Battle-Tested.
Documented.
Reproducible.
Every number on this page comes from a four-book walk-forward backtest of the protocol across six and a half years of public exchange data. 66.3% win rate across 6,147 trades on Bitcoin, Ethereum, Solana and Gold, +0.293R per trade net of commission, and a −20.15R portfolio drawdown along the way — with a worst losing run of eight trades in a row. BTC’s book alone ran 65.8% on n=2,009 for +502.9R with a deeper −15.51R trough; all four books are published separately below. The 73.8% BTC peak figure quoted elsewhere is the confluence tier (n=1,084); the 70.9% ETH peak is the same tier on ETH (n=574). Exits follow the published ladder — half the position at 1R, stop to break-even, the remainder to 4R. No cherry-picked trades. No curve-fitted models. Walk-forward, no refit, every table published below. Every ladder you will see is the caps’ built-in grading engine at work — setups scored across the confluence stack, tiered by strength, sized by tier.
4,076 Wins.
2,071 Losses.
6,147 Trades.
The grid below shows the proportion of wins to losses across all 6,147 trades — one dot for every one of the 6,147 trades, wins massed first so the proportion reads at a glance. Orange = win. Dark = loss. A win means the trade finished above break-even after commission, not that it touched a target. At n=6,147 the 95% confidence interval on the win rate is 65.1–67.5% — narrow enough to be a measurement rather than a suggestion, which is the whole reason the sample was rebuilt this size.
The Rules That
Define a Win
Performance data means nothing without a precise definition. These are the exact rules used to log every result on this page — applied identically to every setup, without exception.
A Win Is Recorded When
- All 5 CAP gates confirmed before entry
- Price reaches the first defined target (min 1:2 R) before stop is hit
- Stop loss was set at entry and never adjusted
- Setup occurred within a valid NY / London / Overlap session
A Loss Is Recorded When
- Price hits the pre-set stop loss before reaching first target
- Setup is disqualified at any gate after entry (logged as loss)
- No partial closes counted as wins — full position measured
- Re-entries on same setup treated as separate setups
What Is Excluded
- Setups that failed any gate before entry (never entered the market)
- Off-session price action (outside NY/London/Overlap)
- Setups during major macro events (FOMC, CPI release days)
- Setups where execution was technically unavailable
How 120 Scans
Become 12 Trades
The 73.8% peak win rate is not achieved by finding better entries — it is achieved by eliminating worse ones. The 5-gate sequence is a progressive filter. Each gate removes setups that carry higher-probability failure modes.
6,147 Trades.
One Month Under.
+1,804R.
Four books — Bitcoin, Ethereum, Solana and Gold — run side by side across six and a half years at 1% risk per trade. The interesting part is not the final number, it is the shape of the path: the portfolio closed one of eighty months below its starting line — its first — and its deepest peak-to-trough drawdown measured trade by trade was −20.15R, with a worst run of eight consecutive losing trades. The books do not move together, and each one’s own curve, drawdown and confidence interval is published separately below. We publish this in R, not in dollars. R is a unit of risk, not a return — what it compounds to depends entirely on how much you risk per trade and how many positions you carry at once, and those are decisions this backtest does not make for you.
Every exit in the tables above was taken by the rule set, not by a person. A backtest can only test what is mechanical, so the only exits available to it are the TP ladder and the 30-bar timer — and on BTC it was the timer, not a target, that closed half the sample. Nothing in this test was held longer because the structure said to hold it.
That is a real limit, and it is worth naming precisely. Runner management on this desk is done on Elliott Wave structure — a read of where a move sits in its wave count. That is a judgement about context, not a condition that can be written as IF/THEN, so it cannot be encoded into the five gates and it cannot be encoded into this backtest. It is excluded here for exactly that reason, not because it is unimportant.
So read every figure on this page as what the mechanical protocol produces on its own, with no judgement applied at any point after entry. A discretionary overlay moves the outcome, and this page publishes no number for it — there is no backtest of a judgement call, and we are not going to invent one. What is published here is the part that can be checked.
The Mathematics of a Systematic Edge
The full documented record: 6,147 trades across all four books — BTC (n=2,009), ETH (n=1,618), SOL (n=724) and Gold (n=1,796) — plotted month by month from January 2020 to August 2026. 4,076 wins, 2,071 losses. Every point is the sum of that month’s closed trades, and every trade is generated by the same rules published above, so the curve and the tables cannot disagree. Nothing smoothed and nothing removed — including the opening month below the line and every drawdown since. It draws identically every time because it is the measured backtest, not a simulation.
| Year | Trades | W / L | WR | Net R | Cumulative |
|---|---|---|---|---|---|
| 2020 | 267 | 183W / 84L | 68.5% | +117.2 | +117.2R |
| 2021 | 659 | 448W / 211L | 68.0% | +212.0 | +329.2R |
| 2022 | 815 | 531W / 284L | 65.2% | +210.6 | +539.8R |
| 2023 | 860 | 557W / 303L | 64.8% | +213.8 | +753.6R |
| 2024 | 1,202 | 783W / 419L | 65.1% | +328.3 | +1081.9R |
| 2025 | 1,352 | 894W / 458L | 66.1% | +397.7 | +1479.6R |
| 2026 | 992 | 680W / 312L | 68.5% | +324.2 | +1803.8R |
| Total | 6,147 | 4,076W / 2,071L | 66.3% | +1803.8 | +1803.8R |
Every row is the sum of that year’s four per-book rows in the Monthly Equity Attribution tables published further down this page. Nothing is reconstructed trade by trade, and the column sums to the published +1,803.8R total. Portfolio view: the four books run concurrently at 1% risk per trade, so up to 4% of the account can be at risk at once; each book’s standalone curve, drawdown and confidence interval is published separately below. Every calendar year is positive and none falls below 64.8% — including 2022, a full crypto bear market, and 2020, which contains the March crash. One month of eighty closed below the starting line, and it was the first.
How the Protocol
Was Actually Born
The CAP Framework did not come from intuition. It did not come from theory. It was derived — ruthlessly — by applying advanced AI analysis to 12 months of historical price action across BTCUSDT, ETHUSDT and XAUUSD simultaneously, on every timeframe from the daily down to the 15-minute. The 1-hour is the configuration that survived, and it is the only one carried forward into a published table.
How the CAP Protocol
Is Battle-Tested
Every result published here was produced by running the same 5-gate decision engine — no discretion, no adjustment. If the setup doesn't meet all five conditions, it doesn't count.
Strict Gate Compliance
Only setups where all 5 CAP gates confirmed sequentially were included. Partial setups — even high-conviction ones — are excluded. The protocol is binary: all gates, or no trade.
