The graveyard

1,000+ tests. 2 shipped.

Across 17 ML paradigms, 50+ named sprints, and roughly a thousand hyperparameter configs since 2024, two strategies cleared every audit gate (R86b · R99b). The rest live here — categories, then 21 recent rejections.

1,000+
tests run
50+
sprints
17
ML paradigms
250+
event tests
2
shipped

By research direction

Each row is a paradigm we systematically tested — full post-mortems live in our internal experiments log.

A

ML overlay (meta-learning)

5 tested·0 wins

Add a model on top of existing strategies — pick winners, scale by confidence, skip losers. All five attempts hurt Sharpe via diversification loss.

B

Direct ML direction prediction

4 tested·0 wins

Predict next-bar direction with LightGBM / XGBoost on price + cross-asset features. AUC ceiling 0.50-0.51 — no extractable signal at hourly resolution.

C

Position sizing & filter schemes

5 tested·0 wins

Adjust position size or skip trades by volatility, regime, or time-of-day. Vol-target had 60 sub-configs; one passed but overfit a single hyperparameter.

D

Statistical arbitrage

1 tested·0 wins

Trade the spread between cointegrated assets reverting to long-run mean. XAU/XAG not cointegrated (p=0.26), half-life 142 days — untradeable.

E

Alternative data

2 tested·0 wins

Non-price signals (news, sentiment, social) for direction. GDELT raw + smart variants: no |IC| > 0.05 stable in both IS and OOS.

F

Volatility prediction

2 tested·0 wins

Predict next-period vol and scale exposure. Vol IS predictable (Spearman 0.17) — but unexploitable on retail XAU spot without options access.

G

HMM regime classifier

1 tested·0 wins

3-to-5-state Gaussian Hidden Markov Models to detect trend/chop/crisis. Train +3.8 → OOS +0.18 collapse. Overfits on 75 train trades.

H

COT sentiment filter

1 tested·0 wins

CFTC Commitment-of-Traders speculator positioning as entry gate. Pre-2025 pattern broke post-2025 — random null P=49.6%.

I

Sprint D LightGBM grids

35 tested·0 wins

Direction prediction on raw bars across 35 hyperparameter configs. Ceiling -0.5 Sharpe. Inversely-predictive in some folds.

J

Liquidity-grab / microstructure

1 tested·0 wins

Trade liquidity hunts at swing highs/lows. Sharpe -0.97 at hourly resolution. Needs Level-2 / tick data — retail-impossible.

K

Event-window search

250 tested·1 wins

Brute-force test every economic event × horizon combination. ~1000 tests with strict Bonferroni; only 42-Day Bill Auction survived (t=4.74).

21 recent sprint-level rejections

Detailed rejections from the 2026 sprint cycles, with the audit reason that killed each one.

May 2026 · 3 rejected
R111v318
Armed ratchet exitno-edge
OOS Sharpe +0.31 — below ship threshold +0.7
R111v212
Trailing ratchet (vanilla)overfit
IS +1.4 → OOS +0.2 after walk-forward purge
R110a05
MFE-trip exit variantoverfit
IS +1.6 → OOS +0.4 — replaced by R110b
Apr 2026 · 4 rejected
R10928
5m event-window predictorcapacity
Slippage assumption broke at 5m timeframe
R10821
BB squeeze v2 + regime gateregime
Failed cross-year stability check (3/5 folds negative)
R10714
XAG/XAU pair-tradeno-edge
Cointegration breakdown post-2020
R10607
COT-only signal generatorno-edge
Negative IS Sharpe (-0.4) · failed precommit
Mar 2026 · 4 rejected
R10531
Volatility regime classifieroverfit
k-fold instability — 3/5 folds negative OOS
R10424
Multi-symbol momentumcapacity
BTC capacity insufficient post-ETF inflow
R10317
CPI release follow-throughno-edge
Edge gone after 2020 — already arbitraged
R10210
NFP gap fader (5m)look-ahead
Volume features leaked next-bar information
Feb 2026 · 4 rejected
R10128
Forward-vol breakoutno-edge
Negative carry after R83 deflation
R10021
Implied-RV crossoveroverfit
Walk-forward 2/5 folds passed — under threshold
R99a14
COT capitulation (vanilla)regime
Era-blind logic fired in wrong regime · replaced by R99b
R9807
XAU daily breakout (gc) v2overfit
Parameter sweep failed Bonferroni correction
Jan 2026 · 6 rejected
R9731
Cross-asset RSI divergenceno-edge
OOS Sharpe -0.12 — IS gain was noise
R9624
Risk-off bond-equity rotationcapacity
Insufficient sample size at 1d timeframe
R9517
Cluster mean-reversion (XAU)regime
Cluster instability across market regimes
R9410
GARCH-vol breakout filteroverfit
Filter killed more signal than noise — net negative
R9308
Sentiment-weighted entrylook-ahead
News sentiment timestamp leaked future info
R8605
Wide event window (vanilla)no-edge
Replaced by R86b tighter window — shipped

Killed sprints stay published forever — both as a discipline forcing function and a public record of what we tried.