Research evidence
What each module was measured to do Every signal module we built or tested is judged against one bar, fixed before any of them was run. This is the full register - including everything that failed. A profit figure on an unvalidated strategy is not information, so no module sends orders on enthusiasm: it runs only as far as its evidence allows.
55
Modules
0
Validated
9
Candidates
1
Observe
45
Failed
10
Running in IEB
The bar A module is validated only when all three hold: clustered t > 3.0 , a positive result in at least 4 of 5 chronological windows, and at least 70% of symbols agreeing in sign. The bar was fixed before any module was run and is not adjustable.
VALIDATED
Cleared the pre-set bar.
CANDIDATE
Measured and real, but below the bar. Runs only as a labelled experiment.
OBSERVE
Not measured yet. The mechanism is implemented and verified; the edge is unknown. Runs on demo to collect forward data.
FAILED
Measured, no edge. Sends no orders; a module may still be watched as a paper observer.
The register Select a module for its tests, the measured numbers and its caveats. Snapshot of 2026-10-17; register file SHA-256 8d914ff131c0 . Terms: t = t-statistic; R = result in units of the stop distance; bp = basis points; “toll” = trading costs (spread, commission).
ALL · 55 CANDIDATE · 9 OBSERVE · 1 FAILED · 45
QHMR - quiet-hour fade, market vs limit (IC Markets) qhmr_xcross + M1 · GBPCAD, GBPAUD · 16:00-16:34 NY · 30 min / 16:44 exit · paired A/B FORWARD TEST (demo, paired market + limit). Entries beat random by +0.15 R (t 5.8) but net -0.08 R at market costs; this measures whether limit entries close the gap. Overnight Drift overnight_drift + Daily · SPY/QQQ/ES/NQ · hold the close-to-open gap 8,457 SPY and 6,913 QQQ sessions. Filtered Sharpe 0.90/1.09, retention 1.00/0.96. OOS t 2.36/2.47 against a 2.73 threshold - BELOW it; carried by retention and a 30-year prior, not by clearing the bar. Mandated Month-End Rebalancing Flow cme_rebal_flow + Daily - ES/YM/RTY - 4 sessions/month, signed by intra-month divergence First token to clear Gates 1-3 and its binding control. Edge +30.06 bp, 95% CI [+1.89, +58.23] excludes zero, z +2.63 vs the sign-permutation null (99.3rd pct), MaxDD -3.96% vs -15.37% for the passive control. FAILS Gate 4 on one era with t -0.24. Indeterminate, not refuted. Primitive: FVG (H1) prim_fvg_h1 + H1 · 12 instruments · 10-bar hold REAL-BUT-UNTRADEABLE. 10/12 positive (corr-adj t=+2.02, p=0.0193), but edge/toll 0.31x covers toll on 0/12. Mean Reversion meanrev + H4 · 9 FX pairs · 8-day hold Real but weak. Passed both kill tests; below the bar. Institutional VWAP institutional_vwap + Execution benchmark · traded-volume instruments only NOT BACKTESTED in this project. It is an execution benchmark, not an anomaly - its value is the reference level, not a directional claim. FVG H1 - XAUUSD only (IC Markets) fvg_xau_ic + H1 · XAUUSD · 10-bar hold · 0.5% risk sizing FORWARD TEST of the one net-positive FVG cell (XAUUSD at IC, +0.07 bp net, t<1.91). IC Markets venue arm. FVG H1 - XAUUSD only (Pepperstone) fvg_xau_pep + H1 · XAUUSD · 10-bar hold · 0.5% risk sizing FORWARD TEST of the one net-positive FVG cell (XAUUSD at IC, +0.07 bp net, t<1.91). Pepperstone venue arm. FVG H1 - XAUUSD only (FTMO) fvg_xau_ftmo + H1 · XAUUSD · 10-bar hold · 0.5% risk sizing FORWARD TEST of the one net-positive FVG cell (XAUUSD at IC, +0.07 bp net, t<1.91). FTMO demo venue arm (1514960358, armed 2026-10-15). VPA-SMC Panel v2 (full port) vpa_smc_v2 + M5 - IEB watchlist on IC Markets - structural SL/TP, 288-bar backstop UNMEASURED. Complete port of the 2,391-line Pine panel: the v1 engine plus the eight subsystems it was missing. Armed