1% Account Risk Per Trade
Every setup uses 1% of account equity as the risk unit. This mirrors disciplined live-trading position sizing and compounds accurately across the 92-setup run. The three books run concurrently, so up to 3% can be at risk at once.
The Published Backtest, Not A Simulation
The equity curve above plots the 92 qualifying setups from the portfolio monthly table beneath it, month by month. Every value in that table is the sum of three rows published in the per-book tables below — nothing is reconstructed trade by trade. It draws identically every time because it is derived from those tables, drawdown and all. If a book’s table changes, the curve changes with it; they cannot disagree.
Live Session Window Only
Gate 1 restricts valid entries to NY Open, London Open, or their overlap. This time-filter alone eliminates the majority of noise-driven setups from the dataset.
The 5-Gate
CAP Protocol
Every result on this page passed through all five gates in sequence. No gate can be skipped. This is the architecture that produces a documented 73.8% peak win rate on BTC (n=1,084) at the top confluence tier — backtested.
NY Open · London Open · Overlap only. Off-session setups are automatically disqualified regardless of chart structure. Session timing is Gate 1 because liquidity context determines whether institutional order flow is active.
Clean candle close above the most recent swing high. Wicks don't count. The structural shift must be decisive — a body close that commits price above resistance, not a temporary spike.
Fibonacci 0.236–0.382 retracement of the BOS impulse move. Sweet spot: 0.295. Price must retrace into this zone — not above, not below. This is where institutional re-entry accumulates before continuation. How the retracement zone is measured, and what the deep-retrace subset did →
Confluence must confirm the reversal at the OTE zone, and it is scored rather than switched on or off: CVD divergence, Open Interest behaviour, OBV, RSI strength or divergence, order-block respect, FVG mitigation, a liquidity sweep of the stops beyond the zone, an A=C measured move, or a completed fifth of a C leg. No single layer is mandatory — the sweep included. The more that agree, the higher the grade and the larger the authorised size. A liquidity sweep of the stops beneath the zone — Wyckoff's Spring formalised — is one of the heaviest layers when it prints, but it is a grade multiplier, not a precondition.
Candle close back through the reaction high that formed inside the zone — and where a sweep printed, that is the close back above the sweep wick. The Change of Character is the market confirming the confluence read was right. All 5 gates confirmed: execute with pre-defined size, targets, trailing stop. No discretion required.
Results by
Protocol
The CAP Framework runs on four dedicated protocols — each tuned to the specific liquidity and volatility profile of its asset.
BTCUSDT perpetuals. 10-module Foundation or 16-module Masterwork structure. 73.8% peak WR (n=1,084) at the top confluence tier; 65.8% across the 2,009-trade book. Every gate is objectively defined and reproducible on chart.
BTC Masterwork →ETHUSDT perpetuals. 70.9% peak WR (n=574) at the top confluence tier; 68.3% across the 1,618-trade book, net +0.321R per trade — the strongest book of the four. 12-module Foundation or 17-module Masterwork.
ETH Masterwork →SOLUSDT perpetuals. ATR-anchored stops calibrated for SOL’s higher volatility. 67.4% book win rate on n=724 · 76.1% on the peak tier (n=247). Profit factor 1.75 · net expectancy +0.297R per trade · roughly 277 trades a year · 63.6% across all 7,543 zone touches on every timeframe.
SOL Protocol →The CAP Framework applied to Gold. XAUUSD spot, Dukascopy hourly data, 2020–2026 — the longest history of the four books. 64.7% book win rate on n=1,796 · 72.4% on the peak tier (n=181) · net +0.315R per trade. The same protocol translated to commodity volatility, and the deepest history of the four books. Gold’s table is now published in full below — the 72% previously shown here was never supported by one and has been retired.
Gold Protocol →No Cherry-Picked Reviews.
Just Things You Can Check.
The CAP Framework is a new release, and we would rather show you what is independently verifiable than fill this page with testimonials you have no way to confirm. Everything below you can check for yourself — that is the entire point of a rule-based system.
Every result on this page comes from a fully rule-based protocol. The five gates are objectively defined — session window, Break of Structure, OTE zone, confluence (the sweep plus its confirmation), CHoCH. Pull up BTCUSDT on TradingView or Bitget and apply the same gates to historical price, and the same setups print. You do not have to take our word for the win rate; you can re-derive it.
The complete four-book walk-forward backtest is published below this section — every asset, every monthly result, maximum drawdown, Sharpe and Sortino ratios, and confidence intervals. We show the losing months, not just the winning ones, because a results page that only shows wins is not a results page. The drawdown in the curve above is real and intentional.
If the methodology is not what this page describes, you are covered by the refund window on every purchase — so the risk of finding out is on us, not you. And as a founding member, your honest review (good or critical) goes straight to a public review profile as real buyers come on board. We are building proof the slow, real way, in public.
Already running the protocol? Your verified review helps the next trader decide — share it here and it goes on the public profile.
Mapped On The 16th.
Filled At The Line.
Every table on this page is a backtest — setups measured after the event, with the outcome already sitting in the file. This is the other kind of evidence, and it is worth exactly one setup: a chart drawn on 16 August 2026, before anything had happened, and the same chart eleven days later. Both carry TradingView’s own export timestamp in the top-left corner. Tap or click either one to open it full size and read the price axis for yourself.
- It is one setup. n = 1. A single sequence has no win rate, no expectancy and no confidence interval, and it is not evidence of an edge. The sample that argues for the edge is the 92-setup walk-forward test further down this page — losing months included.
- We are not turning this into a percentage. The entries were laddered across a range and position sizes are not published, so any return figure printed here would be one you could not check. The two prices are on the charts; the arithmetic is yours to do.
- Two exchanges, deliberately unmatched. The map is Binance spot, the result is Bybit spot — same asset, same 4H timeframe, quotes a few dollars apart. The first chart was not re-drawn to agree with the second.
- Nothing was added to the images. They are the exported screenshots, cropped only where the two were stacked into one file. The export timestamps are TradingView’s own.
One documented example is an illustration of the process, not a result you should expect to repeat. Trading perpetual futures involves substantial risk of loss. Past performance — including documented win rates — does not guarantee future results.
The Three Pillars
Behind the Numbers
The CAP Framework synthesises three independently validated methodologies. Understanding why each one is included — and what it eliminates — is why the protocol produces a consistent edge rather than random outcomes.
Wyckoff Method
Market regime classification. Accumulation vs Distribution phase identification. The Spring and UTAD patterns behind the liquidity-sweep layer of Gate 4's confluence stack. Wyckoff gives CAP its structural context.
Read the full guide →Elliott Wave
Impulse vs corrective sequence identification. Wave 2 and Wave 4 retracements map directly to the OTE Fibonacci zone in Gate 3. Elliott Wave gives CAP its structural entry timing within a broader sequence.
