How we test, and how we bury.

01Rules are written before the data is opened

Before running anything we write down the entry and exit rules, every variant we plan to test, the instruments and periods, and the pass/fail gates (section 04). That specification is saved with a date and published with the post.

• No parameter tweaking after seeing results. If we change a rule, it's a new variant, it's reported, and it counts toward the total tried.
• Variants are declared up front. We don't add "just one more" after a disappointing run.
• Deviations are flagged. If we ever break a rule (for example by opening a holdout early), the post says so in plain words.

02Data and the three periods

Each test splits history into three windows. Which dates apply to each strategy is stated in its post.

• Instruments. Futures (the E-mini S&P 500, ES, the E-mini Nasdaq-100, NQ, and gold futures, GC, in one test), spot gold and the four major FX pairs from a broker's MT4 demo history, and Bitcoin from Binance's public history. Micro contracts are used only to size results for small accounts. Each post names its exact instrument, timeframe and data source.
• Futures, spot and CFD data are not the same thing. Where a test uses spot or CFD price history (for example gold or FX from a broker), the post says so and does not call it a futures result.
• Data quality. We note gaps, session definitions and any bars removed. Results that change when a data gap is excluded are called out. From 8 Oct 2026 each test lists the known holes in its 1-minute data, reports how many trades were open across a hole and recomputes the gates and the verdict without them; if the verdict changes, the result is Inconclusive. Continuous futures series are back-adjusted: results in R and dollars are right, but older prices are not the prices that traded, so percentage changes on them are not used.
• Raw price data is not republished (vendor licences). Trade logs and derived results are.

03Costs are not optional

Every number on this site is net of costs unless it's explicitly labelled gross.

• Futures (ES, NQ, GC): round-turn commission plus at least 1 tick of slippage per side.
• Gold: a fixed cost per ounce for spread and commission ($0.30 a round trip), with no slippage, which is optimistic; stated in the post.
• FX majors: a fixed cost in pips for spread and commission, stated in the post.
• Bitcoin: a percentage of the price on every fill, stated in the post.
• Fills at the next available price. Entries fill at the next 1-minute open after the signal, not at the price that "should have" filled. Futures add a tick of slippage on top; spot gold and FX do not. If a signal is only confirmed late, the late fill is what we test.

When the gross result is positive but the net result is negative, the cause of death is Costs.

04The pass/fail gates

A strategy has to clear all of these. Clearing most of them is still a burial.

Passing every gate does not mean "trade this". It means the idea survived our tests, and it moves to forward testing on a simulator before anyone should risk money.

05The control group: random timing

A strategy can look good simply because the market drifted, or because the exits are doing the work. So each test also runs a control: the same exits, stops, sizing, session and costs, but with the timing of the entries randomised.

• We run 500 control simulations and place the real strategy within that distribution. Two designs are used, and each post says which: moved entries, where all the real entries are shifted together by a random number of days (or sessions) to the same time of day, keeping their direction and order; and random entries in the same state, drawn where the signal's own condition holds (for example RSI below 50), keeping side and trade count.
• Where it applies we also run placebo versions of the idea (shifted levels, sign-flipped returns) to check that our own code doesn't invent edge on random data.
• A strategy that can't beat the 95th percentile of the random runs is reported as having No proven edge, however good its equity curve looks. This means the test did not demonstrate a statistically convincing edge. It is not proof that none exists.
• The control has to match what the signal conditions on. Random entries at random hours are not enough if the real signal only fires in a particular state, such as RSI below 50 or price far from its average. Two of our early controls matched the side and the trade count but not that state, and when we checked, the answer moved (see the changelog). For every test from 8 Oct 2026 on, the frozen specification defines its control, and the post says what the control randomises and what it keeps. Most of these tests use moved entries, which keep the time of day and direction but not the signal's state; a few use random entries in the same state.

06We report everything we tried

Every post states the count: variants tried, variants that passed, instruments and periods covered. Multiple tries make a lucky result more likely, so the count is part of the verdict.

• Failures are published with the same detail as passes.
• Mixed results are shown as mixed. One good year, one good instrument or one lucky outlier is not a pass.
• A running scoreboard tracks totals across the whole site.
• Several markets: support means more than one pair. When a strategy is run on other markets, every variant is declared in advance in a dated specification, all results are published, and a variant counts as supported across markets only if it passes every gate on at least 3 of 4 pairs. One lucky pair is reported as a single-pair result. This is the bar for these studies, not a universal definition of an edge: an edge can exist in one market only. The first such study is the FX study.
• Variants added later are labelled "Round 2". If we add versions after seeing a result (for example a fixed stop and target), they are flagged as added after the fact, reported in full, counted in the total, and judged against the stricter bar above. The original numbers never change.

