ejaguiar1_stocks · hard first-touch TP/SL only, sentinels excluded, dedup-checked| Class | n (90d hard) | WR | PF | avg/trade | Verdict | Note |
|---|---|---|---|---|---|---|
| CRYPTO | 10,727 | 40.3% | 1.00 | +0.002% | COIN-FLIP | 135 strategies pooled ≈ breakeven. Class edge does not exist yet; only sleeves do. |
| EQUITY | 1,808 | 55.5% | 1.98 | +6.01% | UNTRUSTED | Inflated by frozen-exit resolver bug (QUBT/IONQ/PLTR clusters, 1 exit day × 180+ wins). Resolver fix deployed 21:26 today (305f9a79); historical rows NOT rewritten. |
| FUTURES | 2,420 | 48.3% | 1.32 | +0.27% | CANDIDATE | Thin but positive, no artifact clusters found. Needs n-growth + costs. |
| FOREX | 2,710 | 37.5% | 0.79 | −0.075% | LOSING | Dragged by copy-trader-intel contamination (99.8% of picks, 2.3% WR). Retail-contrarian sleeve (WF PASS, n=306) is the only exception. |
| ETF | 277 | 8.7% | 0.13 | −2.30% | TOXIC | Small-n + resolver artifacts. Do not size. |
| BOND | — | — | 0.62 | neg | LOSING | The "BOND PF 1.79" seen elsewhere comes from a replay-artifact file (MISPRICED_ENTRY/REPLAY rows), not real fills. |
| COMMODITY | — | — | — | — | REFUTED | copper HG=F "PROVEN" collapses: 751/815 signals overlap prior positions (peak 53 concurrent); one-position subset n=137 PF 1.21; last-30d WR 34% avg −0.90%. |
| MEMECOIN | — | 28% | 0.73 | neg | LOSING | mercury2 ensemble; dormant since 05-17. |
| Claim | Claimed | My hard-only re-measure | Verdict |
|---|---|---|---|
OpenCode: high_trust_momentum CRYPTO | n=206, PF 1.81, P=0.9997 | Oct WR 38.1%, avg −0.13% (decayed); certificate pools 79 soft TIME_EXITs; entry-anchored day-clustered CI crosses 0 before costs (GHCOPILOT) | REFUTED as proven |
OpenCode: commodity_term_cot HG=F | n=815, PF 2.29, P=1.0000 | 751/815 arrived with a prior position still open (peak 53× same-instrument LONG); one-position n=137 PF 1.21; last-30d hard WR 34% avg −0.90% | REFUTED |
OpenCode: luxalgo_confluence SHORT | n=404, PF 1.57, P=1.0 | n=1,484 hard: WR 51.1%, PF 1.39, exp +0.79%/trade, pooled CI [0.37,1.20] excludes 0 — BUT forward-60d CI [−0.15,+0.45] crosses 0 and WR-vs-coinflip p=0.41 | CANDIDATE — best pooled evidence in system, forward not yet proven |
Poolside/Kilocode: mega_mutation CRYPTO | n=171, p=9e-17 | Hard-only n=38: WR 52.6%, PF 1.73, CI [−0.51,+3.84] crosses 0 — but forward-60d WR 61.5% (only forward-positive sleeve in system) | CANDIDATE — sealed-tracked by me hourly |
| Poolside: CRYPTO alpha_engine cell (conf .75-.80/S50) | n=185, WR 75.1%, PF 3.90 | Pends resolver-fix re-measure (frozen-exit clusters live in that window) | UNVERIFIED — re-measure after fix |
| Kilocode cell | n=100, PF 2.75 | All from one month; n=60 & PF 0.75 without September (Copilot adversarial audit) | REFUTED |
AAPL vt_equity_two_day_rsi_reversal | n=13, WR 100% | Detector-clean (107 TP / 4 SL, positive Jun→Oct) but n_symbols=1 — single-ticker hypothesis, not a class edge | CANDIDATE (n=1 symbol) |
mega_mutation SHORT (CRYPTO): every new hard outcome is hashed + appended; pooled CI can hide decay, so the forward-60d CI is the headline. Status: n=38 seeded historical + forward accruing. Current honest verdict: NOT_PROVEN (1/4 gates). Also seeding luxalgo_confluence SHORT once the resolver-fix window accumulates outcomes.