Read the full guide →Order Flow · CVD
Cumulative Volume Delta divergence, Open Interest analysis, and ETF flow regime gating. CVD divergence at Gate 4 is the institutional confirmation layer — separating genuine absorption from false rejections.
Read the full guide →Which Timeframe
Is This For?
Every table below this section is stamped 1H, and that stamp is doing honest work — it says where the numbers were measured, not where the protocol is allowed to operate. Those are two different claims, and conflating them would misrepresent the framework in both directions. So here is the distinction, drawn plainly.
The gates are structural, not temporal
A Break of Structure is a body close through a prior swing. An OTE zone is a 0.236–0.382 retracement of that impulse. A CHoCH is a close back through the reaction high. None of those definitions contains a candle duration. They describe the shape of price, and price makes that shape on the 15-minute chart, the 4-hour and the daily alike. Nothing in the CAP logic is welded to a 60-minute bar.
It already reads more than one timeframe
CAP is not a single-timeframe system even as published. The confluence stack scores daily trend, a 4-hour momentum proxy and 1-hour structure together — higher-timeframe context gating a lower-timeframe entry. The 1H stamp names the execution chart, the one the entry bar closes on. The context has always come from above it.
The statistics, however, are 1H
A win rate measured on 1H is a claim about 1H. It does not transfer to 15m or D1 by assumption, and we will not imply that it does — n=6,147 is the 1-hour execution sample and nothing else. Trade the protocol on another timeframe by all means; just do not carry these numbers across with you. Build your own sample there before you size up.
Which way we would lean, and why
Higher. A daily or 4-hour BOS survives noise that shakes a 15-minute one apart, the runs that follow tend to last longer, and the spread and funding you pay stop being a meaningful fraction of the move. You get fewer setups of better quality — which is less screen time, not more. Drop lower and you gain sample speed, but you buy it with noise, cost and decisions per day.
The short version: the protocol travels, the statistics do not. Learn it on the 1-hour chart, where you can check every number on this page against your own screen — then take the same five gates up to the 4-hour or the daily, where the signal is cleaner and the runs are longer, and let your own log earn the numbers. And note what a longer run actually gives you: more of the move left to hold. Finding the entry mechanically is what the framework does — holding the back half of a large trend and scaling out of it is a different skill entirely, and it is the one that coaching exists to teach.
The Full Backtest
For The Curious
Headline numbers are easy to claim. So the protocol was re-derived from scratch across four markets and six and a half years — Bitcoin, Ethereum, Solana and Gold — on public exchange data, with no look-ahead, no parameter refit, and bootstrap confidence intervals on every figure. 46,036 zone touches and the 6,147-trade book drawn from them are published below in full: every timeframe, every tier, every calendar year, every caveat, and the places where the result went against what we expected — including one where it went against the product. If you want to see how this was actually built and where its limits are, this is where you live.
2,009 trades over 2021–2026 years, one position at a time, net of commission. Deepest drawdown 15.51R; longest losing run 8 trades. Across every zone this book touched on every timeframe — 63.8% on n=13,650. Its best-defined tier, three or more confluence layers on a zone under 0.6% of price, runs 73.8% on n=1,084.
1,618 trades over 2021–2026 years, one position at a time, net of commission. Deepest drawdown 14.70R; longest losing run 6 trades. Across every zone this book touched on every timeframe — 64.4% on n=13,919. Its best-defined tier, three or more confluence layers on a zone under 0.6% of price, runs 70.9% on n=574.
724 trades over 2024–2026 years, one position at a time, net of commission. Deepest drawdown 12.94R; longest losing run 7 trades. Across every zone this book touched on every timeframe — 63.6% on n=7,543. Its best-defined tier, three or more confluence layers on a zone under 0.6% of price, runs 76.1% on n=247.
1,796 trades over 2020–2026 years, one position at a time, net of commission. Deepest drawdown 7.69R; longest losing run 7 trades. Across every zone this book touched on every timeframe — 60.8% on n=10,924. Its best-defined tier, three or more confluence layers on a zone under 0.6% of price, runs 72.4% on n=181.
01 · Methodology — Zone, Entry, Stop and Exit▾
Every setup in this book is generated the same way, in this order:
1. The zone is drawn before price arrives. Either an order block — the last opposing candle before a displacement of at least 1.2×ATR(14) that closes beyond the prior 10-bar swing — or a fair value gap, a three-candle imbalance of at least 0.25×ATR. The zone becomes usable only once the bar that created it has closed.
2. A limit rests at the zone midpoint. Nothing is chased. If price never returns, there is no trade. If price passes through the far side of the zone before touching the midpoint, the zone is dead and no fill is taken — invalidation is checked before the fill on every single bar.
3. The stop sits 0.3×ATR beyond the far edge of the zone. 1R is that distance. On this book the median 1R is 0.38% of price.
4. Exits are fixed. Half the position closes at 1R and the stop moves to break-even; the remainder targets 4R. A position that reaches neither within 60 bars of the zone’s own timeframe is closed at the market.
5. One position at a time. A new setup is only taken once the previous one has closed. Without that rule the sample is a population of overlapping zone touches, not a book anyone could run.
Costs. 0.11% (taker) round trip, charged in R against each trade’s own stop distance — so a tight zone pays proportionally more. A “win” here means the trade finished above break-even after that cost, not that it touched a target.
02 · The Book — One Position At A Time▾
| Statistic | Value |
|---|---|
| Trades | 2,009 |
| Period | 2021-01-01 → 2026-08-28 |
| Trades per year | 355 |
| Win rate | 65.8% |
| 95% CI on win rate | [63.7% , 67.8%] |
| Net expectancy | +0.250R |
| 95% CI on expectancy | [+0.193R , +0.307R] |
| Total | +502.9R |
| Profit factor | 1.62 |
| Average win | +0.991R |
| Average loss | −1.173R |
| Max drawdown | 15.51R |
| Longest losing run | 8 trades |
| Median 1R | 0.38% of price |
What a win actually pays. 26.1% of these trades reached both targets, 43.8% banked the partial at 1R and then closed at break-even, and 30.2% took the full stop. That mix — not the win rate on its own — is what +0.250R per trade looks like.
03 · Every Zone Touched — And Every Timeframe▾
Restricting to one position at a time answers “what could a person run?”. Dropping that rule and taking every qualifying zone on every timeframe answers a different question — how much edge is in the structure itself. These overlap heavily, so they are not a track record; they are the evidence base the book is drawn from.