07Verdicts and causes of death

Every strategy ends up with one verdict and, if buried, one primary cause of death.

08What this protocol can't do

• It can't prove a strategy never works. It shows it didn't work on this data, with these rules, under these costs.
• Discretionary parts are not tested. Judgement calls, "reading the market", news filters and mental stops can't be backtested. We test the mechanical version and say what was left out.
• Drawdown is measured on closed trades unless a table says it includes open losses (an approximation from the worst point reached inside each trade). A closed-trade drawdown understates real account risk, especially for strategies without a stop.
• Backtests are not live trading. Liquidity, queue position and psychology are missing.
• History is finite. A holdout opened once is one sample. That's why survivors go to forward testing.
• Our first random controls were not all matched to the signal. We have now re-checked all seven strategy tests with controls matched on the signal's state, and every post shows its check. The answers changed the reading of two tests (RSI divergence, Bollinger), softened claims in two (Donchian, and Supertrend's long-only version), showed a result that depends on the control in one (RSI(2)) and left two unchanged (MACD, gold). No frozen gate result or verdict changed, because those results stay as published and every variant had already failed other gates. The matched controls are a sensitivity check beside the frozen result, not a replacement for it.

We test rules and public claims, never people. No ratings of individuals, only of strategies.

09Reproducibility

Each post links to the frozen specification, the code used and the full trade log. If you run the same rules on the same kind of data and get a different answer, tell us. A corrected result gets published as a correction, not quietly edited.

Where the idea and the code come from. Every post names the source of the claim being tested, so you can check the original. It also states the provenance of the code: our own code, written from public rules; open-source code used under its licence, with credit; or not published, when the original is closed or paid, in which case we rebuild the rules from the author's public description. We never publish anyone else's closed, paid or decompiled code.

10Before we publish: the checklist

Every post, and every change to a post, goes through these ten checks.

• Summaries match tables. Words like "all", "every" and "none" are checked against each variant's number. If three of six ended positive, the text cannot say all six lost.
• Every loss says what it measures. A closed trade or a loss open inside a trade. A drawdown on closed trades or including open losses.
• The counts agree everywhere. Variants tried, variants passed, and posts published are the same on the post, the home page and the About page. Round 2 variants are counted.
• The holdout is described honestly. If it was already used, the post says so and the table marks it.
• Careful wording. "Not demonstrated", never "does not exist". The label is "No proven edge".
• Numbers come from the results files, not from memory or by hand.
• Changes after publication are logged in the post's Corrections section.
• Cause labels follow their definitions. "Costs" only where gross is positive and net is negative. A variant that stays profitable after costs is "No proven edge" or another cause.
• Claims about stops are checked against both loss measures, the worst closed trade and the worst loss open inside a trade, for every configuration. "The stop capped the worst loss" must be true for all of them or be scoped to the ones where it is.
• The control is described exactly. The post says what the random runs randomise (entries, exits or both) and what they match (trade count, side, entry state), and any check of an alternative control is shown beside the frozen result, not instead of it.

An automated checker compares the published pages with the results files and flags anything inconsistent for a person to review. It never rewrites a conclusion. It also has a self-test, which confirms it still catches the kinds of error it exists for. A person reviews every flag before anything goes live.

11Changelog

2026-11-02 · Every published version reversed (68 tests, 435 versions; a reading of the published trade logs, no new backtest; chosen after seeing every result, disclosed) completed and published as specified: 0 of 435 reversed versions passed every gate; of 110 versions chosen on development data, 27 ended positive after costs on later data. The self-test failed its first run on an error in the check itself (it read another market's summary from the same folder), fixed before any reversed result; a mutation check showed three weak checks, which were strengthened; an independent second script gave identical numbers on every version.

2026-11-02 · VWAP standard deviation bands on gold, 5-minute bars (the band rule buried on ES, chosen knowing that result) completed and published as specified: 2 variants (London and New York sessions), 0 passed every gate, both 1 of 6, buried, cause Costs by our rule, though the same trades moved to other sessions did at least as well. The self-test failed its first runs on errors in the test itself, fixed before any result and noted in it; a mutation check showed three weak checks, which were strengthened.