for forward observation at the user request. Its own family FAILED: see vpa_smc. XAUUSD High-Vol London ORB (Phase 4 Frozen) xau_orb_phase4 + M15 setup · XAUUSD.raw · Passive Limit B at 0.10 ATR · 8-bar (120m) hold HOLDOUT FAIL. The module rule on the untouched FTMO 2023 year: 94 trades, net -0.005 R (t -0.04), 0.16 R worse than random entries in the same window, 1/3 quarters positive. The earlier +0.4253 ATR / t 2.80 backtest read the unfinished day (look-ahead). Artemis Squeeze Momentum v1.4.0 artemis + M5-H4 · gold + index CFDs · Confluence Buy/Sell and Squeeze Release · DeCasteljau v10.7 levels No cell clears the bar. History test on IC M1 2021-12..2026-09, 50 cells: best Squeeze Release XAUUSD M15 +55R over 3,000 signals (t 1.7), XAUUSD H4 +11.7R (t 1.6); Confluence negative on 20 of 25 cells; every M5 cell loses. Time Series Momentum (CME Futures) tsmom_futures + Daily · ES/YM/RTY/CL/GC · 12-month Moskowitz TSMOM Toll is non-binding (1.89 bp/yr on CLOB), but Sharpe 0.27 loses to static long-only Risk Parity (0.52) from post-2010 equity whipsaws. Quantum Gold Sovereign Zones qgs_zones + H1 · multi-asset · support/resistance zones FAILED. Best cell gross t +1.09, cost coverage 0.40 under prop-firm terms. 26% of entries fired on >3x spread blowout bars. Replaced by ORB. Real-Yield Shock -> Gold Globex Session real_yield_gold + H1 - GC/MGC - 18:00 ET entry, 10:00 ET next-day exit Near miss (edge +4.99 bp vs 6.68 bp true hurdle); Era 4 inversion reflects post-2022 sovereign gold decoupling from US real yields. Cash-Open Absorption (CME equity CLOB) clob_open_absorb + H1 - ES/YM/RTY - intraday 10:00-16:00 ET, flat overnight Toll slack on CME CLOB (659 bp / 13.5y), but conditional edge (+2.96 bp) misses hurdle (5.92 bp); Control C refutes overnight absorption mechanism. Chandelier Exit (TV) tv_chandelier_06 + H1 - 9 FX/metal - 24-bar proxy hold NULL. 66.7% of symbols positive, corr-adj t=+0.55, edge/toll +0.09x against a 4.0x bar. Primitive: Prior Week H/L (H1) prim_prior_week_hl_h1 + H1 · 12 instruments · 10-bar hold NULL. 8/12 positive, corr-adj t=+0.41, raw pooled t=+0.67. Below significance hurdle. WaveTrend Oscillator (TV) tv_wavetrend_03 + H1 - 9 FX/metal - 12-bar proxy hold NULL. 55.6% of symbols positive, corr-adj t=+0.28, edge/toll -0.03x against a 4.0x bar. US Index Cash-Close MOC Imbalance scalp_cash_close + M1 · US500, US100, US30 · 9-min hold (15:50-15:59 ET) REFUTED on gross: +0.30 bp, t=+0.20 before any cost. The 3.91 bp spread and the Control C "drift" in the first write-up were harness defects (see caveats). AlphaTrend (TV) tv_alphatrend_07 + H1 - 9 FX/metal - 12-bar proxy hold NULL. 33.3% of symbols positive, corr-adj t=+0.17, edge/toll -0.10x against a 4.0x bar. Volume-Weighted MACD (TV) tv_vwmacd_08 + M30 - 3 US ETFs - 12-bar proxy hold NULL. 33.3% of symbols positive, corr-adj t=+0.13, edge/toll -3.70x against a 4.0x bar. Primitive: Round Numbers (H1) prim_round_number_h1 + H1 · 12 instruments · 10-bar hold NULL. 6/12 positive, corr-adj t=+0.10, raw pooled t=+0.16. Indistinguishable from noise. Primitive: Asian Range (H1) prim_asian_range_h1 + H1 · 11 instruments · 10-bar hold NULL. 6/11 positive, corr-adj t=-0.13, raw pooled t=-0.22. Indistinguishable from noise. Stan Weinstein Stage 2 Cycle weinstein_stage + Daily · SPY (1993-2023) · 30-week MA trend Truncates max drawdown by half (-27.5% vs -56.7%), but Sharpe is flat (+0.01 gross, -0.01 fut) and CAGR falls 8.04% -> 5.28% from 33% cash drag. Ehlers SuperSmoother (TV) tv_ehlers_super_05 + H1 - 9 FX/metal - 8-bar proxy hold NULL. 44.4% of symbols positive, corr-adj t=-0.20, edge/toll -0.02x against a 4.0x bar. Primitive: Prior Day H/L (H1) prim_prior_day_hl_h1 + H1 · 12 instruments · 10-bar hold NULL. 5/12 positive, corr-adj t=-0.67, raw pooled t=-1.09. Indistinguishable from noise. Schaff Trend Cycle (TV) tv_stc_09 + H1 - 9 FX/metal - 12-bar proxy hold NULL. 22.2% of symbols positive, corr-adj t=-0.71, edge/toll -0.11x against a 4.0x bar. LETF Closing Auction Momentum (US30/US500) scalp_close_letf + H1 - ES/YM (MT5 US500/US30 tolls) - 15:00->16:00 ET after a >=1 sd day move Conditioning on >=1sd day move adds only +0.94 bp over quiet days (t +0.69); gross edge (+1.49 bp) fails Gate 1 before any toll. Bollinger Squeeze -> Directional Expansion squeeze_expansion + M15 - 12 FX/metals/WTI - 2-sigma break out of volatility compression REFUTED, and not by cost. Gross expectancy -0.90 bp at t -0.24 on 1,263 trades before any toll. Random entry with the same 1:2.5 bracket returns +1.11 bp, so the four-filter stack does not beat coin flips. Net after measured toll -5.91 bp, 0 of 12 instruments positive. Primitive: Order Block (H1) prim_order_block_h1 + H1 · 12 instruments · 10-bar hold NULL. 3/12 positive, corr-adj t=-1.04, raw pooled t=-1.69. Indistinguishable from noise. Volatility-Conditioned G10 Cross-Sectional Momentum volmom_g10 + Daily-from-H1 - 7 G10 USD pairs - 22-day rebalance, K=2 long / K=2 short REFUTED. Net -12.28 bp/month at t -1.08 over 144 non-overlapping monthly periods, 2010-2023. The volatility conditioner has the WRONG SIGN once correctly aligned (calm -17.56 bp vs stressed +1.99 bp, Welch -0.59) and makes the portfolio worse on both gross and net. Loses to random selection. Williams VIX Fix (TV) tv_vixfix_04 + H1 - 9 FX/metal - 10-bar hold NULL. 66.7% of symbols positive, corr-adj t=-1.19, edge/toll -0.07x against a 4.0x bar. DeCasteljau v12.1 decasteljau + M5 · intraday Port repaired against the Pine source; still negative at every setting. VWAP Dislocation 2.75-sigma Mean Reversion scalp_vwap_disloc + M1 · US500, US100, GER40 · 30-bar hold bracket Gross negative (-0.60 bp, t=-1.56); stops hit (38%) exceed mean reversion targets (30%). Primitive: Equal Highs/Lows (H1) prim_equal_highs_h1 + H1 · 12 instruments · 10-bar hold NULL. 6/12 positive, corr-adj t=-1.57, raw pooled t=-2.56. Anti-predictive. WMR Fix Momentum (16:00 London) scalp_wmr_momentum + M1 · EURUSD, GBPUSD · 5-min hold Gross negative on both pairs (-0.47 bp, t=-1.58); erased by 1.15 bp spread+commission. Month-End USD Hedge Flow into the 16:00 London Fix wmr_hedge + H1 · 6-pair USD basket · last London business day 15:00-16:00 London REFUTED before cost: basket gross -1.85 bp (t -1.64) against the pre-registered hedge-flow sign on 86 month-ends, 2010-2021. 0/6 legs in the signal direction. Neural Brain (LSTM direction classifier) brain + M15 - 11 FX/metals - 5-bar direction, 3-way Worse than always predicting HOLD on 8 of 11 instruments. Lorentzian k-NN Classifier (TV) tv_lorentzian_01 + H1 - 9 FX/metal - 4-bar hold NULL. 22.2% of symbols positive, corr-adj t=-1.99, edge/toll -0.06x against a 4.0x bar. Primitive: Fractal Pivot (H1) prim_fractal_pivot_h1 + H1 · 12 instruments · 10-bar hold NULL. 5/12 positive, corr-adj t=-2.16, raw pooled t=-3.52. Significantly anti-predictive. 08:30 London Judas Sweep Fade (ICT) judas_sweep_0830 + M5/M15 - 10 FX + gold + WTI - fade the 08:30 London sweep of the Asian range REFUTED. Fading the 08:30 London sweep of the Asian range is negative GROSS on 10 of 10 instruments, before toll. EURGBP t -2.39 (226 sessions), EURJPY t -2.00 (158), EURUSD -1.00 bp (167). Net after realistic toll -2.56 to -6.12 bp. The sweep continues; it does not reverse. 