Every timeframe, published:
| Timeframe | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 1H | 4,332 | 65.4% | +0.219R | +946.9R |
| 2H | 2,673 | 64.7% | +0.185R | +493.4R |
| 3H | 1,954 | 63.6% | +0.167R | +327.3R |
| 4H | 1,539 | 65.0% | +0.196R | +300.9R |
| 6H | 1,098 | 63.8% | +0.177R | +194.7R |
| 8H | 871 | 64.3% | +0.204R | +178.1R |
| 12H | 606 | 59.1% | +0.098R | +59.2R |
| 1D | 322 | 50.6% | −0.031R | −9.8R |
The edge does not live on one timeframe. What changes with the timeframe is how many setups you get, how much screen time they cost, and how much of each one commission takes — 1R is a fixed percentage of price at every step, so a shorter timeframe pays a larger share of its gross away.
04 · Zone Precision — And Where Tighter Stops Stop Helping▾
How tightly the zone is drawn is the single most useful thing to grade a setup on. A thin zone is a precise level; a thick one is a smear, and the entry inside it is a guess.
| Zone precision | Setups | Win rate | Net EV |
|---|---|---|---|
| Every zone | 13,650 | 63.8% | +0.182R |
| Zone under 1.2% of price | 9,029 | 66.7% | +0.220R |
| Zone under 0.6% of price | 4,610 | 66.9% | +0.246R |
| Zone under 0.3% of price | 1,459 | 59.8% | +0.173R |
Tighter is better — but only up to a point, and the reason is commission. 1R is the distance to the stop, so a tighter zone means a smaller 1R, and a fixed percentage cost eats a larger share of it. Under about 0.6% of price the cost drag starts taking back more than the precision adds. That is why 0.6% is the tier this site publishes and the one the book above is built on — not the tightest tier available.
05 · Confluence — The Five Layers That Measurably Stack▾
Every item on the Masterwork checklist was scored mechanically against all 13,650 zone touches. Five of them carry a lift that survives its own confidence interval:
- a trending 4-hour regime — ADX above 22 with intact structure
- a corrective retracement into the zone — overlapping, small-bodied candles, not an aggressive counter-impulse
- open interest falling as price retraces into the zone
- a 1-hour RSI divergence at the zone
- a change of character also printing inside the zone — the reactive entry confirming the resting one
They stack, and they stack monotonically:
| Layers firing | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| 0 layers | 138 | 58.7% | [50.7% , 66.7%] | +0.106R |
| 1 layer | 3,620 | 58.0% | [56.4% , 59.6%] | +0.047R |
| 2 layers | 6,476 | 63.5% | [62.4% , 64.7%] | +0.189R |
| 3 layers | 3,218 | 70.5% | [68.9% , 72.0%] | +0.301R |
| 4 layers | 198 | 74.7% | [68.7% , 80.8%] | +0.551R |
The peak tier — three or more layers on a zone under 0.6% of price — 73.8% on n=1,084, 95% CI 71.2%–76.4%, +0.362R net, and 26.7% of them reach both targets. None of these layers is required. The base is positive without any of them, which is the point: the criteria are more lenient than the gate stack demands, and the layers are what move a setup from acceptable to the best available.
06 · The Grade Ladder, Measured — And What It Does Not Do▾
The Masterwork grades every setup S through F from a weighted 51-item checklist. Those grade win rates have always been labelled estimated. We scored all 13,650 setups against the checklist mechanically — 40 of the items are computable from price, volume and order-flow data; the rest (macro calendars, ETF flow, on-chain, tick-level footprint) are reported as uncovered rather than guessed, so each setup is scored out of the 345 points that could actually be answered.
| Grade | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| B | 267 | 59.9% | [54.3% , 65.5%] | +0.077R |
| C | 2,399 | 64.4% | [62.4% , 66.3%] | +0.217R |
| D | 9,413 | 63.4% | [62.4% , 64.4%] | +0.172R |
| F | 1,568 | 66.1% | [63.8% , 68.5%] | +0.212R |
It does not discriminate, and we are publishing that rather than burying it. A higher checklist score did not produce a higher win rate on this book, and the same is true on the other three. The five layers in the previous section do. The honest reading is that the 51-item score is a discipline and review instrument — it stops you taking setups you have not checked — and not a probability estimate. It is being repositioned on that basis rather than given a measured number it has not earned.
07 · Year By Year — And What Does Not Separate▾
| Year | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 2021 | 2,908 | 63.9% | +0.195R | +566.6R |
| 2022 | 2,272 | 65.2% | +0.219R | +497.1R |
| 2023 | 1,812 | 65.9% | +0.203R | +368.7R |
| 2024 | 2,323 | 64.7% | +0.231R | +537.1R |
| 2025 | 2,724 | 59.0% | +0.068R | +184.1R |
| 2026 | 1,611 | 66.4% | +0.208R | +335.3R |
No year was excluded and no parameter was changed between them. Order blocks and fair value gaps are close to indistinguishable on this book — 63.1% (n=8,233) against 64.9% (n=5,417). Longs beat shorts, 64.6% against 63.0%; the whole sample sits in a period that was net bullish, so read that gap as regime, not edge.
08 · Data Provenance — Where These Numbers Come From▾
| Item | Value |
|---|---|
| Data feed | Binance public archive — BTCUSDT USD-M perpetual, 1-hour klines |
| Period | 2021-01-01 → 2026-08-28 |
| Zone timeframes | 1H, 2H, 3H, 4H, 6H, 8H, 12H, 1D, 2D, 3D |
| Execution timeframe | 1H |
| Order-flow inputs | Perpetual taker split, spot klines, funding rate history, 5-minute open-interest metrics |
| Venue note | Binance, not Bybit. Bybit’s archive is not reachable from the machine this was run on. The venue is stated rather than silently substituted. |
| Look-ahead | None. A zone is unusable until its bar closes; invalidation is checked before the fill on every bar; every indicator is computed causally. |
| Parameter refit | None. Every threshold was fixed before the run and is unchanged across all four books. |
| Random seed | 42 — used only for bootstrap confidence intervals |
| Costs | 0.11% round trip (taker both sides) |
| Survivorship | None. Every zone the detector produced in the period is in the sample, winners and losers alike. |
01 · Methodology — Zone, Entry, Stop and Exit▾
Every setup in this book is generated the same way, in this order:
1. The zone is drawn before price arrives. Either an order block — the last opposing candle before a displacement of at least 1.2×ATR(14) that closes beyond the prior 10-bar swing — or a fair value gap, a three-candle imbalance of at least 0.25×ATR. The zone becomes usable only once the bar that created it has closed.
2. A limit rests at the zone midpoint. Nothing is chased. If price never returns, there is no trade. If price passes through the far side of the zone before touching the midpoint, the zone is dead and no fill is taken — invalidation is checked before the fill on every single bar.
3. The stop sits 0.3×ATR beyond the far edge of the zone. 1R is that distance. On this book the median 1R is 0.46% of price.