2026-10-29 · The Forex Factory “Monthly Profit - Daily Trading using SMA” system (SMA 5 cross on daily candles with a Bollinger filter) on EUR/USD completed and published as specified: 3 variants (closed at the end of the day, 3R target with breakeven, trailing stop with a band exit), 0 passed every gate, best 2 of 6, all buried (no proven edge; the one-day version labelled Costs by rule, with almost nothing before costs). Moving-average and trend rules tried before on this file; disclosed in the specification. The self-test passed after fixing two hand checks written wrong in the test itself.

2026-10-28 · Gold 5-minute trend pullback (1-hour 200 EMA, EMA 20/50, pullback and confirmation candle, $2 minimum stop) completed and published as specified: 4 variants (long and short, 1.5R and 2R), 0 passed every gate, all 1 of 6, buried (the longs with no proven edge, the shorts by costs). One documented clarification of the window's description after freezing, before any result. The self-test failed its first runs on errors in the test itself, fixed before any result and noted in it.

2026-10-28 · TheStrat on gold (XAUUSD) 60-minute bars, the ES rules replicated unchanged, completed and published as specified: 4 variants, 0 passed every gate, best 2 of 6, all buried (three by costs, one losing before costs); the 2-2 reversal positive on ES lost on gold. Chosen after the ES result; disclosed in the specification. The self-test passed on its first run.

2026-10-28 · TheStrat (Rob Smith) on ES 60-minute bars completed and published as specified: 4 variants (2-1-2, 3-1-2 and 2-2 reversal with full timeframe continuity, and the 2-1-2 without it), 0 passed every gate, best 3 of 6, all buried, no proven edge (two by costs, one losing before costs, the 2-2 reversal fragile). Entry, stop and target taken from secondary sources; disclosed in the specification. The self-test failed its first runs on errors in the test itself, fixed before any result and noted in it.

2026-10-28 · Trader Dale's weekly anchored VWAP pullback (from a YouTube webinar) on EUR/USD 30-minute bars completed and published as specified: 2 variants (a target before the next barrier, a trailing stop), 0 passed every gate, best 2 of 6, both buried, cause Costs. The VWAP strategies on ES were tested and buried before; disclosed in the specification. The self-test passed on its first run.

2026-10-27 · Forward test of MACD + RSI with a 200 EMA filter, MRE-15-1.5R, on NQ: protocol v1.0 frozen with the owner's approval (a paper series from Monday 2 November 2026, computed monthly by the backtest engine on new NQ exports; evaluations at 300, 600 and 900 closed trades). The page shows the protocol and the monthly journal.

2026-10-26 · ICT Silver Bullet on ES 5-minute bars completed and published as specified: 4 variants (New York AM or PM window, 2R or liquidity target), 0 passed every gate, best 2 of 6, all buried, no proven edge (the 2R versions by costs, the liquidity versions fragile). Chosen after the change of character test on NQ failed; disclosed in the specification. The self-test failed once on a descriptive counter (no trade changed) and passed after the fix; one hand-made case was added after a deliberate code change showed it was missing.

2026-10-26 · MACD + RSI crossover with a 200 EMA trend filter on NQ 5-minute bars completed and published as specified: 4 variants (trend on 15-minute or 1-hour bars, target 1.5R or 2R), 0 passed every gate, best 4 of 6 (all four passed the random-entry control and stayed positive without their 10 best trades; all failed the validation profit factor and the development t), all buried, no proven edge. The self-test passed on its first run.

2026-10-26 · Change of character (ChoCh / market structure shift) after a liquidity sweep on NQ completed and published as specified: 4 variants (1-minute and 5-minute structure, entry on the shift or on the retest), 0 passed every gate, best 2 of 6, all buried, no proven edge. The self-test passed on its first run; one hand-made case was added afterwards when a deliberate code change showed it was missing, and it passed. The same day, a descriptive change-of-character split was added to the EUR/USD order block post (computed after its results were published; not a test).

2026-10-23 · Liquidity sweep + order block + fair value gap (smart money concepts, from a YouTube video) on EUR/USD 5-minute bars completed and published as specified: 2 variants (an aggressive and a conservative entry), 0 passed every gate, both 1 of 6, buried, cause No proven edge. The self-test passed on its first run.