10:00 ET Macro Judas Reversal ten_am_macro + M5 · XAUUSD, EURUSD, US100, US30 · session-anchored REFUTED on gross: -2.40 bp (t=-2.39, -0.080R). Stop-out rate (55%) dominates targets (30%); sweep continuation beats reversal. Nadaraya-Watson Envelope (TV) tv_nadaraya_02 + H1 - 9 FX/metal - 12-bar hold NULL. 22.2% of symbols positive, corr-adj t=-2.49, edge/toll -0.32x against a 4.0x bar. Kaufman Adaptive MA (TV) tv_kama_10 + H1 - 9 FX/metal - 24-bar proxy hold NULL. 11.1% of symbols positive, corr-adj t=-2.84, edge/toll -0.26x against a 4.0x bar. QuantNeuralEdge Scalper scalper + M5 · scalp Win rate below the breakeven point of its own R:R. 9-Agent Consensus consensus_9 + FX M15 · 9-agent quorum REFUTED. 34,000 trades on EURUSD/GBPUSD M15. All 16 testable cells negative to t -6.94; GROSS also negative in all 8 checked, so costs were not what killed it. Loses to its own bias-matched control (Welch negative in every cell, to -2.41). min_conviction 0.55 -> 0.62 makes it WORSE in 7 of 8 cells: the conviction score runs backwards. Candle Range Theory crt + XAUUSD/XAGUSD M5-M15 · sweep and fade REFUTED. 7,557 trades, two specs, two metals. Best cell is +0.021R at t 0.43 against a control of -0.024R (Welch +0.92). Validation symbol fails outright at t -3.83 and -4.75. 0 of 16 grid cells clear the control at Welch 2.00. Donchian Breakout Failure donchian_fail + Equities · fade the failed break REFUTED on 15 years of IWM after looking strongest in the programme on 2 years of SPY. 1,318 IWM trades: t 1.83, out-of-sample retention -0.13 - the holdout LOST money. By era t +3.69 (2011-15), +0.28 (2016-20), -0.68 (2021-26). Textbook post-publication decay. NOT verified on FX. Opening Range Breakout orb + Session open · 15-min range REFUTED outside SPY. Charging the spread quoted on the entry bar collapses it: sp500 pays 0.299R per trade, taking +0.500R to +0.201R and retention from 0.79 to 0.15. Only SPY survives, and only as an observer. Session ORB (London/NY) orb_session + M15 · 08:30 London and 09:30 NY REFUTED. 3,782 trades on GBPUSD M15 primary / EURUSD validation. Zero of four cells pass. GROSS expectancy is statistically nil in all four (best t 1.40), so there is no edge for the 0.05-0.07R toll to eat. The filed measured-move target is only 0.52R from entry once the 1.32x entry overshoot is counted - it needs a 66% win rate to break even and gets 66%. Session Drift CONTROL session_drift + H1 · long at a fixed hour, no rule CONTROL, not a candidate. It goes long at 13:00 UTC and exits 12 bars later with no logic of any kind. It exists as the BENCHMARK a real strategy must beat. Breadth 94%, mean t +1.76 across the matrix - which is the point: a rule that cannot beat this is adding nothing over directional drift. VPA / SMC Confluence vpa_smc + FX M15 · structural SL/TP REFUTED unconditioned AND macro-conditioned. 1,761 trades on EURUSD/GBPUSD M15, six cells, using the engine's own structural levels. All six negative (-0.065 to -0.171R) with Welch negative in every cell - it loses to a bias-matched random entry. Conditioning on Fed policy epochs makes it worse in 5 of 6, and against 1,000 circular label rotations p runs 0.399-0.911. Sentinel Watchtower sentinel + Multi-detector · intraday & swing Negative expectancy in both modes; detector weights never fitted. FX Carry Filtered by D1 Trend (EMA50/200) fx_carry_trend + D1 · USDJPY, USDCHF · carry >= 0.50% and EMA50/200 trend · SL 2.5 ATR, TP1 3.75 ATR, EMA50 exit, 60 days FAILED on the 2012-2021 design years: 92 trades, net -$7,373 at $500 risk per trade, daily Sharpe -0.70 against a 0.70 gate. Lost to unfiltered carry and to random entry dates. Holdout not read. Backtests are where a module starts, not where it earns trust. Its forward signals - recorded before their results - are measured in the Copilot's Signal Lab and published on Telegram.