4. Exits are fixed. Half the position closes at 1R and the stop moves to break-even; the remainder targets 4R. A position that reaches neither within 60 bars of the zone’s own timeframe is closed at the market.
5. One position at a time. A new setup is only taken once the previous one has closed. Without that rule the sample is a population of overlapping zone touches, not a book anyone could run.
Costs. 0.11% (taker) round trip, charged in R against each trade’s own stop distance — so a tight zone pays proportionally more. A “win” here means the trade finished above break-even after that cost, not that it touched a target.
02 · The Book — One Position At A Time▾
| Statistic | Value |
|---|---|
| Trades | 1,618 |
| Period | 2021-01-01 → 2026-08-28 |
| Trades per year | 286 |
| Win rate | 68.3% |
| 95% CI on win rate | [66.1% , 70.6%] |
| Net expectancy | +0.321R |
| 95% CI on expectancy | [+0.259R , +0.385R] |
| Total | +519.5R |
| Profit factor | 1.86 |
| Average win | +1.014R |
| Average loss | −1.172R |
| Max drawdown | 14.70R |
| Longest losing run | 6 trades |
| Median 1R | 0.46% of price |
What a win actually pays. 26.5% of these trades reached both targets, 44.2% banked the partial at 1R and then closed at break-even, and 29.4% took the full stop. That mix — not the win rate on its own — is what +0.321R per trade looks like.
03 · Every Zone Touched — And Every Timeframe▾
Restricting to one position at a time answers “what could a person run?”. Dropping that rule and taking every qualifying zone on every timeframe answers a different question — how much edge is in the structure itself. These overlap heavily, so they are not a track record; they are the evidence base the book is drawn from.
Every timeframe, published:
| Timeframe | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 1H | 4,504 | 65.5% | +0.241R | +1085.5R |
| 2H | 2,686 | 65.9% | +0.250R | +671.5R |
| 3H | 1,931 | 64.7% | +0.202R | +390.2R |
| 4H | 1,556 | 64.3% | +0.210R | +326.7R |
| 6H | 1,158 | 61.8% | +0.179R | +207.0R |
| 8H | 900 | 62.7% | +0.209R | +188.2R |
| 12H | 617 | 64.0% | +0.225R | +138.5R |
| 1D | 317 | 55.5% | +0.063R | +19.8R |
The edge does not live on one timeframe. What changes with the timeframe is how many setups you get, how much screen time they cost, and how much of each one commission takes — 1R is a fixed percentage of price at every step, so a shorter timeframe pays a larger share of its gross away.
04 · Zone Precision — And Where Tighter Stops Stop Helping▾
How tightly the zone is drawn is the single most useful thing to grade a setup on. A thin zone is a precise level; a thick one is a smear, and the entry inside it is a guess.
| Zone precision | Setups | Win rate | Net EV |
|---|---|---|---|
| Every zone | 13,919 | 64.4% | +0.221R |
| Zone under 1.2% of price | 7,263 | 67.3% | +0.279R |
| Zone under 0.6% of price | 3,104 | 69.9% | +0.339R |
| Zone under 0.3% of price | 718 | 66.3% | +0.310R |
Tighter is better — but only up to a point, and the reason is commission. 1R is the distance to the stop, so a tighter zone means a smaller 1R, and a fixed percentage cost eats a larger share of it. Under about 0.6% of price the cost drag starts taking back more than the precision adds. That is why 0.6% is the tier this site publishes and the one the book above is built on — not the tightest tier available.
05 · Confluence — The Five Layers That Measurably Stack▾
Every item on the Masterwork checklist was scored mechanically against all 13,919 zone touches. Five of them carry a lift that survives its own confidence interval:
- a trending 4-hour regime — ADX above 22 with intact structure
- a corrective retracement into the zone — overlapping, small-bodied candles, not an aggressive counter-impulse
- open interest falling as price retraces into the zone
- a 1-hour RSI divergence at the zone
- a change of character also printing inside the zone — the reactive entry confirming the resting one
They stack, and they stack monotonically:
| Layers firing | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| 0 layers | 164 | 54.9% | [47.0% , 62.2%] | +0.039R |
| 1 layer | 4,590 | 62.0% | [60.7% , 63.4%] | +0.174R |
| 2 layers | 6,443 | 64.8% | [63.6% , 65.9%] | +0.225R |
| 3 layers | 2,539 | 68.1% | [66.3% , 69.9%] | +0.306R |
| 4 layers | 183 | 68.3% | [61.7% , 74.9%] | +0.247R |
The peak tier — three or more layers on a zone under 0.6% of price — 70.9% on n=574, 95% CI 67.2%–74.6%, +0.442R net, and 31.0% of them reach both targets. None of these layers is required. The base is positive without any of them, which is the point: the criteria are more lenient than the gate stack demands, and the layers are what move a setup from acceptable to the best available.
06 · The Grade Ladder, Measured — And What It Does Not Do▾
The Masterwork grades every setup S through F from a weighted 51-item checklist. Those grade win rates have always been labelled estimated. We scored all 13,919 setups against the checklist mechanically — 40 of the items are computable from price, volume and order-flow data; the rest (macro calendars, ETF flow, on-chain, tick-level footprint) are reported as uncovered rather than guessed, so each setup is scored out of the 387 points that could actually be answered.
| Grade | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| B | 722 | 66.8% | [63.3% , 70.1%] | +0.254R |
| C | 4,971 | 64.5% | [63.2% , 65.8%] | +0.228R |
| D | 8,075 | 64.0% | [62.9% , 65.0%] | +0.209R |
| F | 151 | 72.2% | [64.9% , 78.8%] | +0.545R |
It does not discriminate, and we are publishing that rather than burying it. A higher checklist score did not produce a higher win rate on this book, and the same is true on the other three. The five layers in the previous section do. The honest reading is that the 51-item score is a discipline and review instrument — it stops you taking setups you have not checked — and not a probability estimate. It is being repositioned on that basis rather than given a measured number it has not earned.
07 · Year By Year — And What Does Not Separate▾
| Year | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 2021 | 2,903 | 60.2% | +0.189R | +549.3R |
| 2022 | 2,609 | 64.5% | +0.227R | +591.0R |
| 2023 | 1,974 | 66.5% | +0.225R | +445.1R |
| 2024 | 2,464 | 66.2% | +0.267R | +656.9R |
| 2025 | 2,501 | 64.6% | +0.220R | +549.8R |
| 2026 | 1,468 | 66.5% | +0.197R | +288.7R |
No year was excluded and no parameter was changed between them. Order blocks and fair value gaps are close to indistinguishable on this book — 64.7% (n=8,167) against 64.1% (n=5,752). Longs beat shorts, 66.4% against 62.3%; the whole sample sits in a period that was net bullish, so read that gap as regime, not edge.