2026-10-23 · Correction to two gold posts (best time to trade gold; H1 Open Trap): their notes said the clock correction moved the weeks of 2004 to 2010 that the 9 Oct clock audit found an hour early; the correction they used (from the EMA 9/20 gold test) moves 20 clock-change weeks of 2020 to 2026 instead. Both tests were re-run on the audit's clock and the comparison added to each post: no verdict changed; on the night/day post the rise from 10 a.m. and the long-short version's small profit after costs do not hold on that clock.

2026-10-23 · The "sneaky pivot" on gold (XAUUSD), 15-minute bars from the COMEX open (the YM rule replicated, chosen after the YM result; four versions fixed in advance) completed and published as specified: 4 variants, 0 passed every gate, best 2 of 6, all buried, causes Costs (2) and No proven edge (2). After the results, a third run on the clock of our 9 Oct clock audit was added as a check, because the frozen loader's clock correction does not move the 2004-2010 weeks the audit found an hour early: no verdict changed.

2026-10-23 · Forward test of the Sneaky Pivot SP-STRICT-T2 on YM: protocol v1.0 frozen with the owner's approval (a paper series from Monday 26 October 2026, computed monthly by the backtest engine on new YM exports; evaluations at 100, 200 and 300 closed trades). The page shows the protocol and the monthly journal.

2026-10-23 · The "sneaky pivot" on YM, 15-minute bars (a YouTube day-trading rule at the previous day's high and low; four versions fixed in advance) completed and published as specified: 4 variants, 0 passed every gate, three passed 5 of 6 (failing only the development t-statistic), all buried, cause No proven edge. Before any result was reported, a bias in our moved-trade control was found and fixed (it entered at a price often at the top of the minute); two hand-made expectations in the self-test were corrected; the engine was not changed.

2026-10-22 · Trendline breakout and bounce on YM, 1-hour bars (the rules of a YouTube masterclass made mechanical; four versions fixed in advance; first test on the YM file exported and checked the same day) completed and published as specified: 4 variants, 0 passed every gate, all buried (Fragile, No proven edge). The self-test passed after one hand-made expectation in it was corrected; the engine was not changed.

2026-10-22 · The NR7 / NR4 breakout on ES (three versions fixed in advance, with the next-day range statistics) completed and published as specified: 3 variants, 0 passed every gate, all buried (Fragile, No proven edge, Costs). The self-test passed after one hand-made expectation in it was corrected (the flat test day was itself an NR7 day); the engine was not changed. The specification's warning that the NR7 versions would have fewer than 50 validation trades proved wrong for NR7-BO and is corrected in the article.

2026-10-22 · Stop loss vs no stop loss across our backtests (183 pairs of published versions with the same entries, listed before any pair was compared; no new backtest) completed and published as specified. Its checks passed on the second run, after an over-strict inventory check (one log unit per test, where a two-market test has one per market) was corrected to one unit per pair; an independent second script, reading the trade logs itself, gave identical numbers.

2026-10-22 · Trend following vs mean reversion across all our backtests (387 versions labelled by rule kind before any result was joined; no new backtest) completed and published as specified, with one documented fix: one test missing from the frozen label table was labelled by the frozen gold-guide rule before any result was read. Its checks passed on the first run; an independent second script gave identical numbers.

2026-10-22 · Cross-market validation scoreboard (the same rules on several markets in 9 families of published tests, 682 market pairs; no new backtest; rules and prediction frozen before any cross-market number was computed) completed and published as specified. Its checks passed on the first run; an independent second script gave identical numbers.

2026-10-22 · Trading costs across all our backtests (a comparison of the published gross and net results, no new backtest; rules frozen before any number across the tests was computed) completed and published as specified: 54 tests, 387 versions. Its checks passed on the second run, after a fix in the checker's reading of one CSV cell with a comma; an independent second script gave identical numbers.

2026-10-22 · The ES opening gap fill (three versions fixed in advance, with fill statistics by gap size) completed and published as specified: 3 variants, 0 passed every gate, all buried for costs. The self-test passed on its first run.

2026-10-22 · Trading Made Simple on EUR/USD (the TDI author's complete rules: candle rule, bounces, stochastic filter, night routine), chosen after the TDI result and stated as such, completed and published as specified: 3 variants, 0 passed every gate. The self-test passed on its first run.

2026-10-22 · The TDI (Traders Dynamic Index) on EUR/USD 4-hour bars (three readings of a Forex Factory method, fixed in advance) completed and published as specified: 3 variants, 0 passed every gate, all buried for costs. The self-test caught a wrong expectation in our own hand-made example before any result; corrected and passed.

