 28.6%, TREND UP 20.6%, TREND DOWN 14.1%, EXPANSION OTHER 3.4%, SPIKE REVERSE 约 0.9%。趋势上涨日是最大的单一方向类别——这个货币对在样本窗口内上涨趋势日多于下跌日，这是套息时代的印记——但十天里仍有六天是区间或安静日。
First-hour continuation第一小时延续
DEAD已否定
n = 770 instances. Median forward effect vs the matched random null at the primary horizon (rest-of-day): discovery +0.018 ATR, holdout +0.008 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 770。在主要周期（rest-of-day）相对匹配随机基线的中位前向效应：发现期 +0.018 ATR，留出期 +0.008 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.2 per month (≈ 254 per year) · Dukascopy USD/JPY M15 · n = 770频率：每月约 21.2 次（每年约 254 次） · n = 770
Time & Volatility时间与波动
Hour of day (D1, descriptive). The widest M15 bars cluster at 14:00 UTC (US data hour, median bar range ≈ 13.7 pips), 13:00 and — uniquely among the FX volumes — 00:00 UTC, the Tokyo open (≈ 12.7 pips). This pair has two opens that matter.
Asia width (D2, descriptive). Narrow-Asia days (median width 0.21 prior-day ATR) still finish with a median full-day range of 0.68 ATR — 3.26× the Asia width — and close beyond the Asia range 77.5% of the time. Compare EUR/USD's 89.9%: Tokyo's range is a real range here, so a narrow one coils less violently.
Weekday (D3) and ATR regime (E1) are descriptive conditioning studies in the census files: no weekday carries a tradeable directional claim, and forward volatility rises with the trailing ATR regime while forward direction does not.
n = 56 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.763 ATR, holdout +0.058 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 56。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.763 ATR，留出期 +0.058 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.8 per month (≈ 21 per year) · Dukascopy USD/JPY M15→D1 derived · n = 56频率：每月约 1.8 次（每年约 21 次） · n = 56
Compression episodes (daily)压缩区间（日线）
DEAD已否定
n = 18 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.054 ATR, holdout -0.230 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 18。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.054 ATR，留出期 -0.230 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.6 per month (≈ 7 per year) · Dukascopy USD/JPY M15→D1 derived · n = 18频率：每月约 0.6 次（每年约 7 次） · n = 18
Bollinger-width squeezes (daily)布林带宽挤压（日线）
WEAK弱
n = 11 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.577 ATR, holdout -0.208 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 11。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.577 ATR，留出期 -0.208 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.4 per month (≈ 5 per year) · Dukascopy USD/JPY M15→D1 derived · n = 11频率：每月约 0.4 次（每年约 5 次） · n = 11
After >2×ATR days — continuation-signed>2×ATR 大日之后——延续方向
DEAD已否定
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.528 ATR, holdout -0.423 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.528 ATR，留出期 -0.423 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
After >2×ATR days — |return|>2×ATR 大日之后——波动幅度
WEAK弱
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.580 ATR, holdout +0.270 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.580 ATR，留出期 +0.270 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
Mean Reversion均值回归
Stretched H1 moves — no reliable snapback hereH1 过度延伸——在此无可靠回拉
WEAK弱
The extension snapback that CONFIRMED in EUR/USD (+0.209 ATR) is only WEAK in USD/JPY (+0.044 holdout). Stretched yen moves do not reliably snap back — the snapback is a EUR/USD behaviour, not an FX universal. n = 1005 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.024 ATR, holdout +0.044 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
在欧元/美元里确认过的延伸回拉（+0.209 ATR）在美元/日元里只有弱（留出期 +0.044）。日元的过度延伸走势并不可靠回拉——回拉是欧元/美元的行为，不是外汇的普遍规律。样本 n = 1005。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.024 ATR，留出期 +0.044 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 27.7 per month (≈ 332 per year) · Dukascopy USD/JPY M15→H1 derived · n = 1005频率：每月约 27.7 次（每年约 332 次） · n = 1005
London range reversion in US hours美国时段对伦敦区间的回归
DEAD已否定
n = 728 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.056 ATR, holdout -0.016 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 728。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.056 ATR，留出期 -0.016 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 20.0 per month (≈ 240 per year) · Dukascopy USD/JPY M15 · n = 728频率：每月约 20.0 次（每年约 240 次） · n = 728
Daily RSI(2) extremes, faded日线 RSI（2）极端——反打
WEAK弱
n = 59 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.088 ATR, holdout +0.224 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 59。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.088 ATR，留出期 +0.224 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.7 per month (≈ 21 per year) · Dukascopy USD/JPY M15→D1 derived · n = 59频率：每月约 1.7 次（每年约 21 次） · n = 59
Daily RSI(14) extremes, faded日线 RSI（14）极端——反打
DEAD已否定
n = 27 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.015 ATR, holdout +0.674 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 27。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.015 ATR，留出期 +0.674 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.8 per month (≈ 10 per year) · Dukascopy USD/JPY M15→D1 derived · n = 27频率：每月约 0.8 次（每年约 10 次） · n = 27
Macro Events — US releases & the Bank of Japan宏观事件——美国发布与日本央行
Yield conditioning (F7-JPY, descriptive) — this volume's cleanest measurement. On US event days when the US 10Y moved ≥ 5bp, USD/JPY moved WITH the yield 84.2% of the time (n=76), vs a 48.9% quiet-day baseline; with the broad dollar index, 87.0% (n=46). Yields up ≥ 5bp → the pair's median day was +72.8 pips; yields down ≥ 5bp → −106.3 pips. The yield → dollar → pair chain is not lore in this market; it is the dominant event-day fact. (Volume II measured the mirror from the other side: EUR/USD moved OPPOSITE the yield on 77.3% of such days.)
The Bank of Japan (JPY extras, measured 2026-10-16). Decision dates were sourced from the BoJ's official meeting schedule as reproduced by MNI + Kenanga, cross-checked (25 dates, 2023-10 → 2026-10; provenance in boj_meetings.json). F_daily_BoJ — WEAK (n=23 usable): decision days lean pair-DOWN over the following 5 days (discovery −0.286 ATR, holdout −0.902 ATR) — a yen-bid lean after BoJ days, with a small discovery half, so it is recorded as a flag, not a finding. The Tokyo session's median range on BoJ days is 0.718 ATR vs 0.400 ATR on other days (descriptive): decision days are ~1.8× wider in the home session.
Suspected interventions (descriptive proxy, pre-registered). A Tokyo-hours 60-minute shock of ≥ 3× the trailing daily ATR against the prior 5-day drift: 4,728 windows scanned, exactly 1 instance in three years. At the pre-registered threshold the class is nearly empty — intervention-scale Tokyo shocks are rarer than the folklore says, or smoother than a 60-minute window can see. No verdict by design.
The F6 warning. 'Hot' and 'cold' below use the crude proxy (released m/m change vs the prior month's, NOT a consensus surprise), with directions flipped for this pair as documented before measurement (hot print = strong dollar = pair up). Every F6 row is WEAK here — as in gold and EUR/USD, no hot/cold trade survives.
All 08:30 releases — the first spike is faded全部 08:30 发布——第一波被回吐
CONFIRMED — NEGATIVE已确认 — 负向
Pooled across CPI/PPI/NFP/Retail: the release spike's direction is given back over the next two hours (−0.650 ATR holdout, ≈ −4.6 pips). This now CONFIRMS in all three POSEIDON markets measured so far — gold, EUR/USD and USD/JPY. Directions here are signed for the pair: a strong-dollar spike up is faded down. n = 142 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.433 ATR, holdout -0.650 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 4.0 per month (≈ 47 per year) · Dukascopy USD/JPY M15 · n = 142频率：每月约 4.0 次（每年约 47 次） · n = 142
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=142.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=142。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2023-10-17T12:30:00+00:00', 'bar_index': 1114, 'dir': 1.0, 'primary_ret_ATR': 0.0, 'primary_ret_usd': 0.0, 'path_rmse_vs_median': 0.4779992989735174, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-09-11T12:30:00+00:00', 'bar_index': 73594, 'dir': -1.0, 'primary_ret_ATR': 0.2520858806955098, 'primary_ret_usd': 0.044999999999987494, 'path_rmse_vs_median': 0.6605424007162718, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-16T12:30:00+00:00', 'bar_index': 38458, 'dir': -1.0, 'primary_ret_ATR': -1.1463612299906079, 'primary_ret_usd': -0.21000000000000796, 'path_rmse_vs_median': 0.7408647903009951, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2023-10-17T12:30:00+00:00', 'bar_index': 1114, 'dir': 1.0, 'primary_ret_ATR': 0.0, 'primary_ret_usd': 0.0, 'path_rmse_vs_median': 0.4779992989735174, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-09-11T12:30:00+00:00', 'bar_index': 73594, 'dir': -1.0, 'primary_ret_ATR': 0.2520858806955098, 'primary_ret_usd': 0.044999999999987494, 'path_rmse_vs_median': 0.6605424007162718, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-16T12:30:00+00:00', 'bar_index': 38458, 'dir': -1.0, 'primary_ret_ATR': -1.1463612299906079, 'primary_ret_usd': -0.21000000000000796, 'path_rmse_vs_median': 0.7408647903009951, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
CPI releases — the first spikeCPI 发布——第一波
WEAK弱
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.132 ATR, holdout -0.703 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.132 ATR，留出期 -0.703 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15 · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
PPI releases — the first spikePPI 发布——第一波
DEAD已否定
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -1.439 ATR, holdout +0.564 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -1.439 ATR，留出期 +0.564 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15 · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
NFP releases — the first spike非农发布——第一波
DEAD已否定
n = 36 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.044 ATR, holdout -1.023 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 36。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.044 ATR，留出期 -1.023 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15 · n = 36频率：每月约 1.0 次（每年约 12 次） · n = 36
Retail Sales releases — the first spike零售销售发布——第一波
WEAK弱
n = 36 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.345 ATR, holdout -0.975 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 36。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.345 ATR，留出期 -0.975 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15 · n = 36频率：每月约 1.0 次（每年约 12 次） · n = 36
FOMC releases — the first spikeFOMC 发布——第一波
DEAD已否定
n = 24 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.114 ATR, holdout +1.639 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 24。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.114 ATR，留出期 +1.639 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy USD/JPY M15 · n = 24频率：每月约 0.7 次（每年约 8 次） · n = 24
CPI days — direction persistenceCPI 日——方向持续性
DEAD已否定
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.150 ATR, holdout -0.542 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.150 ATR，留出期 -0.542 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
PPI days — direction persistencePPI 日——方向持续性
WEAK弱
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.326 ATR, holdout +0.288 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.326 ATR，留出期 +0.288 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
NFP days — direction persistence非农日——方向持续性
DEAD已否定
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.154 ATR, holdout -0.630 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.154 ATR，留出期 -0.630 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
Retail Sales days — direction persistence零售销售日——方向持续性
DEAD已否定
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.237 ATR, holdout +0.339 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.237 ATR，留出期 +0.339 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy USD/JPY M15→D1 derived · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
FOMC days — direction persistenceFOMC 日——方向持续性
DEAD已否定
n = 23 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.297 ATR, holdout +1.494 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 23。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.297 ATR，留出期 +1.494 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy USD/JPY M15→D1 derived · n = 23频率：每月约 0.7 次（每年约 8 次） · n = 23
Hot prints (crude proxy, pair-flipped)热数据（粗略代理，按货币对翻转）
WEAK弱
n = 69 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.010 ATR, holdout +0.114 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 69。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.010 ATR，留出期 +0.114 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.0 per month (≈ 24 per year) · Dukascopy USD/JPY M15→D1 derived · n = 69频率：每月约 2.0 次（每年约 24 次） · n = 69
Cold prints (crude proxy, pair-flipped)冷数据（粗略代理，按货币对翻转）
WEAK弱
n = 66 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.017 ATR, holdout -0.739 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 66。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.017 ATR，留出期 -0.739 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.9 per month (≈ 23 per year) · Dukascopy USD/JPY M15→D1 derived · n = 66频率：每月约 1.9 次（每年约 23 次） · n = 66
Hot CPI only (crude proxy)仅热 CPI（粗略代理）
WEAK弱
n = 17 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.207 ATR, holdout -0.868 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 17。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.207 ATR，留出期 -0.868 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy USD/JPY M15→D1 derived · n = 17频率：每月约 0.5 次（每年约 6 次） · n = 17
Cold CPI only (crude proxy)仅冷 CPI（粗略代理）
WEAK弱
n = 15 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.132 ATR, holdout -0.889 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 15。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.132 ATR，留出期 -0.889 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy USD/JPY M15→D1 derived · n = 15频率：每月约 0.5 次（每年约 6 次） · n = 15
Cross-Asset States跨资产状态
After broad-dollar spike days广义美元冲高日之后
WEAK弱
n = 27 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.616 ATR, holdout +0.282 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 27。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.616 ATR，留出期 +0.282 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.8 per month (≈ 10 per year) · Dukascopy USD/JPY M15→D1 derived · n = 27频率：每月约 0.8 次（每年约 10 次） · n = 27
After DXY spike days (2y window)DXY 冲高日之后（2 年窗口）
DEAD已否定
n = 49 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.440 ATR, holdout -0.211 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 49。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.440 ATR，留出期 -0.211 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.3 per month (≈ 27 per year) · Dukascopy USD/JPY M15→D1 derived · n = 49频率：每月约 2.3 次（每年约 27 次） · n = 49
After yield-spike days收益率冲高日之后
DEAD已否定
n = 41 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.414 ATR, holdout +0.005 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 41。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.414 ATR，留出期 +0.005 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.2 per month (≈ 14 per year) · Dukascopy USD/JPY M15→D1 derived · n = 41频率：每月约 1.2 次（每年约 14 次） · n = 41
Quiet days安静日
DESCRIPTIVE描述性
n = 157 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.010 ATR, holdout +0.077 ATR. Verdict: DESCRIPTIVE under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 157。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.010 ATR，留出期 +0.077 ATR。按章程规则判定为：DESCRIPTIVE（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 4.5 per month (≈ 54 per year) · Dukascopy USD/JPY M15→D1 derived · n = 157频率：每月约 4.5 次（每年约 54 次） · n = 157
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-06-03T00:00:00+00:00', 'bar_index': 434, 'dir': 1.0, 'primary_ret_ATR': 0.57512078232749, 'primary_ret_usd': 0.9350000000000023, 'path_rmse_vs_median': 0.5065789381626667, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-05-20T00:00:00+00:00', 'bar_index': 164, 'dir': 1.0, 'primary_ret_ATR': 0.4820507969878172, 'primary_ret_usd': 0.6430000000000007, 'path_rmse_vs_median': 0.5997858451041901, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-18T00:00:00+00:00', 'bar_index': 682, 'dir': 1.0, 'primary_ret_ATR': 0.13010845477913907, 'primary_ret_usd': 0.13700000000000045, 'path_rmse_vs_median': 0.6942111879447442, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-06-03T00:00:00+00:00', 'bar_index': 434, 'dir': 1.0, 'primary_ret_ATR': 0.57512078232749, 'primary_ret_usd': 0.9350000000000023, 'path_rmse_vs_median': 0.5065789381626667, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-05-20T00:00:00+00:00', 'bar_index': 164, 'dir': 1.0, 'primary_ret_ATR': 0.4820507969878172, 'primary_ret_usd': 0.6430000000000007, 'path_rmse_vs_median': 0.5997858451041901, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-18T00:00:00+00:00', 'bar_index': 682, 'dir': 1.0, 'primary_ret_ATR': 0.13010845477913907, 'primary_ret_usd': 0.13700000000000045, 'path_rmse_vs_median': 0.6942111879447442, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Sessions in Their Own Right时段本身
Session anatomy, 788 trading days (median ranges, pips). The Overlap is widest (50), then London (46) — but Overnight (42), New York (42) and Asia/Tokyo (41) are all real sessions in this pair. Compare EUR/USD, where Asia carries just 18 of the Overlap's 32. For USD/JPY, Asia is home turf, exactly as Volume III's definitions warned.