08 · Data Provenance — Where These Numbers Come From▾
| Item | Value |
|---|---|
| Data feed | Binance public archive — ETHUSDT USD-M perpetual, 1-hour klines |
| Period | 2021-01-01 → 2026-08-28 |
| Zone timeframes | 1H, 2H, 3H, 4H, 6H, 8H, 12H, 1D, 2D, 3D |
| Execution timeframe | 1H |
| Order-flow inputs | Perpetual taker split, spot klines, funding rate history, 5-minute open-interest metrics |
| Venue note | Binance, not Bybit. Bybit’s archive is not reachable from the machine this was run on. The venue is stated rather than silently substituted. |
| Look-ahead | None. A zone is unusable until its bar closes; invalidation is checked before the fill on every bar; every indicator is computed causally. |
| Parameter refit | None. Every threshold was fixed before the run and is unchanged across all four books. |
| Random seed | 42 — used only for bootstrap confidence intervals |
| Costs | 0.11% round trip (taker both sides) |
| Survivorship | None. Every zone the detector produced in the period is in the sample, winners and losers alike. |
01 · Methodology — Zone, Entry, Stop and Exit▾
Every setup in this book is generated the same way, in this order:
1. The zone is drawn before price arrives. Either an order block — the last opposing candle before a displacement of at least 1.2×ATR(14) that closes beyond the prior 10-bar swing — or a fair value gap, a three-candle imbalance of at least 0.25×ATR. The zone becomes usable only once the bar that created it has closed.
2. A limit rests at the zone midpoint. Nothing is chased. If price never returns, there is no trade. If price passes through the far side of the zone before touching the midpoint, the zone is dead and no fill is taken — invalidation is checked before the fill on every single bar.
3. The stop sits 0.3×ATR beyond the far edge of the zone. 1R is that distance. On this book the median 1R is 0.51% of price.
4. Exits are fixed. Half the position closes at 1R and the stop moves to break-even; the remainder targets 4R. A position that reaches neither within 60 bars of the zone’s own timeframe is closed at the market.
5. One position at a time. A new setup is only taken once the previous one has closed. Without that rule the sample is a population of overlapping zone touches, not a book anyone could run.
Costs. 0.11% (taker) round trip, charged in R against each trade’s own stop distance — so a tight zone pays proportionally more. A “win” here means the trade finished above break-even after that cost, not that it touched a target.
02 · The Book — One Position At A Time▾
| Statistic | Value |
|---|---|
| Trades | 724 |
| Period | 2024-01-01 → 2026-08-28 |
| Trades per year | 277 |
| Win rate | 67.4% |
| 95% CI on win rate | [64.1% , 70.7%] |
| Net expectancy | +0.297R |
| 95% CI on expectancy | [+0.201R , +0.394R] |
| Total | +215.1R |
| Profit factor | 1.75 |
| Average win | +1.031R |
| Average loss | −1.220R |
| Max drawdown | 12.94R |
| Longest losing run | 7 trades |
| Median 1R | 0.51% of price |
What a win actually pays. 25.6% of these trades reached both targets, 42.4% banked the partial at 1R and then closed at break-even, and 32.0% took the full stop. That mix — not the win rate on its own — is what +0.297R per trade looks like.
03 · Every Zone Touched — And Every Timeframe▾
Restricting to one position at a time answers “what could a person run?”. Dropping that rule and taking every qualifying zone on every timeframe answers a different question — how much edge is in the structure itself. These overlap heavily, so they are not a track record; they are the evidence base the book is drawn from.
Every timeframe, published:
| Timeframe | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 1H | 2,436 | 65.0% | +0.253R | +616.0R |
| 2H | 1,513 | 65.5% | +0.276R | +418.0R |
| 3H | 1,061 | 64.2% | +0.244R | +258.6R |
| 4H | 867 | 61.5% | +0.160R | +138.6R |
| 6H | 612 | 59.0% | +0.083R | +50.7R |
| 8H | 446 | 63.5% | +0.197R | +87.7R |
| 12H | 325 | 60.3% | +0.079R | +25.6R |
| 1D | 155 | 64.5% | +0.182R | +28.2R |
The edge does not live on one timeframe. What changes with the timeframe is how many setups you get, how much screen time they cost, and how much of each one commission takes — 1R is a fixed percentage of price at every step, so a shorter timeframe pays a larger share of its gross away.
04 · Zone Precision — And Where Tighter Stops Stop Helping▾
How tightly the zone is drawn is the single most useful thing to grade a setup on. A thin zone is a precise level; a thick one is a smear, and the entry inside it is a guess.
| Zone precision | Setups | Win rate | Net EV |
|---|---|---|---|
| Every zone | 7,543 | 63.6% | +0.216R |
| Zone under 1.2% of price | 3,616 | 67.3% | +0.297R |
| Zone under 0.6% of price | 1,302 | 68.7% | +0.329R |
| Zone under 0.3% of price | 223 | 65.5% | +0.325R |
Tighter is better — but only up to a point, and the reason is commission. 1R is the distance to the stop, so a tighter zone means a smaller 1R, and a fixed percentage cost eats a larger share of it. Under about 0.6% of price the cost drag starts taking back more than the precision adds. That is why 0.6% is the tier this site publishes and the one the book above is built on — not the tightest tier available.
05 · Confluence — The Five Layers That Measurably Stack▾
Every item on the Masterwork checklist was scored mechanically against all 7,543 zone touches. Five of them carry a lift that survives its own confidence interval:
- a trending 4-hour regime — ADX above 22 with intact structure
- a corrective retracement into the zone — overlapping, small-bodied candles, not an aggressive counter-impulse
- open interest falling as price retraces into the zone
- a 1-hour RSI divergence at the zone
- a change of character also printing inside the zone — the reactive entry confirming the resting one
They stack, and they stack monotonically:
| Layers firing | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| 0 layers | 93 | 59.1% | [49.5% , 68.8%] | +0.183R |
| 1 layer | 2,214 | 60.8% | [58.7% , 62.8%] | +0.153R |
| 2 layers | 3,514 | 64.3% | [62.7% , 65.8%] | +0.208R |
| 3 layers | 1,605 | 65.9% | [63.6% , 68.2%] | +0.307R |
| 4 layers | 117 | 70.1% | [61.5% , 77.8%] | +0.439R |
The peak tier — three or more layers on a zone under 0.6% of price — 76.1% on n=247, 95% CI 70.9%–81.4%, +0.489R net, and 28.3% of them reach both targets. None of these layers is required. The base is positive without any of them, which is the point: the criteria are more lenient than the gate stack demands, and the layers are what move a setup from acceptable to the best available.