2026-10-22 · The Accumulation/Distribution line on SPY daily bars (the Bitcoin rules, long only, frozen after the Bitcoin result) completed and published as specified: 3 variants, 0 passed every gate; the crossing rules did worse than random timing. The self-test passed on its first run. The Bitcoin post now links it, with a correction note.

2026-10-22 · The Accumulation/Distribution line on Bitcoin 4-hour candles (three versions fixed in advance, Binance volume) completed and published as specified: 3 variants, 0 passed every gate. The self-test passed on its first run.

2026-10-22 · The Parabolic SAR on EUR/USD 1-hour bars (Wilder's defaults, three versions fixed in advance) completed and published as specified: 3 variants, 0 passed every gate. The self-test passed on its first run.

2026-10-21 · The engulfing candle on gold 1-hour bars (four versions fixed in advance) completed and published as specified: 4 variants, 0 passed every gate, all buried for costs. The self-test caught a rounding issue in the 20-bar average before any result; fixed and passed.

2026-10-21 · Heikin Ashi on Bitcoin 4-hour candles (three common versions fixed in advance) completed and published as specified: 3 variants, 0 passed every gate; the no-wick entry with a 200-bar average passed 5 of 6 (it failed the validation profit factor). The self-test passed on its first run.

2026-10-21 · The Initial Balance breakout on ES (the NQ rules unchanged, a replication frozen after the NQ result) completed and published as specified: 3 variants, 0 passed every gate, none beat the control. The self-test passed on its first run. A paragraph and a correction note on this replication were added to the NQ post.

2026-10-21 · The Initial Balance breakout on NQ (break of the first-hour range, three stop and target versions fixed in advance) completed and published as specified: 3 variants, 0 passed every gate; the best two passed 4 of 6 and beat the control. The self-test passed on its first run.

2026-10-21 · The first 5-minute candle with a fair value gap on NQ (the rules of a YouTube scalping video, with the entry window and exit the video leaves open fixed in advance) completed and published as specified: 2 variants, 0 passed every gate. The self-test passed on its first run.

2026-10-21 · The Bitcoin CME gap fill (spot at the CME weekly close and reopen; fill rates, and fading gaps of 1% or more with three stop rules) completed and published as specified: 3 variants, 0 passed every gate. During the self-test, before any result, one expectation written in the test itself was corrected (noted in the self-test file).

2026-10-21 · ADX + DI on the four forex majors, 1-hour bars (Wilder's ADX 14 crossing 20 or 25 in the DI direction, exit on the DI cross) completed and published as specified: 8 pair-variants, 0 passed every gate, 0 of 4 pairs for either threshold under the cross-market rule. During the self-test, before any result, the control's handling of flips was corrected (noted in the self-test file).

2026-10-21 · Inside bar breakout on EURUSD daily bars (next-day stop orders beyond the mother bar, stop at the other side; targets 1R and 2R, and 2R with a 50-day trend filter) completed and published as specified: 3 variants, 0 passed every gate. The self-test passed on its first run.

2026-10-21 · London breakout on GBPUSD (Asian range 00:00 to 07:00 London, stop orders at the open, stop at the other side; targets 1R, 2R or none) completed and published: 3 variants, 0 passed every gate; two beat the random-day control. Before any result the base cost was raised from 1.0 to 1.5 pips and the clock check was replaced (the first method could not tell; no week shifted), both noted in the specification.

2026-10-21 · Gold at night, gold by day (the overnight effect of Blose and Gondhalekar, 2014, on spot gold; day = 08:20 to 13:30 New York) completed and published as specified: 3 variants, 0 passed every gate. The article adds a descriptive table of gold's average range by hour, computed after the test and outside its gates. The self-test passed on its first run.

2026-10-21 · The 200-day moving average rule on SPY (daily, and Faber's monthly 10-month rule; dividends, T-bill interest in cash) completed and published as specified: 2 variants, 0 passed every gate (one Inconclusive with fewer than 30 trades). The site's first stock-market test, with a new market page. During the self-test, before any result, one control case written in the test itself was corrected (noted in the self-test file).

2026-10-21 · H1 Open Trap on gold minute data (an MT4 indicator's trap signals with its default settings; targets 1R, 2R and 3R, the last two with a trailing stop) completed and published as specified: 3 variants, 0 passed every gate. The same day, after its results, a follow-up with a $5 minimum stop was specified and frozen before it was run (1 variant, buried), reported together. During the self-test, before any result, one check written in the test itself (an arbitrary count) was replaced by an exact one (noted in the self-test file).