Session
Median range (pips)
Overnight (23:00-02:00 UTC)
42
Asia (02:00-07:00 UTC)
41
London (07:00-12:00 UTC)
46
Overlap (12:00-15:00 UTC)
50
New York (15:00-20:00 UTC)
42
OffHours (20:00-23:00 UTC)
19
First-hour breaks (Sb). Accepted first-hour breaks are CONFIRMED negative in the sessions where they confirm (strongest −0.653 ATR holdout, ≈ −4.5 pips) and CONFIRMED positive in one (+0.147) — session identity decides the sign; the centre of gravity is fade. Rejections are mixed (one CONFIRMED +0.223, one CONFIRMED −0.114).
Prior-session extremes (Sd). The largest single holdout effect in the batch is a rejection: Asia-range rejections during a later session, CONFIRMED +0.656 ATR (≈ +5.5 pips, n=154). Overlap-range rejections split by the touching session (+0.377 in one, −0.466 in another) — which session does the touching decides the sign, the same lesson EUR/USD taught. Full per-session tables: census files S/<session>/.
n = 783 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery +0.052 ATR, holdout +0.256 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 783。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 +0.052 ATR，留出期 +0.256 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 259 per year) · Dukascopy USD/JPY M15 · n = 783频率：每月约 21.5 次（每年约 259 次） · n = 783
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=783.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=783。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-09-10T13:15:00+00:00', 'bar_index': 48541, 'dir': -1.0, 'primary_ret_ATR': -0.41582746049575975, 'primary_ret_usd': -0.05299999999999727, 'path_rmse_vs_median': 0.3944329460362579, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-29T13:15:00+00:00', 'bar_index': 20605, 'dir': 1.0, 'primary_ret_ATR': 0.47239231586812064, 'primary_ret_usd': 0.07699999999999818, 'path_rmse_vs_median': 0.4723541940180814, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-09-15T13:15:00+00:00', 'bar_index': 73789, 'dir': 1.0, 'primary_ret_ATR': 0.164108168099012, 'primary_ret_usd': 0.01999999999998181, 'path_rmse_vs_median': 0.5725119855311483, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-09-10T13:15:00+00:00', 'bar_index': 48541, 'dir': -1.0, 'primary_ret_ATR': -0.41582746049575975, 'primary_ret_usd': -0.05299999999999727, 'path_rmse_vs_median': 0.3944329460362579, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-29T13:15:00+00:00', 'bar_index': 20605, 'dir': 1.0, 'primary_ret_ATR': 0.47239231586812064, 'primary_ret_usd': 0.07699999999999818, 'path_rmse_vs_median': 0.4723541940180814, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-09-15T13:15:00+00:00', 'bar_index': 73789, 'dir': 1.0, 'primary_ret_ATR': 0.164108168099012, 'primary_ret_usd': 0.01999999999998181, 'path_rmse_vs_median': 0.5725119855311483, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Overlap → New York handoff — it reverses重叠→纽约交接——随后反转
CONFIRMED — NEGATIVE已确认 — 负向
The Overlap session's direction is reversed across the New York session (−0.411 ATR holdout, ≈ −4.2 pips, n=784). The strongest handoff effect in this volume: what the dollar did into the US afternoon, New York undoes. n = 784 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery -0.041 ATR, holdout -0.411 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 21.6 per month (≈ 259 per year) · Dukascopy USD/JPY M15 · n = 784频率：每月约 21.6 次（每年约 259 次） · n = 784
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=784.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=784。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-10-09T14:45:00+00:00', 'bar_index': 50563, 'dir': 1.0, 'primary_ret_ATR': 0.1280254534059117, 'primary_ret_usd': 0.025000000000005684, 'path_rmse_vs_median': 0.44048518828739414, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-06-14T14:45:00+00:00', 'bar_index': 17635, 'dir': 1.0, 'primary_ret_ATR': -0.3758508251928243, 'primary_ret_usd': -0.0730000000000075, 'path_rmse_vs_median': 0.49575860597954374, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-07-22T14:45:00+00:00', 'bar_index': 70051, 'dir': 1.0, 'primary_ret_ATR': 0.6185178612059334, 'primary_ret_usd': 0.03299999999998704, 'path_rmse_vs_median': 0.5012613883167996, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-10-09T14:45:00+00:00', 'bar_index': 50563, 'dir': 1.0, 'primary_ret_ATR': 0.1280254534059117, 'primary_ret_usd': 0.025000000000005684, 'path_rmse_vs_median': 0.44048518828739414, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-06-14T14:45:00+00:00', 'bar_index': 17635, 'dir': 1.0, 'primary_ret_ATR': -0.3758508251928243, 'primary_ret_usd': -0.0730000000000075, 'path_rmse_vs_median': 0.49575860597954374, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-07-22T14:45:00+00:00', 'bar_index': 70051, 'dir': 1.0, 'primary_ret_ATR': 0.6185178612059334, 'primary_ret_usd': 0.03299999999998704, 'path_rmse_vs_median': 0.5012613883167996, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
New York → Off-hours handoff纽约→盘后交接
CONFIRMED — NEGATIVE已确认 — 负向
n = 784 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery -0.090 ATR, holdout -0.123 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 784。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 -0.090 ATR，留出期 -0.123 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.6 per month (≈ 259 per year) · Dukascopy USD/JPY M15 · n = 784频率：每月约 21.6 次（每年约 259 次） · n = 784
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-12-06T19:45:00+00:00', 'bar_index': 29655, 'dir': 1.0, 'primary_ret_ATR': 0.12840958782317227, 'primary_ret_usd': 0.020000000000010232, 'path_rmse_vs_median': 0.5448988105031431, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-12-05T19:45:00+00:00', 'bar_index': 4503, 'dir': 1.0, 'primary_ret_ATR': -1.3878178515651045, 'primary_ret_usd': -0.18000000000000682, 'path_rmse_vs_median': 0.5669788670485798, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-22T19:45:00+00:00', 'bar_index': 67959, 'dir': 1.0, 'primary_ret_ATR': -0.2601075868655195, 'primary_ret_usd': -0.02600000000001046, 'path_rmse_vs_median': 0.5674765431837885, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-12-06T19:45:00+00:00', 'bar_index': 29655, 'dir': 1.0, 'primary_ret_ATR': 0.12840958782317227, 'primary_ret_usd': 0.020000000000010232, 'path_rmse_vs_median': 0.5448988105031431, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-12-05T19:45:00+00:00', 'bar_index': 4503, 'dir': 1.0, 'primary_ret_ATR': -1.3878178515651045, 'primary_ret_usd': -0.18000000000000682, 'path_rmse_vs_median': 0.5669788670485798, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-06-22T19:45:00+00:00', 'bar_index': 67959, 'dir': 1.0, 'primary_ret_ATR': -0.2601075868655195, 'primary_ret_usd': -0.02600000000001046, 'path_rmse_vs_median': 0.5674765431837885, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
London's first 30 minutes伦敦最初 30 分钟
DEAD已否定
n = 782 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery +0.001 ATR, holdout -0.079 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 782。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 +0.001 ATR，留出期 -0.079 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 258 per year) · Dukascopy USD/JPY M15 · n = 782频率：每月约 21.5 次（每年约 258 次） · n = 782
Asia midpoint tests亚洲中点测试
DEAD已否定
n = 780 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery +0.143 ATR, holdout -0.463 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 780。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 +0.143 ATR，留出期 -0.463 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 258 per year) · Dukascopy USD/JPY M15 · n = 780频率：每月约 21.5 次（每年约 258 次） · n = 780
Overnight first hour隔夜第一小时
DEAD已否定
n = 782 instances. Median forward effect vs the matched random null at the primary horizon (close): discovery +0.035 ATR, holdout -0.112 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 782。在主要周期（close）相对匹配随机基线的中位前向效应：发现期 +0.035 ATR，留出期 -0.112 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 258 per year) · Dukascopy USD/JPY M15 · n = 782频率：每月约 21.5 次（每年约 258 次） · n = 782
What USD/JPY Teaches Gold & EUR/USD美元/日元对黄金与欧元/美元的启示
The haven filter is dead too. Gold's confirmed behaviours were re-measured split by USD/JPY's state at the event (XJ1). Gold's accepted breaks (n=323): the 2h continuation is present in BOTH yen states in the holdout — +0.40 ATR when the yen was bid, +0.42 when it was offered. The split separates nothing. Together with Volume II's dead EUR/USD-state filter, conditioning gold on the concurrent FX majors is now closed negatively, twice measured, and the gold chapters say so (logged write-back, 2026-10-16).
The haven mirror (XJ2). Same-session gold vs USD/JPY returns correlate NEGATIVELY in every session — Overlap −0.49, New York −0.48, London −0.30 — the exact mirror of gold × EUR/USD (+0.50 / +0.48). One dollar impulse, opposite signs. No lead-lag pair reaches |r| ≥ 0.10: neither market warns the other.