Tools: tools/buffy_mega_mutation_paper_tracker.py · data: audit_dashboard/data/buffy_paper_trades.json
| Class | # | Candidate (claim source) | My re-measure | Verdict |
|---|---|---|---|---|
| CRYPTO | 1 | luxalgo_confluence SHORT (OpenCode #1) | n=1,484 hard, PF 1.39, pooled CI [0.37,1.20] — but fwd-60d CI [−0.15,+0.45] crosses 0; WR coin-flip p=0.41 | CANDIDATE (best in system) |
| 2 | mega_mutation SHORT (Poolside/Kilocode) | n=38 hard, PF 1.73, fwd-30d WR 61.5% — only forward-positive sleeve; CI crosses 0 (thin n) | CANDIDATE — sealed-tracked hourly | |
| 3 | prediction_market_consensus BNB (OpenCode #3) | artifact-flagged (482 wins / 80 exit days = the known BNB batch-close); fwd-30d exp +0.003%, CI [−0.32,+0.33] | REFUTED | |
| 4 | high_trust_momentum BTC (OpenCode #5) | n=31 hard only, artifact-flagged, fwd-30d n=4; family Oct WR 38.1% | REFUTED | |
| 5 | conf 0.75–0.80 LONG slice (Poolside + Zoo: “strongest statistically-vetted edge”) | n=4,114 hard: WR 39.3%, PF 1.17 — fwd-30d n=475 exp −0.487%, CI [−0.87,−0.09] entirely negative; deduped sym-day CI crosses 0 | REFUTED forward | |
| FOREX | 1 | regime_accumulation USDCAD (OpenCode #2, “PF 3.88→2.74”) | artifact: 44 wins on 5 exit days; hard-only n=80; fwd-30d exp −0.47%, CI [−0.53,−0.36] | REFUTED (artifact) |
| 2 | cta_fx_multifactor USDJPY (OpenCode #4) | artifact: 32 wins / 6 exit days; fwd-30d n=11 exp −0.30% | REFUTED (artifact) | |
| 3 | myfxbook_retail_contrarian (Kilocode) | repeated winning price pairs on closed weekends (GHCOPILOT D4 fail) | REFUTED (data) | |
| 4 | non_crypto_consensus FX (Kilocode) | same weekend-price exposure; class fwd-30d −0.31% | COIN-FLIP | |
| EQUITY | 1 | AAPL vt_equity_two_day_rsi_reversal | detector-clean (107 TP / 4 SL, positive Jun→Oct) but n_symbols=1 | CANDIDATE (single-ticker) |
| 2 | MRBB+200d / IBS+200d (Grok canvas) | OOS backtest only (n=487 PF 1.84 / n=775 PF 1.63); no DB forward rows; next-open entry untested | RESEARCH — seal forward before trusting | |
| 3 | class headline +7.6%/trade prior-60d | frozen-exit artifacts still live in pre-fix rows (fix landed 21:26 today, historical not rewritten) | UNTRUSTED | |
| ETF/BOND | 1 | Tactical Rotation TAA (Kilocode/Grok) | Sharpe 0.88 vs SPY 0.74, monthly, no per-trade ledger | RESEARCH (most defensible non-directional product) |
| 2 | AI-tournament BOND PF 1.79 (Kimi) | replay-artifact file (MISPRICED_ENTRY/REPLAY rows), not real fills; hard-only class PF 0.62 | REFUTED | |
| FUTURES / COMMODITY / MEMECOIN | FUTURES class PF 1.32 pooled but fwd-30d −0.75% (no sleeve passes); copper REFUTED (overlap n=137 PF 1.21 → negative after costs); mercury2 net loser | NO CREDIBLE CANDIDATE | ||
atr_percentile_gate_scanner — 7 dedup-clean per-symbol sleeves PF 1.4–2.2 looked like a hidden multi-symbol edge, but the family battery kills it: fwd-30d n=794 exp −0.46%, CI [−0.76,−0.16] — per-symbol slicing of a decaying family (survivorship). (2) Zoo's DOTUSDT (+282% of alpha_engine PnL) and EURGBP (74.9% of copytrader PnL) are concentration red flags, not edges — remove-the-top-symbol is exactly gate G2's purpose. (3) Orphan pages found (funds.html, strategy_perf_by_class.html, benchmark_proof.html …) contain no additional forward-proven edge.ccxt (Binance/Bybit depth + funding history) — needed for honest perp funding in cost model. EQUITY/ETF/BOND: yfinance total-return bars (dividends!) + pandas_market_calendars for session/exchange-calendar checks (kills weekend FX prints at the source). COMMODITY: zipline-reloaded or CME DataMine free tier for continuous front-month contracts with stated roll method. Cross-cutting: vectorbt for purged/CV embargo cross-validation (PBO), backtrader/freqtrade for independent re-implementation of any candidate (two-codebase replication is the cheapest defense against implementation-lookahead bugs). All free/open-source; no paid feed needed until a candidate is MONEY-READY.All five agents (OpenCode, Poolside/Kilocode, GHCOPILOT, INCEPTION/VLLM, FREEBUFF) converged on the same root cause: the frozen-exit resolver stamped one price snapshot onto every open pick → fake win clusters → every headline number inflated. Fix deployed today 21:26 (commit 305f9a79, per-pick price resolution). Consequence: all pre-fix forward numbers get re-measured; pooled certificates are suspect by default.
Quickest path to a true winner: (1) re-run the 4-gate battery on the post-fix ledger once ≥1 week accrues; (2) run the sealed cohorts on luxalgo SHORT + mega_mutation SHORT to n≥100; (3) only then label PROVEN-FORWARD. Nothing gets sized before that. Tactical rotation TAA (Sharpe 0.88 vs SPY 0.74) and the mutual-fund ranker are the only shippable non-directional products today.