2026-10-17 · Opening range breakout (5 minutes) on NQ futures (the rule of Zarattini and Aziz, 2023, and the classic breakout of the same range; 10 R target or the session close) completed and published as specified: 4 variants, 0 passed every gate. During the self-test, before any result, one hand-made check written wrong in the test itself was corrected (noted in the self-test file).

2026-10-17 · Head and shoulders pattern on 15-minute gold (Osler and Chang's published rules at two sizes; entry on a close beyond the neckline, stop beyond the right shoulder, target the pattern's height) completed and published as specified: 4 variants, 0 passed every gate; the best passed 5 of 6. During the self-test, before any result, a helper that built the hand-made test prices was corrected (noted in the self-test file).

2026-10-16 · WaveTrend oscillator on 1-hour gold (LazyBear's defaults: crosses in the ±53 zones, out at the next cross or at the next opposite signal) completed and published as specified: 4 variants, 0 passed every gate. During the self-test, before any result, two checks written wrong in the test itself were corrected (noted in the self-test file).

2026-10-15 · Noise area breakout on Bitcoin around the clock (the UTC day, every day of the week) completed and published: the 30-minute test as specified (2 variants) and, the same day after its results, a 15-minute test with its own frozen specification (2 variants), reported together; 0 of 4 passed every gate. The engine of the earlier noise-area tests was not changed (its fingerprint is pinned by the NQ forward test): a separate engine had to reproduce all of its ES and NQ trades first. During the self-test, before any result, one check we had written wrong (it expected four weekend days in a sample with five) was corrected.

2026-10-15 · Forward-test protocol for the TTM Squeeze on gold (SQ-L-B) frozen before any new data: a paper series computed by the backtest's engine on the MT4 gold file, start Monday 19 October 2026, evaluations at 100, 200 and 300 trades. It does not change the historical verdict. The same day the causes of death on three posts were relabelled to the protocol's definitions (notes on each post).

2026-10-14 · Forex-year study (fy_spec.md, frozen before any simulation): random-entry traders on real EUR/USD 1-minute prices, 2005 to 2025, three styles, three risk sizes, two cost levels and two leverage limits, 819,000 simulated trading years. Published as the guide “Why Most Forex Traders Lose Money”, with the official figures quoted from their sources. A simulation study, not a strategy test, so it has no gates or verdict.

2026-10-14 · Prop firm challenge study (pc_spec.md, frozen before any simulation): 22.4 million simulated challenges under generic rule types, for four traders and four risk sizes, with a consistency rule added to the simulator (off by default; the earlier prop firm post reproduces exactly). Published as the guide “How to Pass a Prop Firm Challenge”. It is a simulation study, not a strategy test, so it has no gates or verdict.

2026-10-14 · Two tests run before this protocol existed were removed from the home page and the market pages: an opening-range breakout with fair value gaps on ES (“ORB + FVG v2”) and weekly Fibonacci pivots with TDI on gold and four FX pairs. Their results were shown, but their rules, code and trades were never published, so nobody could check them. They were never counted in any total. If either is tested under this protocol, it will be published in full like every other test.

2026-10-14 · Corrections to this page after a site audit, so that it describes what the tests actually do. (1) The sample-size gate said “(200 for long histories)”. No specification ever used that figure; every test used 50 trades per period. Under 200, the 1-hour golden cross on gold (149, 112 and 126 trades in its three periods) would have failed this gate in its run; its verdict is Inconclusive either way. We removed the figure instead of applying it after the fact, and say so here and on that post. (2) The control section said that every test from 8 Oct 2026 uses a control matched on the signal's entry state. Most of them move the real entries in time instead; both designs are now described, and each post already said which it used. (3) Gold costs include no slippage, which is optimistic, so “fills are pessimistic” was not true for gold and FX; corrected. (4) The instruments and costs now list NQ, gold futures and Bitcoin. No gate result or verdict changed.

2026-10-14 · TTM Squeeze on gold, 1-hour bars (Bollinger 20, 2.0 inside Keltner 20, 1.5; long and short or long only; a quick momentum exit or a zero-cross exit) completed and published as specified: 4 variants, 0 passed every gate; the best passed 5 of 6. The same day a bug was found and fixed in the control of the RSI 30/70 gold test (a check hidden in a code comment): its control was rerun, no gate or verdict changed, and the correction is noted on its page.