Event twins (XJ4). On 79 matched 08:30 releases, gold and USD/JPY first spikes are dollar-consistent (opposite raw signs) 81% of the time, and in both markets the spike is then faded over two hours.
The three-market A1 table. Same definition, three characters: gold's accepted prior-day breaks continue, EUR/USD's reverse (−0.36 ATR), USD/JPY's continue (+0.11). Any desk rule that assumes 'FX majors break out like gold' — or the reverse — is wrong in at least one market it will be applied to.
Cost audit. Measured on this corpus: median M15 ATR 10.1 pips, median daily range 111 pips (788 days). At a typical retail spread of 0.7–1.2 pips plus 0.3 slippage, a round trip costs ≈ 1.2 pips = 0.12× the median M15 ATR — cheaper per unit of movement than EUR/USD (0.20×), dearer than gold (0.07×). In R terms: an 8-pip stop pays 0.150R per trade, a 12-pip stop 0.100R, a 20-pip stop 0.060R. Consequence: this volume's CONFIRMED effects below ~1 pip (the candle family) are measurements, not trades. The effects that clear friction with room are the fades and the daily streak: the 08:30 spike (−4.6 pips), the Overlap→NY handoff reversal (−4.2), Asia-range rejections in their session (+5.5), one session's first-hour break fade (−4.5), and the 3-day streak fade (−27.9 pips at 5d).
Methods. Definitions were fixed in writing before measurement (Volume III DEFINITIONS). Deviations from the FX template, all documented in advance: round levels at 0.50 steps; F6 directions flipped (a hot US print is a strong dollar, and this pair rises on a strong dollar); F7 measured as WITH-yield co-movement. Sessions are fixed UTC windows with the 22:00 UTC roll — post-midnight bars belong to the prior evening's trading day, so no behaviour is grouped by calendar date against a same-date range (the look-ahead error that invalidated earlier desk backtests is excluded by construction). BoJ decision dates come from the BoJ's official schedule as reproduced by MNI + Kenanga, cross-checked, with provenance recorded in the data file.
Limits. (1) Three years of daily data, much of it an unusual BoJ-normalisation / carry regime — daily verdicts carry less weight than gold's ten-year ones. (2) Japanese domestic releases (CPI, Tankan, trade) are NOT in the shared calendar — the measured event life is US releases + BoJ decisions. (3) Interventions are covered only by the pre-registered mechanical proxy, which found 1 instance; real MoF action may be smoother than the proxy's window. (4) With 144 behaviours, a few CONFIRMED labels are expected false positives; instance counts and effect sizes are the evidence. (5) Census numbers are gross of costs and are NOT approved strategies — anything trade-shaped must still pass the desk's readiness bar (mechanical definition, costed backtest, ≥50 trades, PF ≥ 1.3 out-of-sample, ≥2 market periods).
9CONFIRMED — survived unseen data已确认——通过未见数据检验13WEAK — a flag, not a finding弱——只是标记，不是结论18DEAD — killed by the holdout已死亡——被留出数据杀掉10DESCRIPTIVE — measured, no direction claim描述性——只测量，不下方向结论
How wide each session runs每个时段跑多宽Median block range in points · 781 cash days. Trend days lean UP nearly 2:1 — the index’s structural drift.区块区间中位数（点）· 781 个现金交易日。趋势日向上倾斜近 2:1——指数的结构性漂移。
After-hours盘后
58 pts
Premarket盘前
124 pts
Overnight隔夜
223 pts
Cash session现金时段
254 pts
The friction tax — cost of one round trip摩擦税——一次往返的成本Round-trip cost as a fraction of one median M15 candle (ATR). Smaller = cheaper to trade.往返成本占一根典型 M15 K 线（ATR）的比例。越小 = 交易成本越低。
Gold黄金
0.07×
NAS100纳斯达克100
0.096×
USD/JPY美元/日元
0.12×
EUR/USD欧元/美元
0.20×
One day on the clock时钟上的一天Blocks of the ET day. The overnight block is nearly as wide as the cash session itself.ET 一天的区块。隔夜区块几乎和现金时段本身一样宽。
Overnight隔夜
223 pts
20:00–04:00 ET
Premarket盘前
124 pts
04:00–09:30
Cash session现金时段
254 pts
09:30–16:00
After-hours盘后
58 pts
16:00–20:00
Laws that survived all four markets在四个市场都活下来的规律
📰
The first spike lies第一波冲高会说谎
On 08:30 ET data days the first spike is faded in gold, EUR/USD and USD/JPY. Never chase it.在 08:30 ET 数据日，第一波冲高在黄金、欧元/美元和美元/日元中都会被反打。永远不要追它。
📉
Streaks snap back连涨连跌会弹回
Three same-direction days lean the other way over the next five — in all four markets.连续三天同方向，接下来五天会倾向反方向——四个市场全部如此。
🧭
Breaks continue — except EUR/USD突破会延续——欧元/美元除外
An accepted break of yesterday’s range keeps going in gold, USD/JPY and NAS100. In EUR/USD it reverses.被接受的昨日区间突破在黄金、美元/日元和纳斯达克100中会继续。在欧元/美元中它会反转。
⏱
Edges make the extremes边缘产生极值
Session highs and lows form in the first or last 30 minutes — everywhere, in every market.时段高低点在最初或最后 30 分钟形成——每个市场、每个地方都如此。
The headline behaviours — look first, read later头条行为——先看图，后读字
The break that sticks守得住的突破
Cash-session breaks of the prior day’s high/low that close beyond keep travelling. Of the four markets, EUR/USD alone fades this.现金时段收盘突破前日高低点后会继续。四个市场中，只有欧元/美元反打这个。
Breaks of the prior WEEK’s high/low, accepted, fade violently over the following days — the largest fade in this volume, and the mirror of its daily behaviour.被接受的前周高低点突破，会在接下来几天剧烈反打——本卷最大的反打效应，也是它日线行为的镜像。
Three same-direction cash days lean the other way for five — the largest streak fade of all four POSEIDON markets.连续三天同方向的现金日，接下来五天倾向反方向——四个 POSEIDON 市场中最大的连涨连跌反打。
Follow the overnight gap? Dead. Fade it? Dead. Holdout effects −0.003 / +0.003 ATR — as close to exactly zero as this census produces. The gap fills 57.6% of the time, but its direction tells you nothing.跟随隔夜跳空？死亡。反打它？死亡。留出效应 −0.003 / +0.003 ATR——这是本普查能产生的最接近零的结果。跳空有 57.6% 的时间被填补，但它的方向什么都告诉不了你。
After an inside bar the next 2 hours move MORE in absolute terms. A volatility fact, not a direction — absent in all three FX volumes.内包线之后，接下来 2 小时的绝对波动更大。这是波动率事实，不是方向——在三个外汇卷中都不存在。
The full chapters follow below — every number above is drawn from them, unchanged.完整章节在下方——上面的每一个数字都来自它们，未作任何改动。
Start Here从这里开始
POSEIDON — Volume IV: NAS100. The same observation-first census as Volumes I–III, now on an equity index — with the deviations the instrument forces, all fixed in writing before measurement. The corpus is the Dukascopy NAS100 CFD (USATECHIDXUSD), M15 aggregated from M1 BID — a CFD, not the cash index and not NQ futures, and it is named as such everywhere. The session of record is the US cash session, 09:30–16:00 America/New_York with real DST rules; the trading day is the ET calendar date. 781 valid cash days (ET weekday, cash range > 0, ≥ 20 cash bars — the feed writes flat repeated quotes through weekends, and 317 such phantom days are excluded and reported, not patched). 50 behaviours: 9 CONFIRMED, 13 WEAK, 18 DEAD, 10 DESCRIPTIVE. All effects gross of costs; the cost and sizing audit is its own chapter and its verdict is blunt.
The one-sentence character of NAS100: an index that trends by day and fades its own streaks. Prior-day and premarket breaks continue during the cash session; prior-week breaks collapse; three-day streaks snap back; and the famous overnight gap predicts nothing at all.
Prior cash-day breaks, accepted — they continue前一现金日突破被接受——随后延续
CONFIRMED已确认
When the cash session first touches the prior cash day's high or low and closes beyond it, the next two hours keep travelling (+0.111 ATR holdout, ≈ +8.7 points). Gold's and USD/JPY's sign — of the four POSEIDON markets, EUR/USD alone fades. n = 557 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.052 ATR, holdout +0.111 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 15.3 per month (≈ 184 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 557频率：每月约 15.3 次（每年约 184 次） · n = 557
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=557.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=557。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2023-10-31T17:45:00+00:00', 'bar_index': 2951, 'dir': 1.0, 'primary_ret_ATR': 0.31481729401061537, 'primary_ret_usd': 11.420000000000073, 'path_rmse_vs_median': 0.46135420496521196, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-22T17:00:00+00:00', 'bar_index': 28388, 'dir': 1.0, 'primary_ret_ATR': 0.5712587189431921, 'primary_ret_usd': 27.8650000000016, 'path_rmse_vs_median': 0.49529644992155747, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-21T17:15:00+00:00', 'bar_index': 92517, 'dir': 1.0, 'primary_ret_ATR': 0.17813428473534756, 'primary_ret_usd': 13.282999999999447, 'path_rmse_vs_median': 0.5026740245652533, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2023-10-31T17:45:00+00:00', 'bar_index': 2951, 'dir': 1.0, 'primary_ret_ATR': 0.31481729401061537, 'primary_ret_usd': 11.420000000000073, 'path_rmse_vs_median': 0.46135420496521196, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-07-22T17:00:00+00:00', 'bar_index': 28388, 'dir': 1.0, 'primary_ret_ATR': 0.5712587189431921, 'primary_ret_usd': 27.8650000000016, 'path_rmse_vs_median': 0.49529644992155747, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-05-21T17:15:00+00:00', 'bar_index': 92517, 'dir': 1.0, 'primary_ret_ATR': 0.17813428473534756, 'primary_ret_usd': 13.282999999999447, 'path_rmse_vs_median': 0.5026740245652533, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Prior cash-day breaks, 'rejected' — the level fails前一现金日突破被'拒绝'——价位失守
CONFIRMED — NEGATIVE已确认 — 负向
A wick beyond the prior cash day's edge that closes back inside does NOT hold the level in the index: forward movement leans INTO the break (−0.185 ATR holdout, signed in the rejection direction). In every FX volume a rejection resolved in the rejection direction; here the first close back inside is a pause, not a rejection. n = 316 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.007 ATR, holdout -0.185 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
刺破前一现金日边缘但收盘回到内部的形态，在指数里并不守住价位：前向运动倾向于顺突破方向（按拒绝方向签署，留出期 −0.185 ATR）。在每个外汇卷里拒绝都按拒绝方向化解；在这里，第一次收回内部是暂停，不是拒绝。样本 n = 316。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.007 ATR，留出期 -0.185 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 8.8 per month (≈ 105 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 316频率：每月约 8.8 次（每年约 105 次） · n = 316
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2024-09-23T18:00:00+00:00', 'bar_index': 34440, 'dir': -1.0, 'primary_ret_ATR': -0.08904758460482932, 'primary_ret_usd': -3.0339999999996508, 'path_rmse_vs_median': 0.2980043772441681, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-07-02T16:15:00+00:00', 'bar_index': 61505, 'dir': -1.0, 'primary_ret_ATR': 0.20858181119548524, 'primary_ret_usd': 6.744999999998981, 'path_rmse_vs_median': 0.33099088317517794, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-10-18T18:45:00+00:00', 'bar_index': 1707, 'dir': 1.0, 'primary_ret_ATR': 0.033821881341574316, 'primary_ret_usd': 1.4200000000000728, 'path_rmse_vs_median': 0.5817686579163358, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2024-09-23T18:00:00+00:00', 'bar_index': 34440, 'dir': -1.0, 'primary_ret_ATR': -0.08904758460482932, 'primary_ret_usd': -3.0339999999996508, 'path_rmse_vs_median': 0.2980043772441681, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-07-02T16:15:00+00:00', 'bar_index': 61505, 'dir': -1.0, 'primary_ret_ATR': 0.20858181119548524, 'primary_ret_usd': 6.744999999998981, 'path_rmse_vs_median': 0.33099088317517794, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2023-10-18T18:45:00+00:00', 'bar_index': 1707, 'dir': 1.0, 'primary_ret_ATR': 0.033821881341574316, 'primary_ret_usd': 1.4200000000000728, 'path_rmse_vs_median': 0.5817686579163358, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Premarket range breaks, accepted — they continue盘前区间突破被接受——随后延续
CONFIRMED已确认
The premarket range is a real range (median 124 points) and its accepted break at the cash open continues over the next two hours (+0.222 ATR holdout, ≈ +10.6 points). The strongest intraday continuation in this volume. n = 606 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.237 ATR, holdout +0.222 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 16.7 per month (≈ 200 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 606频率：每月约 16.7 次（每年约 200 次） · n = 606