06 · The Grade Ladder, Measured — And What It Does Not Do▾
The Masterwork grades every setup S through F from a weighted 51-item checklist. Those grade win rates have always been labelled estimated. We scored all 7,543 setups against the checklist mechanically — 40 of the items are computable from price, volume and order-flow data; the rest (macro calendars, ETF flow, on-chain, tick-level footprint) are reported as uncovered rather than guessed, so each setup is scored out of the 332 points that could actually be answered.
| Grade | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| B | 78 | 71.8% | [61.5% , 80.8%] | +0.308R |
| C | 1,083 | 64.3% | [61.4% , 67.0%] | +0.306R |
| D | 4,956 | 63.0% | [61.6% , 64.3%] | +0.178R |
| F | 1,426 | 64.9% | [62.5% , 67.4%] | +0.276R |
It does not discriminate, and we are publishing that rather than burying it. A higher checklist score did not produce a higher win rate on this book, and the same is true on the other three. The five layers in the previous section do. The honest reading is that the 51-item score is a discipline and review instrument — it stops you taking setups you have not checked — and not a probability estimate. It is being repositioned on that basis rather than given a measured number it has not earned.
07 · Year By Year — And What Does Not Separate▾
| Year | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 2024 | 2,803 | 64.1% | +0.235R | +658.3R |
| 2025 | 3,066 | 63.0% | +0.215R | +657.8R |
| 2026 | 1,674 | 64.0% | +0.189R | +316.2R |
No year was excluded and no parameter was changed between them. Order blocks and fair value gaps are close to indistinguishable on this book — 62.6% (n=4,631) against 65.2% (n=2,912). Longs beat shorts, 65.7% against 61.4%; the whole sample sits in a period that was net bullish, so read that gap as regime, not edge.
08 · Data Provenance — Where These Numbers Come From▾
| Item | Value |
|---|---|
| Data feed | Binance public archive — SOLUSDT USD-M perpetual, 1-hour klines |
| Period | 2024-01-01 → 2026-08-28 |
| Zone timeframes | 1H, 2H, 3H, 4H, 6H, 8H, 12H, 1D, 2D, 3D |
| Execution timeframe | 1H |
| Order-flow inputs | Perpetual taker split, spot klines, funding rate history, 5-minute open-interest metrics |
| Venue note | Binance, not Bybit. Bybit’s archive is not reachable from the machine this was run on. The venue is stated rather than silently substituted. |
| Look-ahead | None. A zone is unusable until its bar closes; invalidation is checked before the fill on every bar; every indicator is computed causally. |
| Parameter refit | None. Every threshold was fixed before the run and is unchanged across all four books. |
| Random seed | 42 — used only for bootstrap confidence intervals |
| Costs | 0.11% round trip (taker both sides) |
| Survivorship | None. Every zone the detector produced in the period is in the sample, winners and losers alike. |
01 · Methodology — Zone, Entry, Stop and Exit▾
Every setup in this book is generated the same way, in this order:
1. The zone is drawn before price arrives. Either an order block — the last opposing candle before a displacement of at least 1.2×ATR(14) that closes beyond the prior 10-bar swing — or a fair value gap, a three-candle imbalance of at least 0.25×ATR. The zone becomes usable only once the bar that created it has closed.
2. A limit rests at the zone midpoint. Nothing is chased. If price never returns, there is no trade. If price passes through the far side of the zone before touching the midpoint, the zone is dead and no fill is taken — invalidation is checked before the fill on every single bar.
3. The stop sits 0.3×ATR beyond the far edge of the zone. 1R is that distance. On this book the median 1R is 0.22% of price.
4. Exits are fixed. Half the position closes at 1R and the stop moves to break-even; the remainder targets 4R. A position that reaches neither within 60 bars of the zone’s own timeframe is closed at the market.
5. One position at a time. A new setup is only taken once the previous one has closed. Without that rule the sample is a population of overlapping zone touches, not a book anyone could run.
Costs. 0.02% (spot spread) round trip, charged in R against each trade’s own stop distance — so a tight zone pays proportionally more. A “win” here means the trade finished above break-even after that cost, not that it touched a target.
02 · The Book — One Position At A Time▾
| Statistic | Value |
|---|---|
| Trades | 1,796 |
| Period | 2020-01-02 → 2026-08-27 |
| Trades per year | 271 |
| Win rate | 64.7% |
| 95% CI on win rate | [62.5% , 66.9%] |
| Net expectancy | +0.315R |
| 95% CI on expectancy | [+0.256R , +0.374R] |
| Total | +566.3R |
| Profit factor | 1.82 |
| Average win | +1.083R |
| Average loss | −1.091R |
| Max drawdown | 7.69R |
| Longest losing run | 7 trades |
| Median 1R | 0.22% of price |
What a win actually pays. 21.8% of these trades reached both targets, 42.9% banked the partial at 1R and then closed at break-even, and 35.3% took the full stop. That mix — not the win rate on its own — is what +0.315R per trade looks like.
03 · Every Zone Touched — And Every Timeframe▾
Restricting to one position at a time answers “what could a person run?”. Dropping that rule and taking every qualifying zone on every timeframe answers a different question — how much edge is in the structure itself. These overlap heavily, so they are not a track record; they are the evidence base the book is drawn from.
Every timeframe, published:
| Timeframe | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 1H | 2,496 | 64.2% | +0.292R | +729.3R |
| 2H | 1,936 | 62.7% | +0.248R | +479.5R |
| 3H | 1,545 | 61.0% | +0.237R | +365.8R |
| 4H | 1,333 | 62.3% | +0.223R | +297.3R |
| 6H | 1,105 | 57.6% | +0.121R | +133.2R |
| 8H | 934 | 58.4% | +0.130R | +121.0R |
| 12H | 721 | 55.3% | +0.050R | +35.7R |
| 1D | 474 | 55.3% | +0.118R | +55.8R |
The edge does not live on one timeframe. What changes with the timeframe is how many setups you get, how much screen time they cost, and how much of each one commission takes — 1R is a fixed percentage of price at every step, so a shorter timeframe pays a larger share of its gross away.
04 · Zone Precision — And Where Tighter Stops Stop Helping▾
How tightly the zone is drawn is the single most useful thing to grade a setup on. A thin zone is a precise level; a thick one is a smear, and the entry inside it is a guess.
| Zone precision | Setups | Win rate | Net EV |
|---|---|---|---|
| Every zone | 10,924 | 60.8% | +0.209R |
| Zone under 1.2% of price | 10,208 | 61.5% | +0.219R |
| Zone under 0.6% of price | 8,109 | 62.9% | +0.249R |
| Zone under 0.3% of price | 4,110 | 64.5% | +0.288R |
Tighter is better — but only up to a point, and the reason is commission. 1R is the distance to the stop, so a tighter zone means a smaller 1R, and a fixed percentage cost eats a larger share of it. Under about 0.6% of price the cost drag starts taking back more than the precision adds. That is why 0.6% is the tier this site publishes and the one the book above is built on — not the tightest tier available.