2026-10-14 · Fibonacci Golden Zone scalping on ES futures (the rules of a Google AI answer: the 09:30 opening range, a pullback to 50% to 61.8% of the push, a 1-minute limit and a 5-minute confirmation, each with and without the VWAP) completed and published as specified: 4 variants, 0 passed every gate.

2026-10-14 · Ichimoku Cloud strategies on EUR/USD (the TK cross with the cloud and the Kumo breakout with the Chikou, closed at the Kijun-sen, on 1-hour, 4-hour and daily bars; the clock check run on a corrected copy of the file) completed and published as specified: 6 variants, 0 passed every gate.

2026-10-14 · 9 EMA strategies on NQ futures (a retrace to the 9 EMA with the VWAP on 5- and 15-minute bars, the 9/20 break and hold on 5-minute bars, as described in a YouTube video; exits fixed by us as the video gives none) completed and published as specified: 3 variants, 0 passed every gate.

2026-10-14 · Stochastic + RSI + MACD strategy on EUR/USD (15-minute, 30-minute and 1-hour bars, as shown in a YouTube video, with the video's own 0.15% stop cap as a second version) completed and published as specified: 6 variants, 0 passed every gate.

2026-10-14 · VWAP strategies on ES futures (5-minute bars, New York session: with the VWAP, against it, and VWAP bands, as described by an Investopedia article) completed and published as specified: 3 variants, 0 passed every gate.

2026-10-14 · Classic RSI(14) 30/70 on gold (5-minute and 1-hour bars, classic and protected versions) completed and published as specified: 8 variants, 0 passed every gate. Before the article one report figure (the worst open loss) was fixed; trades, gates and verdicts were identical.

2026-10-14 · Intraday trend pullback on Bitcoin (15-minute entries, 1-hour trend, 2R target, flat by midnight UTC) completed and published as specified: 2 variants, 0 passed every gate. During the engine self-test, before any result, stop levels were set to round to the cent (noted in the specification).

2026-10-14 · Turtle trading on daily Bitcoin (Binance BTCUSDT, August 2017 to September 2026) completed and published as specified: 4 variants, 0 passed every gate. The first test with a fresh holdout (no earlier test had used Bitcoin data). The Turtle engine got additive options for UTC days, a cost in percent and the Bitcoin swap; the gold and EUR/USD runs reproduced exactly. A Crypto market page was added.

2026-10-11 · Forward-test protocol for the NQ noise area breakout (NA-NQ-NY-LS) frozen before any new data: a paper series computed by the backtest's code on the monthly NQ export, start Monday 12 October 2026 09:30 New York, evaluations at 250 and 500 trades. It does not change the historical verdict.

2026-10-11 · Noise area breakout on ES, NQ and XAUUSD (Asia, London and New York sessions) completed and published as specified, one post per market: 18 variants, 2 passed every gate, both on NQ in the New York session (long and short, and long only), and both keep that verdict under the data rule. The site again has survivors, but the NQ holdout had been used by earlier tests and the same rule failed on ES and gold, so they are candidates for a test on new data. Before publication a bug that marked almost every session as touching a hole in the data was fixed; trades, bands and gates were not affected (logged in the specification).

2026-10-11 · Turtle trading on daily EUR/USD completed and published as specified: 4 variants, 0 passed every gate.

2026-10-11 · Intraday momentum (first half-hour to last half-hour) on ES and NQ completed and published as specified, one post per market: 8 variants, 0 passed every gate.

2026-10-09 · Turtle trading on daily gold completed and published as specified: 4 variants, 0 passed every gate.

2026-10-09 · Fibonacci 61.8% pullback on 1-hour EUR/USD completed and published as specified: 2 variants, 0 passed every gate.

2026-10-08 · EMA 5/8/13 ribbon on NQ (5- and 15-minute bars, US session) completed and published as specified: 8 variants, 0 passed every gate; the 15-minute version is listed as a candidate for a check on new data with the rules unchanged. No rule, cost or gate was changed after the specification was frozen.

2026-10-08 · Data rule applied to the gold survivor of the 1-hour golden cross: without the trades whose signal or fill touches a suspect gap in the thin 2004 to 2012 data, its development t-statistic falls to 2.29 (2.5 needed), so its verdict changed from Survivor to Inconclusive (data quality). The site now has no survivor. The forward test of the rule continues unchanged.