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=606.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=606。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2023-10-27T17:00:00+00:00', 'bar_index': 2564, 'dir': -1.0, 'primary_ret_ATR': 0.4964322602570807, 'primary_ret_usd': 19.44999999999891, 'path_rmse_vs_median': 0.4300870053566662, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-11T16:30:00+00:00', 'bar_index': 53634, 'dir': 1.0, 'primary_ret_ATR': 0.2826208193439403, 'primary_ret_usd': 38.544999999998254, 'path_rmse_vs_median': 0.4510787672889875, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-04-24T15:45:00+00:00', 'bar_index': 89919, 'dir': 1.0, 'primary_ret_ATR': 0.6153058640662653, 'primary_ret_usd': 34.546000000002095, 'path_rmse_vs_median': 0.4619401069992464, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2023-10-27T17:00:00+00:00', 'bar_index': 2564, 'dir': -1.0, 'primary_ret_ATR': 0.4964322602570807, 'primary_ret_usd': 19.44999999999891, 'path_rmse_vs_median': 0.4300870053566662, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-04-11T16:30:00+00:00', 'bar_index': 53634, 'dir': 1.0, 'primary_ret_ATR': 0.2826208193439403, 'primary_ret_usd': 38.544999999998254, 'path_rmse_vs_median': 0.4510787672889875, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-04-24T15:45:00+00:00', 'bar_index': 89919, 'dir': 1.0, 'primary_ret_ATR': 0.6153058640662653, 'primary_ret_usd': 34.546000000002095, 'path_rmse_vs_median': 0.4619401069992464, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Premarket range breaks, rejected盘前区间突破被拒绝
DEAD已否定
n = 546 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.118 ATR, holdout -0.237 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 546。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.118 ATR，留出期 -0.237 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 15.2 per month (≈ 182 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 546频率：每月约 15.2 次（每年约 182 次） · n = 546
Prior-week breaks, accepted — they collapse前周突破被接受——随后崩落
CONFIRMED — NEGATIVE已确认 — 负向
Breaks of the prior week's cash high/low that are accepted during the week fade violently over the following days (−0.671 ATR holdout, ≈ −215 points). The weekly level is where this instrument's breakouts go to die — the mirror of its daily behaviour. n = 99 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.115 ATR, holdout -0.671 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 2.9 per month (≈ 34 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 99频率：每月约 2.9 次（每年约 34 次） · n = 99
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=99.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=99。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-08-24T13:30:00+00:00', 'bar_index': 745, 'dir': -1.0, 'primary_ret_ATR': -0.9298820508939948, 'primary_ret_usd': -427.57800000000134, 'path_rmse_vs_median': 0.8240668216438491, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-20T14:30:00+00:00', 'bar_index': 334, 'dir': 1.0, 'primary_ret_ATR': -1.1512573448435854, 'primary_ret_usd': -405.72299999999814, 'path_rmse_vs_median': 0.8320080831629612, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-01-06T14:30:00+00:00', 'bar_index': 582, 'dir': 1.0, 'primary_ret_ATR': 0.35008585859710456, 'primary_ret_usd': 109.875, 'path_rmse_vs_median': 0.8378358586725903, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-08-24T13:30:00+00:00', 'bar_index': 745, 'dir': -1.0, 'primary_ret_ATR': -0.9298820508939948, 'primary_ret_usd': -427.57800000000134, 'path_rmse_vs_median': 0.8240668216438491, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-20T14:30:00+00:00', 'bar_index': 334, 'dir': 1.0, 'primary_ret_ATR': -1.1512573448435854, 'primary_ret_usd': -405.72299999999814, 'path_rmse_vs_median': 0.8320080831629612, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2026-01-06T14:30:00+00:00', 'bar_index': 582, 'dir': 1.0, 'primary_ret_ATR': 0.35008585859710456, 'primary_ret_usd': 109.875, 'path_rmse_vs_median': 0.8378358586725903, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Prior-week breaks, rejected前周突破被拒绝
DEAD已否定
n = 65 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.277 ATR, holdout -0.190 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 65。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.277 ATR，留出期 -0.190 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.9 per month (≈ 23 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 65频率：每月约 1.9 次（每年约 23 次） · n = 65
Candlesticks in ContextK线形态
Reading the candle chapter honestly. Pins are DEAD (discovery effect exactly 0.0). Engulfing is CONFIRMED but tiny and negative (≈ −2.6 points) — a measurement, not a trade. The one real candle fact is volatility, not direction: inside bars mark an expanding tape (N-B3, measured on |return| by the program's convention, since an inside bar's direction is defined by its own future break).
诚实地读 K 线章节。针形线已被否定（发现期效应恰好为 0.0）。吞没被确认但微小且为负（约 −2.6 点）——是测量结果，不是交易机会。唯一真实的 K 线事实是波动而非方向：内包线标记着扩张的行情（N-B3，按本计划惯例以 |收益| 测量，因为内包线的方向由其自身未来的突破定义）。
Pin bars on cash bars现金 K 线针形线
DEAD已否定
n = 2753 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.000 ATR, holdout -0.019 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 2753。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.000 ATR，留出期 -0.019 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 75.8 per month (≈ 909 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 2753频率：每月约 75.8 次（每年约 909 次） · n = 2753
Pin bars at a level价位处的针形线
DEAD已否定
n = 2727 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.000 ATR, holdout -0.023 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 2727。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.000 ATR，留出期 -0.023 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 75.0 per month (≈ 900 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 2727频率：每月约 75.0 次（每年约 900 次） · n = 2727
Pin bars mid-range区间中部的针形线
WEAK弱
n = 26 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +1.281 ATR, holdout +0.073 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 26。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +1.281 ATR，留出期 +0.073 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.1 per month (≈ 13 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 26频率：每月约 1.1 次（每年约 13 次） · n = 26
Engulfing candles on cash bars现金 K 线吞没形态
CONFIRMED — NEGATIVE已确认 — 负向
n = 2365 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.027 ATR, holdout -0.054 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 2365。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.027 ATR，留出期 -0.054 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 65.1 per month (≈ 781 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 2365频率：每月约 65.1 次（每年约 781 次） · n = 2365
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-07-31T19:45:00+00:00', 'bar_index': 99343, 'dir': -1.0, 'primary_ret_ATR': 0.039213011085751, 'primary_ret_usd': 3.155000000002474, 'path_rmse_vs_median': 0.1700292949409156, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-19T20:45:00+00:00', 'bar_index': 77843, 'dir': 1.0, 'primary_ret_ATR': 0.010540032283534914, 'primary_ret_usd': 0.3470000000015716, 'path_rmse_vs_median': 0.17251053967709, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-08-09T19:45:00+00:00', 'bar_index': 30127, 'dir': 1.0, 'primary_ret_ATR': 0.05199987567326184, 'primary_ret_usd': 2.643000000000029, 'path_rmse_vs_median': 0.1954732865487271, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-07-31T19:45:00+00:00', 'bar_index': 99343, 'dir': -1.0, 'primary_ret_ATR': 0.039213011085751, 'primary_ret_usd': 3.155000000002474, 'path_rmse_vs_median': 0.1700292949409156, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-19T20:45:00+00:00', 'bar_index': 77843, 'dir': 1.0, 'primary_ret_ATR': 0.010540032283534914, 'primary_ret_usd': 0.3470000000015716, 'path_rmse_vs_median': 0.17251053967709, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-08-09T19:45:00+00:00', 'bar_index': 30127, 'dir': 1.0, 'primary_ret_ATR': 0.05199987567326184, 'primary_ret_usd': 2.643000000000029, 'path_rmse_vs_median': 0.1954732865487271, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Engulfing candles at a level价位处的吞没形态
CONFIRMED — NEGATIVE已确认 — 负向
n = 2347 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.019 ATR, holdout -0.058 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 2347。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.019 ATR，留出期 -0.058 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 64.6 per month (≈ 775 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 2347频率：每月约 64.6 次（每年约 775 次） · n = 2347
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-07-31T19:45:00+00:00', 'bar_index': 99343, 'dir': -1.0, 'primary_ret_ATR': 0.039213011085751, 'primary_ret_usd': 3.155000000002474, 'path_rmse_vs_median': 0.1700464307409107, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-19T20:45:00+00:00', 'bar_index': 77843, 'dir': 1.0, 'primary_ret_ATR': 0.010540032283534914, 'primary_ret_usd': 0.3470000000015716, 'path_rmse_vs_median': 0.17341476041059906, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-08-09T19:45:00+00:00', 'bar_index': 30127, 'dir': 1.0, 'primary_ret_ATR': 0.05199987567326184, 'primary_ret_usd': 2.643000000000029, 'path_rmse_vs_median': 0.1960037815452173, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-07-31T19:45:00+00:00', 'bar_index': 99343, 'dir': -1.0, 'primary_ret_ATR': 0.039213011085751, 'primary_ret_usd': 3.155000000002474, 'path_rmse_vs_median': 0.1700464307409107, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-12-19T20:45:00+00:00', 'bar_index': 77843, 'dir': 1.0, 'primary_ret_ATR': 0.010540032283534914, 'primary_ret_usd': 0.3470000000015716, 'path_rmse_vs_median': 0.17341476041059906, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-08-09T19:45:00+00:00', 'bar_index': 30127, 'dir': 1.0, 'primary_ret_ATR': 0.05199987567326184, 'primary_ret_usd': 2.643000000000029, 'path_rmse_vs_median': 0.1960037815452173, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Engulfing candles mid-range区间中部的吞没形态
DEAD已否定
n = 18 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.367 ATR, holdout +0.123 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 18。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.367 ATR，留出期 +0.123 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.8 per month (≈ 10 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 18频率：每月约 0.8 次（每年约 10 次） · n = 18
Inside bars — volatility expands after them内包线——之后波动扩张
CONFIRMED已确认
Measured on |return| by the program's convention: an inside bar's direction is defined by its subsequent break, so a directional measure would be look-ahead by construction. What is real: after an inside bar the next two hours move ~0.36 ATR (holdout) MORE in absolute terms than after a random cash bar. Verified not to be an hour-of-day mix artifact — the gap holds inside every cash hour. The likely mechanism (interpretation): a large mother bar makes 'inside' more likely, so inside bars mark an already-active tape. In the FX volumes this |return| effect was ≈ 0 — a genuine cross-market difference. Direction is NOT predicted. n = 2403 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.532 ATR, holdout +0.359 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
按本计划的惯例以 |收益| 测量：内包线的方向由其随后的突破定义，因此方向性测量在构造上就是前视。真实的是：内包线之后两小时的绝对运动比随机现金 K 线之后多约 0.36 ATR（留出期）。已验证这不是日内小时混合的假象——差距在每个现金时段内都成立。可能的机制（属诠释）：大母线使'内包'更可能发生，所以内包线标记的是已经活跃的行情。在外汇各卷里这个 |收益| 效应约等于 0——这是真实的跨市场差异。方向并未被预测。样本 n = 2403。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.532 ATR，留出期 +0.359 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 66.1 per month (≈ 793 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 2403频率：每月约 66.1 次（每年约 793 次） · n = 2403
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-09-26T19:30:00+00:00', 'bar_index': 69774, 'dir': 1.0, 'primary_ret_ATR': 0.7501036803676847, 'primary_ret_usd': 28.574000000000524, 'path_rmse_vs_median': 0.2224692574370422, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-03T20:15:00+00:00', 'bar_index': 44241, 'dir': -1.0, 'primary_ret_ATR': 0.8705806857628078, 'primary_ret_usd': 32.71099999999933, 'path_rmse_vs_median': 0.22894758624274636, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-05-03T17:45:00+00:00', 'bar_index': 20711, 'dir': 1.0, 'primary_ret_ATR': 0.4403348550276279, 'primary_ret_usd': 16.40099999999802, 'path_rmse_vs_median': 0.22942669323936, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-09-26T19:30:00+00:00', 'bar_index': 69774, 'dir': 1.0, 'primary_ret_ATR': 0.7501036803676847, 'primary_ret_usd': 28.574000000000524, 'path_rmse_vs_median': 0.2224692574370422, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-03T20:15:00+00:00', 'bar_index': 44241, 'dir': -1.0, 'primary_ret_ATR': 0.8705806857628078, 'primary_ret_usd': 32.71099999999933, 'path_rmse_vs_median': 0.22894758624274636, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-05-03T17:45:00+00:00', 'bar_index': 20711, 'dir': 1.0, 'primary_ret_ATR': 0.4403348550276279, 'primary_ret_usd': 16.40099999999802, 'path_rmse_vs_median': 0.22942669323936, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Inside bars at a level — |return|价位处内包线——|收益|