05 · Confluence — The Five Layers That Measurably Stack▾
Every item on the Masterwork checklist was scored mechanically against all 10,924 zone touches. Five of them carry a lift that survives its own confidence interval:
- a trending 4-hour regime — ADX above 22 with intact structure
- a corrective retracement into the zone — overlapping, small-bodied candles, not an aggressive counter-impulse
- open interest falling as price retraces into the zone
- a 1-hour RSI divergence at the zone
- a change of character also printing inside the zone — the reactive entry confirming the resting one
They stack, and they stack monotonically:
| Layers firing | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| 0 layers | 572 | 59.6% | [55.8% , 63.6%] | +0.163R |
| 1 layer | 5,836 | 60.1% | [58.9% , 61.4%] | +0.182R |
| 2 layers | 4,195 | 61.7% | [60.2% , 63.1%] | +0.244R |
| 3 layers | 321 | 63.2% | [58.3% , 68.5%] | +0.317R |
The peak tier — three or more layers on a zone under 0.6% of price — 72.4% on n=181, 95% CI 65.7%–79.0%, +0.543R net, and 27.1% of them reach both targets. None of these layers is required. The base is positive without any of them, which is the point: the criteria are more lenient than the gate stack demands, and the layers are what move a setup from acceptable to the best available.
06 · The Grade Ladder, Measured — And What It Does Not Do▾
The Masterwork grades every setup S through F from a weighted 51-item checklist. Those grade win rates have always been labelled estimated. We scored all 10,924 setups against the checklist mechanically — 40 of the items are computable from price, volume and order-flow data; the rest (macro calendars, ETF flow, on-chain, tick-level footprint) are reported as uncovered rather than guessed, so each setup is scored out of the 261 points that could actually be answered.
| Grade | Setups | Win rate | 95% CI | Net EV |
|---|---|---|---|---|
| B | 1,168 | 61.0% | [58.2% , 63.9%] | +0.224R |
| C | 2,768 | 58.7% | [56.9% , 60.5%] | +0.189R |
| D | 5,932 | 61.6% | [60.4% , 62.8%] | +0.217R |
| F | 1,037 | 61.5% | [58.5% , 64.4%] | +0.202R |
It does not discriminate, and we are publishing that rather than burying it. A higher checklist score did not produce a higher win rate on this book, and the same is true on the other three. The five layers in the previous section do. The honest reading is that the 51-item score is a discipline and review instrument — it stops you taking setups you have not checked — and not a probability estimate. It is being repositioned on that basis rather than given a measured number it has not earned.
07 · Year By Year — And What Does Not Separate▾
| Year | Setups | Win rate | Net EV | Total |
|---|---|---|---|---|
| 2020 | 1,596 | 64.3% | +0.261R | +416.9R |
| 2021 | 1,480 | 60.4% | +0.194R | +287.7R |
| 2022 | 1,752 | 59.0% | +0.146R | +256.6R |
| 2023 | 1,377 | 58.5% | +0.097R | +133.5R |
| 2024 | 1,574 | 60.9% | +0.229R | +360.8R |
| 2025 | 1,898 | 60.3% | +0.248R | +471.3R |
| 2026 | 1,247 | 62.3% | +0.283R | +352.3R |
No year was excluded and no parameter was changed between them. Order blocks and fair value gaps are close to indistinguishable on this book — 59.7% (n=6,079) against 62.2% (n=4,845). Longs beat shorts, 62.9% against 58.6%; the whole sample sits in a period that was net bullish, so read that gap as regime, not edge.
08 · Data Provenance — Where These Numbers Come From▾
| Item | Value |
|---|---|
| Data feed | Dukascopy XAUUSD spot, hourly bars |
| Period | 2020-01-02 → 2026-08-27 |
| Zone timeframes | 1H, 2H, 3H, 4H, 6H, 8H, 12H, 1D, 2D, 3D |
| Execution timeframe | 1H |
| Look-ahead | None. A zone is unusable until its bar closes; invalidation is checked before the fill on every bar; every indicator is computed causally. |
| Parameter refit | None. Every threshold was fixed before the run and is unchanged across all four books. |
| Random seed | 42 — used only for bootstrap confidence intervals |
| Costs | 0.02% spot spread |
| Survivorship | None. Every zone the detector produced in the period is in the sample, winners and losers alike. |
All numbers reproducible · Public exchange data · Walk-forward · No refit
Run The 5-Gate Protocol→How to Cite
This Study
Every table, figure and confidence interval on this page may be quoted freely in articles, research and reports with attribution. The study is a walk-forward backtest on public exchange data — it is not a third-party-audited live track record, and any citation should say so. Suggested citation:
Charles V. (2026). Four-Book Walk-Forward Test of the CAP Protocol. The Chart Whisperer. n=6,147 trades, January 2020 – August 2026. https://chartwhisperer.ca/protocol-results
- Study period
- January 2020 – August 2026 (80 months, walk-forward, no refit)
- Sample
- 6,147 trades — 4,076 wins, 2,071 losses — pooled across the BTC, ETH, SOL and Gold books
- Headline result
- 66.3% win rate · +0.293R expectancy per trade · +1,803.8R total · −20.15R maximum portfolio drawdown
- Uncertainty
- 95% confidence interval on the win rate: 65.1% – 67.5% (bootstrap, 4,000 resamples). Quote the interval with the point estimate.
- Exits
- Rule set only — TP ladder or the 30-bar timer. No discretionary exits are modelled. See the full backtest.
- Data source
- Binance public archive (BTC, ETH, SOL perpetuals) and Dukascopy XAUUSD spot, hourly candles. Reproducible from the tables on this page.
- Licence
- Quote and reference freely with attribution. Do not present the figures as a live or audited track record.
Citing a specific table? Every section on this page has its own anchor — for example #results-by-book, #monthly-equity or #expectancy — so you can link the exact table you are quoting. Found an error? Tell us via the contact page and it will be re-checked against the source data.
Questions About
The Data
Ready to trade with
this exact protocol?
Every result on this page was generated by the same 5-Gate system available in the BTC, ETH, SOL, and Gold protocols. One methodology. Four assets. Pick your level — Foundation or Masterwork — and every additional cap you add is exactly half price, forever.
SIMULATED PERFORMANCE · PAST RESULTS DO NOT GUARANTEE FUTURE RETURNS · FULL DISCLOSURE →