2026-10-08 · EMA 9/20 crossover on XAUUSD (15-minute and 1-hour bars) completed and published as specified: 24 variants, 0 passed every gate. Before the run the gold data was audited: a clock error in 20 weeks (2020 to 2026) was fixed, the last bar of the file was found to have been still forming when the earlier gold studies read it (their results reproduce exactly with the value they read), and the gaps of the file were listed. No rule, cost or gate was changed after the specification was frozen.

2026-10-08 · EMA 9/21 crossover on 1-hour ES, NQ and GC bars completed and published as specified: 12 variants, 0 passed every gate. A data rule was added for tests from this date on: known holes in the 1-minute files are listed, trades open across a hole are counted, and verdicts are recomputed without them (a changed verdict becomes Inconclusive); the ES studies published before it were rechecked this way and no verdict changed. No rule, cost or gate of the EMA test was changed after its specification was published.

2026-10-08 · Coppock Curve on 1-hour ES bars completed and published as specified: 12 variants, 0 passed every gate, all 12 lost money after costs. It replaces the earlier unprotocolled Coppock tests on ES, which had no frozen specification. No rule, cost or gate was changed after the specification was published.

2026-10-08 · Golden cross on 1-hour EUR/USD bars completed and published as specified, as the first transfer test of the gold survivor: 12 variants, 0 passed every gate, none beat random timing at the 95th percentile. A floating-point tie on bars where the two averages are mathematically equal was found while extending the engine and handled before the run; the cross rule, costs and gates were not changed (logged in the specifications).

2026-10-08 · Golden cross on 1-hour bars completed and published as specified: 24 variants on ES and gold, with and without a stop and targets. One variant (gold, long only, no stop, no target) cleared every gate and 23 did not; it is a single-market result with a narrow pass on the development t-statistic (see the post). No rule, cost or gate was changed after the specification was published.

2026-10-08 · Rule added, for tests from this date on: a variant with fewer than 30 trades in the full sample is Inconclusive whatever its gates show, because a gate failed on a handful of trades is not evidence of anything. Used first in the daily golden cross test, where 7 of 8 results were Inconclusive. With 30 trades or more the usual rule applies.

2026-10-08 · Rule added, for tests from this date on: the control in the frozen specification must be matched on the signal's entry state, not only on the side and the trade count. Reason: re-checking the RSI divergence control showed that about half of the real signals fire in a state where the control never trades (long-only version: 78th percentile against the frozen control, 99th against a matched one), and the Bollinger control flattered its signal (98.8th percentile against the frozen control, 90th against a distance-matched one). The earlier gate results stay as frozen and the alternative controls are reported beside them. Checklist extended from seven to ten checks (cause labels, stop claims, control description).

2026-10-08 · FX study completed and published exactly as specified: 96 variants on four pairs, 0 passed every gate, 0 met the 3-of-4-pairs bar for support. No rule, cost or gate was changed after the specification was published.

2026-10-08 · Added market pages (futures, metals, FX) and pre-registered the FX study before running it: 24 variants on each of four pairs, with the cross-market rule above and the matched-control rule. The specification is public and dated.

2026-10-08 · Control audit completed: the random controls of RSI(2), Supertrend, Donchian, MACD and the gold breakout were re-checked against controls matched on the signal's state (trend and dip, holding pattern, position in the range, entry minute and side). Each post now has a "Control check after review" section. Several sentences were reworded; no gate result, verdict or cause of death changed.

2026-10-08 · Wording, no change to the gates: statements that a stop and target "cannot create an edge" were replaced by statements limited to the combination tested.

2026-10-07 · Added the pre-publication checklist and an automated consistency checker that flags problems for human review. No change to the gates.

2026-10-07 · Wording: the verdict label "No edge" became "No proven edge", meaning no statistically convincing edge was demonstrated in the test. It is not a claim that none exists.

2026-10-07 · Clarification: drawdown is measured on closed trades unless stated, and every post now carries a "what this shows, what remains unknown, why it failed, what you can apply" section. No change to the gates.

2026-10-07 · Rule added: the development t-statistic bar rises from 2.0 to 2.5 when more than four variants of a strategy have been tried in total, and variants added after a result are labelled "Round 2" and counted.

2026-10-07 · Clarification, no change to the gates: code provenance statement added to section 09. Instruments wording widened from "ES and MES" to futures in general, with a note on spot and CFD data.

v1.0 · 2026-10-07 · First public version. Earlier internal tests predate this page and used slightly different gates; they are labelled "legacy" where they appear.