CONFIRMED已确认
n = 2365 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.548 ATR, holdout +0.384 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 2365。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.548 ATR，留出期 +0.384 ATR。按章程规则判定为：CONFIRMED（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 65.1 per month (≈ 781 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 2365频率：每月约 65.1 次（每年约 781 次） · n = 2365
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2025-09-26T19:30:00+00:00', 'bar_index': 69774, 'dir': 1.0, 'primary_ret_ATR': 0.7501036803676847, 'primary_ret_usd': 28.574000000000524, 'path_rmse_vs_median': 0.22411898606330444, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-03T20:15:00+00:00', 'bar_index': 44241, 'dir': -1.0, 'primary_ret_ATR': 0.8705806857628078, 'primary_ret_usd': 32.71099999999933, 'path_rmse_vs_median': 0.2275749138847062, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-05-03T17:45:00+00:00', 'bar_index': 20711, 'dir': 1.0, 'primary_ret_ATR': 0.4403348550276279, 'primary_ret_usd': 16.40099999999802, 'path_rmse_vs_median': 0.2281909946061321, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2025-09-26T19:30:00+00:00', 'bar_index': 69774, 'dir': 1.0, 'primary_ret_ATR': 0.7501036803676847, 'primary_ret_usd': 28.574000000000524, 'path_rmse_vs_median': 0.22411898606330444, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-01-03T20:15:00+00:00', 'bar_index': 44241, 'dir': -1.0, 'primary_ret_ATR': 0.8705806857628078, 'primary_ret_usd': 32.71099999999933, 'path_rmse_vs_median': 0.2275749138847062, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-05-03T17:45:00+00:00', 'bar_index': 20711, 'dir': 1.0, 'primary_ret_ATR': 0.4403348550276279, 'primary_ret_usd': 16.40099999999802, 'path_rmse_vs_median': 0.2281909946061321, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
Inside bars mid-range — |return|区间中部内包线——|收益|
WEAK弱
n = 38 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.548 ATR, holdout +0.028 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 38。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.548 ATR，留出期 +0.028 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.5 per month (≈ 18 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 38频率：每月约 1.5 次（每年约 18 次） · n = 38
Trend & Momentum趋势与动量
3-day close streaks — they fade, hardest of all3 日收盘连串——回吐最狠
CONFIRMED — NEGATIVE已确认 — 负向
Three same-direction cash days are followed by five days leaning the other way (−0.411 ATR holdout, ≈ −130 points). All four POSEIDON markets now agree that daily streaks fade; the index's version is the largest. n = 92 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.374 ATR, holdout -0.411 ATR. Verdict: CONFIRMED under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 2.7 per month (≈ 32 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 92频率：每月约 2.7 次（每年约 32 次） · n = 92
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=92.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=92。
Archetype instances (illustrations, not evidence)原型实例（仅作图示，非证据）
{'timestamp_utc': '2026-08-18T13:30:00+00:00', 'bar_index': 741, 'dir': -1.0, 'primary_ret_ATR': 0.5886583573131435, 'primary_ret_usd': 293.0330000000031, 'path_rmse_vs_median': 0.6884144727753209, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-08-08T13:30:00+00:00', 'bar_index': 477, 'dir': 1.0, 'primary_ret_ATR': 0.39686472779578963, 'primary_ret_usd': 106.22200000000157, 'path_rmse_vs_median': 0.7171363387775633, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-09-25T13:30:00+00:00', 'bar_index': 254, 'dir': 1.0, 'primary_ret_ATR': -0.5264908397149364, 'primary_ret_usd': -167.9559999999983, 'path_rmse_vs_median': 0.8235389736391996, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
{'timestamp_utc': '2026-08-18T13:30:00+00:00', 'bar_index': 741, 'dir': -1.0, 'primary_ret_ATR': 0.5886583573131435, 'primary_ret_usd': 293.0330000000031, 'path_rmse_vs_median': 0.6884144727753209, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2025-08-08T13:30:00+00:00', 'bar_index': 477, 'dir': 1.0, 'primary_ret_ATR': 0.39686472779578963, 'primary_ret_usd': 106.22200000000157, 'path_rmse_vs_median': 0.7171363387775633, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'} · {'timestamp_utc': '2024-09-25T13:30:00+00:00', 'bar_index': 254, 'dir': 1.0, 'primary_ret_ATR': -0.5264908397149364, 'primary_ret_usd': -167.9559999999983, 'path_rmse_vs_median': 0.8235389736391996, 'note': 'archetype = instance whose stacked path is closest to the median path (illustration, not evidence)'}
4-day close streaks4 日收盘连串
WEAK弱
n = 48 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.497 ATR, holdout -0.479 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 48。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.497 ATR，留出期 -0.479 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.4 per month (≈ 17 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 48频率：每月约 1.4 次（每年约 17 次） · n = 48
5-day close streaks5 日收盘连串
WEAK弱
n = 33 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.364 ATR, holdout -1.303 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 33。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.364 ATR，留出期 -1.303 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 11 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 33频率：每月约 1.0 次（每年约 11 次） · n = 33
After a +1% cash day+1% 现金大涨日之后
DEAD已否定
n = 144 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.076 ATR, holdout +0.051 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 144。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.076 ATR，留出期 +0.051 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 4.2 per month (≈ 50 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 144频率：每月约 4.2 次（每年约 50 次） · n = 144
After a −1% cash day−1% 现金大跌日之后
WEAK弱
n = 113 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.330 ATR, holdout -0.155 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 113。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.330 ATR，留出期 -0.155 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 3.4 per month (≈ 40 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 113频率：每月约 3.4 次（每年约 40 次） · n = 113
After a +2% cash day+2% 现金大涨日之后
DEAD已否定
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.470 ATR, holdout -0.567 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.470 ATR，留出期 -0.567 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
After a −2% cash day−2% 现金大跌日之后
DEAD已否定
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.512 ATR, holdout -0.994 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.512 ATR，留出期 -0.994 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.2 per month (≈ 15 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 34频率：每月约 1.2 次（每年约 15 次） · n = 34
Opening Gaps & the First Hour开盘跳空与第一小时
The gap census (descriptive). Overnight gap = cash open minus prior cash close, in prior-day cash ATR units: median |gap| 0.26 ATR over 766 valid days, and the gap FILLS (price touches the prior cash close) inside the cash session 57.6% of the time. Weekend gaps are included in the same census. What the gap does not do is predict: both the follow-through and the fade measurements below are DEAD with holdout effects of −0.003 / +0.003 ATR — as close to exactly zero as this census produces. A 57.6% fill rate with no directional edge is the whole story: gaps usually fill, and knowing one happened tells you nothing tradeable about the session that follows.
n = 766 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.171 ATR, holdout -0.003 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 766。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.171 ATR，留出期 -0.003 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 258 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 766频率：每月约 21.5 次（每年约 258 次） · n = 766
Overnight gaps — fade隔夜跳空——反打
DEAD已否定
n = 766 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.171 ATR, holdout +0.003 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 766。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.171 ATR，留出期 +0.003 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 258 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 766频率：每月约 21.5 次（每年约 258 次） · n = 766
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=766.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=766。
First cash hour → rest of session现金第一小时→时段其余时间
DEAD已否定
n = 780 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +0.203 ATR, holdout -0.007 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 780。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +0.203 ATR，留出期 -0.007 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 21.5 per month (≈ 258 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 780频率：每月约 21.5 次（每年约 258 次） · n = 780
Time & Volatility时间与波动
Day types (N-I1, descriptive). On 781 cash days: QUIET OTHER 30.1%, RANGE 29.8%, TREND UP 24.2%, TREND DOWN 13.2%, EXPANSION OTHER 1.9%, spike-reverse ~0.7%. Six days in ten are range or quiet, as in every POSEIDON market — but the trend days lean UP nearly 2:1. That asymmetry is the index's structural drift, and it lives mostly in one block (see Blocks).
ATR regime (N-E1, descriptive). Forward range rises with the trailing ATR regime tercile while forward direction does not — volatility conditions volatility, not direction, the same conditioning shape the FX volumes found.
Earnings-season flag (descriptive). The pre-registered calendar proxy (weeks 2–5 of Jan/Apr/Jul/Oct) was computed and carried in the census files as conditioning only; no directional claim is made from it in this volume.
日类型（N-I1，描述性）。781 个现金日：QUIET OTHER 30.1%, RANGE 29.8%, TREND UP 24.2%, TREND DOWN 13.2%, EXPANSION OTHER 1.9%，spike-reverse 约 0.7%。十天里有六天是区间或安静日，与每个 POSEIDON 市场一样——但趋势日近 2:1 地偏向上涨。这种不对称是指数的结构性漂移，且它主要活在一个时段块里（见'ET 日的各个时段块'章节）。
n = 11 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.621 ATR, holdout +0.126 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 11。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.621 ATR，留出期 +0.126 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.4 per month (≈ 5 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 11频率：每月约 0.4 次（每年约 5 次） · n = 11
After >2×ATR cash days — continuation>2×ATR 现金大日之后——延续
DEAD已否定
n = 18 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +1.042 ATR, holdout -0.998 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 18。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +1.042 ATR，留出期 -0.998 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 18频率：每月约 0.7 次（每年约 8 次） · n = 18
After >2×ATR cash days — |return|>2×ATR 现金大日之后——波动幅度
WEAK弱
n = 18 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.767 ATR, holdout +0.394 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 18。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.767 ATR，留出期 +0.394 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 18频率：每月约 0.7 次（每年约 8 次） · n = 18
Macro Events — from the Cash Open宏观事件——从现金开盘看
Why this chapter looks different. The 08:30 ET releases land in PREMARKET for this instrument — there is no release-bar spike inside the cash session to fade. What the cash session inherits is a gap and a direction, and the gap-direction measure (F_intra0830_gapdir) is DEAD: the FX spike-fade, confirmed in gold, EUR/USD and USD/JPY alike, has no cash-session analogue here. FOMC is the exception that proves the rule at two horizons: the FOMC DAY's direction reverses over five days (WEAK), while the FOMC BAR's direction continues over two hours (WEAK) — both on n ≈ 24, both flags rather than findings.
Yield conditioning (F7-NAS, descriptive). On event days with a ≥ 5bp 10Y move, the index moved INVERSE to the yield 59.2% of the time (quiet-day baseline 40.6%); yields down ≥ 5bp → index median +108 points (+0.56%) that day; the broad dollar down ≥ 0.3% → +212 points (+1.07%). The growth-stock yield logic is real in the data and weaker than USD/JPY's with-yield transmission (84.2%).
The F6 warning. 'Hot'/'cold' use the crude proxy (released m/m vs prior month, NOT consensus surprise), signed as in gold (hot print → short). Every F6 row here is WEAK or DEAD — the fourth market with the same result. No hot/cold trade survives anywhere in POSEIDON.
为什么这一章看起来不同。08:30 ET 发布对这个品种落在盘前——现金时段内没有可回吐的发布 K 线第一波。现金时段继承的是一个跳空和一个方向，而跳空方向测量（F_intra0830_gapdir）已被否定：在黄金、欧元/美元、美元/日元都确认过的外汇第一波回吐，在这里没有现金时段的对应物。FOMC 是在两个周期上证明规则的例外：FOMC 日的方向在五天内反转（弱），而 FOMC 那根 K 线的方向在两小时内延续（弱）——两者都是 n≈24，都是旗标而非结论。
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.235 ATR, holdout +0.451 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.235 ATR，留出期 +0.451 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
NFP days — direction persistence非农日——方向持续性
WEAK弱
n = 33 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.349 ATR, holdout -0.496 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 33。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.349 ATR，留出期 -0.496 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 33频率：每月约 1.0 次（每年约 12 次） · n = 33
PPI days — direction persistencePPI 日——方向持续性
DEAD已否定
n = 34 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.110 ATR, holdout +0.768 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 34。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.110 ATR，留出期 +0.768 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 34频率：每月约 1.0 次（每年约 12 次） · n = 34
Retail Sales days — direction persistence零售销售日——方向持续性
DEAD已否定
n = 35 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.532 ATR, holdout +0.857 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 35。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.532 ATR，留出期 +0.857 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.0 per month (≈ 12 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 35频率：每月约 1.0 次（每年约 12 次） · n = 35
FOMC days — the day's direction reversesFOMC 日——当日方向随后反转
WEAK弱
FOMC-day direction reverses over the following five days in both halves with unusual consistency (−0.821 / −0.823 ATR, n=23). The holdout count (< 30) keeps this WEAK by the charter rule — a flag for the next census, not a finding. n = 23 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.821 ATR, holdout -0.823 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 23频率：每月约 0.7 次（每年约 8 次） · n = 23
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=23.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=23。
All event days pooled — direction persistence全部事件日汇总——方向持续性
DEAD已否定
n = 159 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.035 ATR, holdout +0.433 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 159。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.035 ATR，留出期 +0.433 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 4.6 per month (≈ 55 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 159频率：每月约 4.6 次（每年约 55 次） · n = 159
08:30 releases — the cash-open gap direction08:30 发布——现金开盘跳空方向
DEAD已否定
n = 141 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery -0.221 ATR, holdout +0.358 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 141。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 -0.221 ATR，留出期 +0.358 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 3.9 per month (≈ 47 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 141频率：每月约 3.9 次（每年约 47 次） · n = 141
The FOMC bar itself — it continuesFOMC 那根 K 线本身——随后延续
WEAK弱
The 14:00 ET FOMC bar's own direction continues strongly over the next two hours (+1.624 ATR holdout, n=24). Tiny sample, WEAK by the charter rule; the mirror image of the FOMC-day reversal above, at a different horizon. n = 24 instances. Median forward effect vs the matched random null at the primary horizon (2h): discovery +1.354 ATR, holdout +1.624 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
14:00 ET FOMC 那根 K 线自身的方向在随后两小时强劲延续（留出期 +1.624 ATR，n=24）。样本很小，按章程规则为弱；它与上面 FOMC 日反转在不同周期上互为镜像。样本 n = 24。在主要周期（2h）相对匹配随机基线的中位前向效应：发现期 +1.354 ATR，留出期 +1.624 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.7 per month (≈ 8 per year) · Dukascopy NAS100 CFD (USATECHIDXUSD) M15 · n = 24频率：每月约 0.7 次（每年约 8 次） · n = 24
Stacked forward paths, behaviour-signed, ATR units. Event at 0. n=24.
叠加前向路径，按行为签署，ATR 单位。事件在 0 点。n=24。
Hot prints (crude proxy)热数据（粗略代理）
WEAK弱
n = 68 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.337 ATR, holdout -0.058 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 68。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.337 ATR，留出期 -0.058 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 2.0 per month (≈ 24 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 68频率：每月约 2.0 次（每年约 24 次） · n = 68
Cold prints (crude proxy)冷数据（粗略代理）
WEAK弱
n = 66 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.103 ATR, holdout -0.251 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 66。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.103 ATR，留出期 -0.251 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 1.9 per month (≈ 23 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 66频率：每月约 1.9 次（每年约 23 次） · n = 66
Hot CPI only (crude proxy)仅热 CPI（粗略代理）
DEAD已否定
n = 17 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery +0.117 ATR, holdout -0.328 ATR. Verdict: DEAD under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 17。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 +0.117 ATR，留出期 -0.328 ATR。按章程规则判定为：DEAD（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 17频率：每月约 0.5 次（每年约 6 次） · n = 17
Cold CPI only (crude proxy)仅冷 CPI（粗略代理）
WEAK弱
n = 15 instances. Median forward effect vs the matched random null at the primary horizon (5d): discovery -0.103 ATR, holdout -0.362 ATR. Verdict: WEAK under the charter rule (same sign both halves; confirmed needs |holdout| ≥ 0.05 ATR, ≥ 50% of discovery, holdout n ≥ 30). Gross of trading costs.
样本 n = 15。在主要周期（5d）相对匹配随机基线的中位前向效应：发现期 -0.103 ATR，留出期 -0.362 ATR。按章程规则判定为：WEAK（两半同号；确认需 |留出| ≥ 0.05 ATR、≥ 发现期的 50%、留出 n ≥ 30）。数据未扣交易成本。
Frequency: ≈ 0.5 per month (≈ 6 per year) · Dukascopy NAS100 CFD cash-session D1 derived · n = 15频率：每月约 0.5 次（每年约 6 次） · n = 15
Blocks of the ET DayET 日的各个时段块
Block anatomy on 781 ET days (medians; ATR = prior-day cash ATR). The surprise of this volume is the OVERNIGHT block: on this CFD it is nearly as wide as the cash session itself and carries a persistent upward drift — the index's structural rise happens substantially while New York sleeps, in the CFD's quotes. This is a fact about USATECHIDXUSD's overnight feed, stated as such; it is not asserted of the cash index.
Block (ET)
Median range (pts)
Median range (ATR)
Median net (pts)
Share net up
Overnight 20:00–04:00
223
0.70
+24.9
55.7%
Premarket 04:00–09:30
124
0.38
+2.6
51.2%
Cash 09:30–16:00
254
0.79
+15.2
53.9%
After-hours 16:00–20:00
58
0.19
+1.4
51.6%
Where the cash extremes form. The cash day's high is first set in the opening 30 minutes on 29.7% of days, its low on 35.2%; the last 30 minutes set 16.4% / 11.8%. Extremes live at the edges — the fourth POSEIDON market with this shape. The cash range is a median 2.03× the premarket range.
The state-filter question is closed — three times. Gold's confirmed behaviours split by NAS100's state (XN1): gold's accepted breaks continue in both NAS up-days (holdout +0.16 ATR) and down-days (+0.08); gold's mid-range engulfing is near-identical across NAS states (+0.08 vs +0.16). The concurrent-2h NAS states are too rare during gold events (n = 7–10) to condition on at all. With the EUR/USD-state filter (Vol. II) and the yen-state filter (Vol. III) also dead, no concurrent-market state conditions gold's confirmed behaviours. The desk's filter question is answered, negatively, three times over (logged write-back to the gold chapters, 2026-10-16).
Risk co-movement (XN3). Gold's return over the exact NAS cash window (DST-aware) correlates +0.23 with the index's cash return — a mild risk co-movement, far weaker than the dollar channels (|r| ≈ 0.5). NAS cash → gold next-2h: r = +0.04, no lead.
The mirror (XN2). NAS A1 breaks split by gold's state flip between discovery and holdout — gold's state does not condition the index's breaks either. Two markets that share macro weather but not tradeable states.
状态过滤问题已关闭——三次。按纳斯达克100 的状态拆分黄金的已确认行为（XN1）：黄金被接受的突破在 NAS 上涨日（留出期 +0.16 ATR）和下跌日（+0.08）都延续；黄金的中位吞没在两种 NAS 状态下几乎相同（+0.08 对 +0.16）。黄金事件期间同时出现的 2 小时 NAS 状态太稀少（n=7–10），根本无法据此条件化。加上欧元/美元状态过滤器（第二卷）和日元状态过滤器（第三卷）也已否定，没有任何同期市场状态能条件化黄金的已确认行为。本桌的过滤问题已被否定地回答了三次（写入日志至黄金章节，2026-10-16）。
风险共同运动（XN3）。黄金在 NAS 现金窗口（按日期感知夏令时）内的收益与指数的现金收益相关性为 +0.23——温和的风险共同运动，远弱于美元通道（|r|≈0.5）。NAS 现金 → 黄金随后 2 小时：r=+0.04，没有领先。
镜像（XN2）。按黄金状态拆分的 NAS A1 突破在发现期与留出期之间翻转——黄金的状态也不能条件化指数的突破。两个共享宏观天气、但不共享可交易状态的市场。
Costs, Sizing, Methods & Limits成本、仓位、方法与局限
Cost audit. Measured on this corpus: median cash M15 ATR 31.4 points, median cash-day ATR 328 points, median cash-day range 254 points (1.19% of price). At the pre-registered mid case (2.0-pt spread + 1.0 slippage), a round trip costs ≈ 3.0 points = 0.096× the median cash M15 ATR — friction per unit of movement is gold-like, not FX-like. In R: a 25-pt stop pays 0.12R per trade, 50-pt 0.06R, 100-pt 0.03R.
Sizing is the wall. At the MT5 minimum (0.01 lot = 1 unit = $1 per point), a 25-pt stop risks $25 — 16.8% of a $149 account; a 50-pt stop risks $50 (33.6%); a 100-pt stop $100 (67.1%). The minimum account for a 50-pt stop at the desk's 1% risk is $5,000. On the desk's current account, NAS100 cannot be traded inside desk risk rules no matter what the census found — this volume is a research asset until the account grows, exactly as its definitions pre-registered. At a proper account size, the confirmed effects that clear friction with room are the streak-3 fade (≈ −130 pts at 5d), the prior-week break fade (≈ −215 pts) and, thinly, the premarket-break continuation (≈ +10.6 pts at 2h vs 3 pts of costs).
Methods. Definitions were fixed in writing before measurement (Volume IV DEFINITIONS + CHARTER_NOTE). Deliberate deviations from the FX volumes: ET/DST-aware sessions instead of fixed UTC; the trading day is the ET date; levels are prior CASH extremes; gaps are first-class behaviours, never dropped. The validity rule (ET weekday + cash range > 0 + ≥ 20 cash bars) matches the FX volumes' valid-day filter and excluded 317 flat-quote weekend/holiday phantoms the feed writes — measured and reported, not silently patched. Inside bars are measured on |return| (program convention; their direction is defined by the subsequent break). Nulls are random cash bars; the null is matched on the cash block, not on hour-of-day — the inside-bar result was therefore re-checked by hand within every cash hour and holds in each.
Limits. (1) CFD ≠ index ≠ futures: overnight quotes are the broker feed's, and the overnight block's drift is a fact about this feed. (2) ~750–780 cash days is a thin daily sample; daily verdicts (streaks, weekly levels, FOMC) carry proportionally less weight, and several rest on n ≈ 100 or less. (3) The earnings flag is a calendar proxy, not a company calendar. (4) With 50 behaviours, a few CONFIRMED labels are expected false positives; counts and effect sizes are the evidence. (5) Census numbers are gross of costs and are NOT approved strategies — anything trade-shaped must still pass the desk's readiness bar (mechanical definition, costed backtest, ≥ 50 trades, PF ≥ 1.3 out-of-sample, ≥ 2 market periods).
POSEIDON, drawn simply — the movements that repeat, market by marketPOSEIDON 简单版——四个市场最重复出现的走势
The honest part first先说实话
These are the most repetitive movements the POSEIDON census found in three years of data, drawn as simply as possible so a total beginner can see them.这些是 POSEIDON 普查在三年数据里找到的最重复出现的走势，用最简单的方式画出来，让完全的新手也能看懂。
They are learning patterns, NOT approved strategies. Nothing here has passed this desk’s backtest bar (50+ trades, profit factor ≥ 1.3 on unseen data). Practise on demo first. If you trade them live at all, trade tiny.它们是学习用的形态，不是已批准的策略。这里没有任何一条通过本交易台的回测门槛（50 笔以上交易、在未见数据上利润因子 ≥ 1.3）。先在模拟账户练习；如果一定要实盘，只能用极小的仓位。
A pattern that repeated 500 times can still fail the next 50 times. The “it fails when” line on every card is your exit — that line, not hope, is what keeps you in the game.一个重复了 500 次的形态，接下来 50 次仍然可能失效。每张卡上的「什么时候失效」就是你的出口——让你留在牌桌上的，是那条线，不是希望。
1 · Draw the lines1 · 画线→2 · Wait for the CLOSE2 · 等收盘确认→3 · Small size3 · 小仓位→4 · The “fails when” line is the exit4 · 失效线就是出口
Gold黄金
Gold trends inside the day and loves a trap at the edges.黄金在日内喜欢走趋势，也最喜欢在边缘设陷阱。
The break that sticks守得住的突破才算数
Draw this画这个
Yesterday's high and yesterday's low — two dashed lines across your chart.昨天的最高价和最低价——在图上画两条虚线。
Wait for this等这个信号
A 15-minute candle CLOSES beyond a line, and the next candles stay outside. A wick poking out does not count.一根 15 分钟 K 线收盘站到线外，之后的 K 线也留在外面。只有影线戳出去不算。
The rule规律
A break that closes and holds usually keeps going the same day. Follow the close, not the poke.收盘站稳的突破，当天通常会继续走。跟随收盘，不要跟随影线。
It fails when什么时候失效
Price closes back inside yesterday's range — the break was a trap and the idea is dead.价格收盘回到昨日区间里面——突破是陷阱，这个想法作废。
Census: confirmed in POSEIDON Vol I — accepted prior-day breaks kept going in both the discovery data and the unseen holdout data (3 years of gold, M15).普查：POSEIDON 第一卷已确认——被接受的前日突破在发现数据和未见留出数据中都会延续（三年黄金 M15 数据）。
The first news spike lies第一波新闻冲高是假的
Draw this画这个
Nothing new — just mark the clock: big US data lands at 08:30 ET (20:30 MYT; 21:30 after 1 Nov).不用画新东西——标好时间：美国重磅数据在美东 08:30 公布（马来西亚 20:30；11 月 1 日后为 21:30）。
Wait for this等这个信号
Let the first 15-minute spike finish. Watch what the second and third candles do instead.让第一波 15 分钟冲高走完。看第二、第三根 K 线怎么做。
The rule规律
The first spike is usually the fake — price followed it only ~43% of the time (on CPI days, 39%). Never chase spike number one.第一波冲高通常是假的——价格跟随它的概率只有约 43%（CPI 日只有 39%）。永远不要追第一波。
It fails when什么时候失效
You enter during the spike candle itself. That is exactly the mistake this pattern exists to stop.你在冲高那根 K 线还没收盘时就进场。这正是这个形态要拦住的错误。
Census: confirmed in POSEIDON Vol I — the pooled 08:30 ET first spike was faded across CPI, PPI and jobs releases.普查：POSEIDON 第一卷已确认——08:30 ET 第一波冲高在 CPI、PPI 和就业数据中普遍被反打。
London's first breakout is a trap伦敦第一波突破多半是陷阱
Draw this画这个
Asia's high and low — draw the pre-London range as a box.亚洲时段的最高和最低——把伦敦开盘前的区间画成一个框。
Wait for this等这个信号
London's first hour pokes out of the box — then candles start CLOSING back inside the box.伦敦第一小时戳出框外——然后 K 线开始收盘回到框内。
The rule规律
London's first breakout usually fails and comes back into the range.伦敦的第一波突破通常会失败，回到区间里面。
It fails when什么时候失效
Candles keep closing outside the box, one after another — that breakout is real. Do not fade it.K 线一根接一根收在框外——那次突破是真的。不要反打它。
Census: confirmed in POSEIDON Vol I — London first-hour breakouts fade.普查：POSEIDON 第一卷已确认——伦敦第一小时突破会被反打。
The reliable handoff最可靠的交接
Draw this画这个
Mark London's direction — on balance, was London rising or falling into the overlap?标出伦敦的方向——总体上，伦敦进入重叠时段时是涨还是跌？
Wait for this等这个信号
The London–New York overlap (12:30–16:30 UTC) starts moving the same way London was leaning.伦敦–纽约重叠时段（UTC 12:30–16:30）开始朝伦敦倾斜的方向走。
The rule规律
Of all gold's session handoffs, this one continues most reliably — swim with it, not against it.在黄金所有时段交接中，这个延续最可靠——顺着它游，不要逆着它。
It fails when什么时候失效
A red US release lands inside the overlap — news overrules the handoff.重叠时段里落下一条红色美国数据——新闻会推翻交接规律。
Census: +0.42 ATR median continuation, confirmed on two separate corpora in POSEIDON Vol I.普查：延续中位数 +0.42 ATR，在 POSEIDON 第一卷的两个独立语料上都得到确认。
Three days one way = tired连走三天就会累
Draw this画这个
On the daily chart, count the closes: 3 or more green days in a row — or 3 or more red.在日线图上数收盘：连续 3 天或以上收涨——或连续收跌。
Wait for this等这个信号
The streak itself is the warning. Do not chase day four in the streak's direction.连涨连跌本身就是警告。不要在第四天追着连胜的方向跑。
The rule规律
Streaks exhaust — after 3+ one-way days, price usually snaps back within days.连胜会耗尽——单边走 3 天以上后，价格通常在几天内弹回。
It fails when什么时候失效
A major news trend is driving the streak — check the calendar before fading anything.有重大新闻趋势在推动这波连胜——反打之前先看日历。
Census: confirmed in POSEIDON Vol I — daily streaks exhaust and fade.普查：POSEIDON 第一卷已确认——日线连胜会耗尽并反转。
The morning raid早晨的扫盘
Draw this画这个
The Asia base — the flat floor price sat on for hours — and the line where price reclaims it.亚洲底座——价格横盘数小时的平坦底部——以及价格收复它的那条线。
Wait for this等这个信号
A sudden flush dives $3–4 under the base… then 15-minute candles CLOSE back above it. That reclaim close is the moment.一波急跌突然扎到基座下方 $3–4……然后 15 分钟 K 线收盘回到基座上方。那个收复收盘就是时机。
The rule规律
The flush was the trap; the reclaim is the trade. Never buy the falling flush — buy the close back above.急跌是陷阱，收复才是交易。永远不要接下跌中的急跌——买收盘站回上方。
It fails when什么时候失效
Candles close and STAY below the base — that is a real breakdown, not a raid. Stand aside.K 线收盘并停留在基座下方——那是真破位，不是扫盘。站到一边。
Desk log: from live desk trades, Oct 2026 — NOT census. Seen live on the mornings of 15–16 Oct 2026.交易台记录：来自 2026 年 10 月的实盘观察——不是普查数据。2026 年 10 月 15–16 日早晨亲眼所见。
EUR/USD欧元/美元
Where gold continues, EUR/USD fades.黄金延续的地方，欧元/美元反打。
Yesterday's break fails昨日的突破会失败
Draw this画这个
Yesterday's high and low — the same two dashed lines as gold.昨天的最高和最低——和黄金一样的两条虚线。
Wait for this等这个信号
A candle closes beyond the line... then closes start coming back inside.一根 K 线收盘站到线外……然后收盘开始回到线内。
The rule规律
In EUR/USD the accepted break usually REVERSES — the break gold traders follow, euro traders fade.在欧元/美元，被接受的突破通常会反转——黄金交易者跟随的突破，欧元交易者反打。
It fails when什么时候失效
The break comes on a big news day and keeps closing further out — that time acceptance is real. Do not step in front of it.突破发生在重磅新闻日并且不断收在更外侧——那一次是真的被接受。不要挡在它前面。
Census: watched 540 times over 3 years — average -0.359 ATR in the unseen holdout data.普查：三年里观察了 540 次——在未见留出数据中平均 -0.359 ATR。
The first news spike lies — even harder第一波冲高，假得更彻底
Draw this画这个
Nothing — just the clock: 08:30 ET (20:30 MYT; 21:30 after 1 Nov).什么都不用画——只有时间：美东 08:30（马来西亚 20:30；11 月 1 日后 21:30）。
Wait for this等这个信号
The first 15-minute spike completes. The fade usually starts right after it.第一波 15 分钟冲高走完。反打通常紧接着开始。
The rule规律
This is the strongest simple pattern in this market: the first spike reverses about 4.3 pips on average.这是这个市场最强的简单形态：第一波冲高平均反转约 4.3 点。
It fails when什么时候失效
Two spikes in a row in the same direction, with closes holding the extreme — rare, but then it is a real repricing, not a fake.同方向连续两波冲高且收盘守住极值——很少见，但那时是真重定价，不是假动作。
EUR/USD moves are small: a typical 15-minute candle is only ~5.5 pips, and the spread eats about 0.20× of a typical move — roughly 3× gold’s friction. Only the big fades above are worth a beginner’s money. If a pattern’s expected move is not clearly bigger than the spread, skip it.欧元/美元的波动很小：一根典型的 15 分钟 K 线只有约 5.5 点，而点差吃掉约 0.20 倍的典型波动——大约是黄金摩擦的 3 倍。只有上面那些大幅反打才值得新手花钱。如果一个形态的预期波动没有明显大于点差，跳过它。
USD/JPY美元/日元
Gold's haven twin, the dollar's yield slave.黄金的避险孪生兄弟，美元收益率的奴隶。
The break that sticks守得住的突破才算数
Draw this画这个
Yesterday's high and low — two dashed lines, exactly like gold.昨天的最高和最低——两条虚线，和黄金完全一样。
Wait for this等这个信号
A 15-minute candle CLOSES beyond a line and the next candles hold outside it.一根 15 分钟 K 线收盘站到线外，之后的 K 线守在外面。
The rule规律
Like gold (and UNLIKE EUR/USD), an accepted break here usually CONTINUES.和黄金一样（和欧元/美元相反），这里被接受的突破通常会延续。
It fails when什么时候失效
Closes come back inside yesterday's range — the break failed; the idea is dead.收盘回到昨日区间内——突破失败，这个想法作废。
Census: watched 524 times — +0.112 ATR average in the unseen holdout data.普查：观察了 524 次——在未见留出数据中平均 +0.112 ATR。
The evening flip傍晚的反转
Draw this画这个
Mark what the London–New York overlap did — its net direction (12:30–16:30 UTC).标出伦敦–纽约重叠时段做了什么——它的净方向（UTC 12:30–16:30）。
Wait for this等这个信号
The New York afternoon starts moving the OTHER way.纽约下午开始朝反方向走。
The rule规律
This market's strongest handoff is a reversal: the overlap leans one way, the NY afternoon flips it.这个市场最强的交接是反转：重叠时段朝一边倾斜，纽约下午把它翻过来。
It fails when什么时候失效
Big US news lands in the NY afternoon — news flow beats the flip.纽约下午落下重磅美国新闻——新闻流会压过反转规律。
Census: watched 784 days — average -0.411 ATR ≈ -4.2 pips.普查：观察了 784 天——平均 -0.411 ATR ≈ -4.2 点。
The first news spike lies第一波新闻冲高是假的
Draw this画这个
Nothing — just the clock: 08:30 ET (20:30 MYT; 21:30 after 1 Nov). Third market, same clock.什么都不用画——只有时间：美东 08:30（马来西亚 20:30；11 月 1 日后 21:30）。第三个市场，同一个时间。
Wait for this等这个信号
Let the first 15-minute spike finish before you even think.让第一波 15 分钟冲高走完，之后再考虑任何事。
The rule规律
Same lesson as gold and the euro: the first spike is the trap, the fade is the pattern.和黄金、欧元同样的教训：第一波冲高是陷阱，反打才是形态。
It fails when什么时候失效
Closes hold the spike's extreme and a second spike extends it — a real repricing is under way.收盘守住冲高的极值且第二波继续延伸——真重定价正在发生。
Nothing on the price chart — open the US 10-year yield next to it.价格图上什么都不画——在旁边打开美国 10 年期收益率。
Wait for this等这个信号
A day when the 10-year yield jumps hard — up or down.10 年期收益率大幅跳动的一天——无论涨跌。
The rule规律
USD/JPY is the dollar's yield slave: when yields spike up, the pair wants up; when yields drop, the pair wants down.美元/日元是美元收益率的奴隶：收益率飙升，货币对想涨；收益率下跌，货币对想跌。
It fails when什么时候失效
It is a Bank of Japan day — Tokyo can overrule the yield for a session.正值日本央行日——东京可以在一个时段内推翻收益率的指引。
Census: on 76 big-yield event days, the pair moved WITH the yield 84% of the time.普查：在 76 个收益率大动的日子里，货币对有 84% 的时间与收益率同向。
Three up days = tired三连涨就会累
Draw this画这个
On the daily chart, count 3 straight up closes — or 3 straight down.在日线图上数：连续 3 天收涨——或连续 3 天收跌。
Wait for this等这个信号
The streak completes. Do not chase day four.连胜完成。不要追第四天。
The rule规律
Three-day streaks fade hard in this market — the snapback is the biggest of the FX volumes.三日连胜在这个市场反打得很猛——弹回幅度是外汇各卷中最大的。
It fails when什么时候失效
The streak is riding Bank of Japan or intervention headlines — check before fading.连胜正骑着日本央行或干预头条——反打之前先查清楚。
Census: watched 91 streaks — average -27.9 pips over the next 5 days.普查：观察了 91 次连胜——接下来 5 天平均 -27.9 点。
NAS100纳斯达克100
Trends by day, fades its own streaks.按天走趋势，反打自己的连胜。
The break that sticks守得住的突破才算数
Draw this画这个
Yesterday's high and low — two dashed lines, same as gold.昨天的最高和最低——两条虚线，和黄金一样。
Wait for this等这个信号
Cash-session candles CLOSE beyond the line and hold outside it.现金时段的 K 线收盘站到线外并守在外面。
The rule规律
Accepted prior-day breaks CONTINUE here — NAS100 is a continuation market at the daily line. (EUR/USD is the only fader of the four.)这里被接受的前日突破会延续——纳斯达克100 在日线价位上是延续型市场。（四个市场中只有欧元/美元是反打型。）
It fails when什么时候失效
Closes fall back inside yesterday's range — failed break; the idea is dead.收盘跌回昨日区间内——突破失败，这个想法作废。
Census: watched 557 times — +0.111 ATR ≈ +8.7 pts in the unseen holdout data.普查：观察了 557 次——在未见留出数据中 +0.111 ATR ≈ +8.7 点。
The premarket break is real盘前突破是真的
Draw this画这个
The premarket box — the high and low before the 09:30 ET cash open (21:30 MYT).盘前框——美东 09:30 现金开盘（马来西亚 21:30）之前的最高和最低。
Wait for this等这个信号
After the open, candles CLOSE beyond the box and hold outside it.开盘之后，K 线收盘站到框外并守住。
The rule规律
Premarket breaks usually continue — this market respects its premarket range.盘前突破通常会延续——这个市场尊重自己的盘前区间。
It fails when什么时候失效
The break closes back inside the box within the first hour — failed break; the idea is dead.突破在第一小时内收盘回到框内——突破失败，这个想法作废。
Census: watched 606 times — +0.222 ATR average continuation.普查：观察了 606 次——平均延续 +0.222 ATR。
Last WEEK's break collapses上周的突破会崩
Draw this画这个
Last week's high and low — the two big lines on your chart.上周的最高和最低——图上那两条大线。
Wait for this等这个信号
Price closes beyond a weekly line... then cannot hold it, and closes fall back inside.价格收盘站到周线之外……然后守不住，收盘跌回线内。
The rule规律
Daily breaks continue here, but WEEKLY breaks collapse — the biggest fade in this whole book.这里日线突破会延续，但周线突破会崩——这是全书最大的反打。
It fails when什么时候失效
The weekly break is riding a huge earnings or news week — check what is driving it before fading anything.周线突破正骑着重磅财报周或新闻周——反打之前先查清驱动因素。
Census: watched 99 times — average collapse -0.671 ATR ≈ -215 pts.普查：观察了 99 次——平均崩跌 -0.671 ATR ≈ -215 点。
Three-day streaks snap三日连胜会断
Draw this画这个
On the daily chart, count 3 straight up closes — or 3 straight down.在日线图上数：连续 3 天收涨——或连续 3 天收跌。
Wait for this等这个信号
The streak completes. Do not chase day four.连胜完成。不要追第四天。
The rule规律
Streaks snap back here harder than in any of the other three markets.这里连胜的弹回比其他三个市场都猛。
It fails when什么时候失效
The streak is riding a major earnings week — check the calendar before fading.连胜正骑着重磅财报周——反打之前先看日历。
The gap means NOTHING — do not trade it跳空什么都不说明——不要交易它
Draw this画这个
Yesterday's close and today's open — the overnight gap between them.昨天的收盘和今天的开盘——它们之间的隔夜跳空。
Wait for this等这个信号
Nothing. There is no trigger — that IS the lesson.什么都不等。没有触发信号——这正是这一课。
The rule规律
Gap direction carries ZERO information. Follow-the-gap: tested dead. Fade-the-gap: tested dead. Gaps fill ~58% of the time, but which way price goes first tells you nothing.跳空方向携带的信息为零。跟随跳空：测试已死。反打跳空：测试已死。跳空约 58% 的时间会被回补，但价格先往哪边走什么都说明不了。
It fails when什么时候失效
It does not fail — traders fail it, by betting on gap direction. Ignore the gap; trade the cash session’s own levels instead.它不会失效——失效的是交易者，他们赌跳空方向。忽略跳空；改为交易现金时段自己的价位。
Census: tested both ways — follow ≈ 0.000 and fade ≈ 0.000 in unseen holdout data. The cleanest bust in POSEIDON.普查：两个方向都测过——在未见留出数据中跟随 ≈ 0.000、反打 ≈ 0.000。POSEIDON 中最干净的证伪。
Sizing warning — read before any NAS100 trade仓位警告——交易纳斯达克100 之前必读
On MT5, 0.01 lot of NAS100 = $1 per point. A normal 50-point stop = $50 of risk on ONE trade. Until an account is roughly $5,000, NAS100 is a research market only: study it, paper-trade it, do not fund it.在 MT5 上，纳斯达克100 的 0.01 手 = 每点 $1。一个正常的 50 点止损 = 单笔交易 $50 风险。在账户达到约 $5,000 之前，纳斯达克100 只是研究市场：研究它、模拟交易它，不要用真金白银做它。
Set up for easy scalps — the beginner’s desk轻松剥头皮的布置——新手交易台
Five steps, in order. The setup is the strategy’s bodyguard: most beginner losses are setup losses, not idea losses.五个步骤，按顺序来。布置是策略的保镖：新手的大多数亏损是布置的亏损，不是想法的亏损。
Honesty first:先说诚实话：everything on this page is a census pattern for demo / tiny-size practice — NOT an approved strategy. The desk readiness bar (≥50 backtested trades, profit factor ≥ 1.3 out-of-sample, ≥2 market periods) is still unmet by everything here.本页的一切都是供模拟 / 极小仓位练习的普查模式——不是已批准的策略。交易台的就绪门槛（回测交易 ≥50 笔、样本外利润因子 ≥ 1.3、≥2 个市场时期）在这里仍然没有任何东西达到。
1
Pick ONE window选一个窗口
One session, on the clock below. Outside it, the charts are closed — a beginner trades a window, not a day.在下方时钟上选一个时段。时段之外，图表关闭——新手交易的是一个窗口，不是一整天。
2
Set up the chart布置图表
M15 candles. Yesterday’s high and low as two lines. Session boxes. Nothing else — no indicator stacks.M15 K 线。昨日的最高和最低画两条线。时段框。其他什么都不要——不要堆指标。
3
One play per market每个市场一个打法
The cards under step 3 are the whole menu. If your play is not on a card, it is not a trade today.步骤 3 下面的卡片就是全部菜单。你的打法如果不在卡上，今天它就不是交易。
4
Run the fresh ticket执行新票据清单
Every box on the checklist, ticked BEFORE entry — agreed in chat first. No fresh ticket, no trade.清单上的每一格，在入场前打勾——先在聊天里商定。没有新票据，就没有交易。
5
Size it tiny — on purpose故意把仓位做小
The stop decides the dollars BEFORE the entry does. If the size feels like nothing, it is probably right.止损在入场之前就决定了美元数。如果仓位小到感觉没什么，那大概是对的。
Step 1 · the clock (MYT)步骤 1 · 时钟（MYT）
One day on the clock时钟上的一天Malaysian time. One window is enough — the Overlap (20:00–23:00) is where London and New York fight, and where the range is widest.马来西亚时间。一个窗口就够了——重叠时段（20:00–23:00）是伦敦和纽约交锋的地方，也是区间最宽的地方。
Quiet安静
sleep
04–10 MYT
Asia亚洲
warming up
10–15
London伦敦
moving
15–20
Overlap重叠
MOST movement
20–23
New York纽约
moving
23–04
Step 2 · the chart — two lines and a box, that’s all步骤 2 · 图表——两条线和一个框，仅此而已
If a beginner’s chart needs a manual, it has too much on it. These three drawings carry every play in this book.如果新手的图表需要说明书，那它上面的东西太多了。这三样画法承载了本书的每一个打法。
The two-tranche shape两批的形状
One idea, two half positions at the same entry. Tranche 1 takes profit early, inside the zone; tranche 2 rides further. One shared stop, beyond the structure.一个想法，两个半仓，同一个入场价。第一批早早止盈，在区域之内；第二批走得更远。一个共用止损，在结构之外。
Step 3 · the whole menu — one play per market步骤 3 · 全部菜单——每个市场一个打法
GOLD · Play 1 — the first spike lies黄金 · 打法 1——第一波冲高说谎
On 08:30 ET data days, do not chase the first spike — it is followed only 43% of the time (CPI: 39%). Wait for the spike, then trade the fade as it stalls and closes back.在 08:30 ET 数据日，不要追第一波冲高——它只有 43% 的时候被跟随（CPI：39%）。等冲高出现，然后在它停滞并收回时做反打。
GOLD · Play 2 — the reliable handoff黄金 · 打法 2——可靠的接力
London’s direction is handed to the Overlap and kept. Trade WITH the handoff, not against it.伦敦的方向被交给重叠时段并被守住。顺着接力交易，不要逆着它。
CENSUS · +0.42 ATR median普查 · 中位数 +0.42 ATR
EUR/USD · the spike fade欧元/美元 · 冲高反打
Same play, second market: fade the pooled 08:30 ET first spike over the next 2 hours. One of the few EUR/USD effects big enough to beat its 0.20× friction.同样的打法，第二个市场：反打汇总的 08:30 ET 第一波冲高，接下来 2 小时。它是少数能打败欧元/美元 0.20 倍摩擦的效应之一。
CENSUS · −4.3 pips · n=142普查 · −4.3 点 · n=142
USD/JPY · the evening flip美元/日元 · 晚间反转
The Overlap pushes one way; New York reverses it. Expect the flip, position for the reversal, judge it on closes.重叠时段朝一个方向推；纽约把它反转。预期反转、为反转布局、用收盘来判断。
At 0.01 lot NAS100 moves $1 per point — a normal 50-point stop risks $50 on one trade. Until the account is roughly $5,000 this market is research-only: study it, paper-trade it, do not fund it.纳斯达克100 的 0.01 手每点动 $1——一个正常的 50 点止损单笔就风险 $50。在账户达到约 $5,000 之前，这个市场只做研究：研究它、模拟它，不要用真钱做它。
RESEARCH ONLY · min. account ≈ $5,000仅供研究 · 最低账户 ≈ $5,000
Step 4 · the fresh-ticket checklist步骤 4 · 新票据清单
Agreed in chat BEFORE entry — no fresh ticket, no trade.入场前先在聊天里商定——没有新票据，就没有交易。
Entry price written down. Trigger judged on candle CLOSES, not wicks.写下入场价。触发用 K 线收盘判断，不用影线。
Stop placed beyond the structure — where the idea dies, plus a small wick buffer. Never inside a level, never widened later.止损放在结构之外——想法死亡的地方，再加一点影线缓冲。永不放在价位之内，之后永不加宽。
Take-profit set INSIDE the zone, a few dollars before the crowded obvious level.止盈设在区域之内，在拥挤的显眼价位之前几美元。
Two tranches, 50/50: one MT5 position each, each with its own TP, both sharing the same SL.两批，各 50%：每个 MT5 仓位独立，各有自己的止盈，共用同一个止损。
Size chosen so the stop = the planned dollar loss. The stop decides the size — never the other way round.仓位大小让止损 = 计划中的美元亏损。由止损决定仓位——永不反过来。
Kill gates set: −3R ends the day, −6R ends the week. No exceptions, no "one more".设好熔断：−3R 结束当天，−6R 结束本周。没有例外，没有「再来一单」。
After a winning session the book is CLOSED. The next trade needs a fresh ticket.盈利时段之后，账本关闭。下一笔交易需要一张新票据。
Step 5 · sizing reality步骤 5 · 仓位的现实
$1
per $1 move — what 0.01 lot of gold earns or loses per dollar每动 $1——黄金 0.01 手每美元的盈亏
$10
risk on a $10 stop at 0.01 lot. Two tranches = $20 total0.01 手、$10 止损的风险。两批 = 总共 $20
−3R
daily kill gate — hit it and the day is over每日熔断——碰到它，今天结束
−6R
weekly kill gate — hit it and the week is over每周熔断——碰到它，本周结束
Never size up to “feel” the trade. Feeling it is what the stop is for — and the stop was priced before the entry existed.永不为了「有感觉」而加大仓位。有感觉是止损的工作——而且止损在入场存在之前就已经定价了。
