Every player here was tested once, sealed, and refused; that is why the active roster means something.

equity_pair_v3_orb (DRL) · retired 2026-07-08

SUPERSEDED_LEAK_LINEAGE — archived; live wiring hard-blocked

Trained on the pre-T7-fix observation build; the sealed backtest-era pooled Sharpes were leak-era artifacts and must not be quoted as evidence.

From the sealed disclosure: Policy trained on the pre-T7-fix observation build. Leaking channels recent_rrs + base_sharpe (future-dependent at decision time; pit_leak_adapter FAILED future_invariance on 140/140 sampled decisions) + residual day_pnl_pct (equity marked at entry vs sealed future exit). Sealed backtest-era Sharpes are leak-era training/eval artifacts and MUST NOT be quoted as evidence. NO live v3-ORB lineage exists.

equity_pair_v3_orb (DRL) · retired 2026-07-05

VOID_PREMISE_REFUTED — promoted 2026-07-04, voided by its own sealed premise-check the next day

The lone passing seed of 9 was promoted, then its PIT-clean retrain collapsed to zero — the edge was the residual leak, not the policy; fresh seeds replicated 1/10 and pre-declared deflation erased the lone pass.

From the sealed disclosure: ARM-A: the PIT-clean (causal_equity=True) retrain of seed 7 collapsed — per-instrument G3 0/492. Per the sealed rule 'A REFUTED -> s7 promotion is a correctable VOID regardless of B'. ARM-B: DOES NOT REPLICATE — 1/10 fresh seeds pass the frozen bar, below the sealed >=2/10 REPLICATES bar; best-of-10 DSR 0.213 (DEFLATED-AWAY). A registry honesty gap (verdict on disk 2026-07-05, entry updated 2026-07-08) is disclosed in the record.

retail playbook — break-and-retest · retired 2026-06-29

SUPPORTED=FALSE (sealed)

Anti-selective: 0/481 US500 names net-positive at cost and the entry picked worse spots than matched random — the edge, if any, lived in untestable discretionary level selection.

retail playbook — break-and-retest · retired 2026-06-30

NO-GO (smoke)

The 'better levels + no-trade-zone + confirmation candle' refinement stayed anti-selective — stricter selection did not rescue the engine.

retail playbook — order-flow · retired 2026-06-30

bar-proxy NO-GO; real order-flow construction UNTESTED (data locked)

The strategy's core is order-flow confirmation that bar data cannot express; the honest bar-only proxy was anti-selective — untestable is not refuted, and the gap was flagged up front.

retail playbook — ICT/SMC · retired 2026-06-30

NO-GO (smoke + bias/exit-package re-test)

Positive absolute Sharpe but 0/3 beat the matched-random control; restoring the taught bias and exits doubled performance and the bias-matched control absorbed all of it — the profit was the bias/exit package, never the ICT entries.

retail playbook — SMB scalp · retired 2026-06-30

NO-GO (smoke)

A VWAP-reversion scalp targeting a nearby reference has per-trade R at the spread, structurally — all 12 cells negative at cost.

retail playbook — SMB scalp · retired 2026-07-01

NO-GO (smoke)

Back$ide's twin: per-trade R below the spread — selection gates cannot fix a structural tiny-R-vs-spread problem.

retail playbook — SMB scalp · retired 2026-07-01

NO-GO (smoke)

Negative even at zero cost: the taught target sat a median ~22R away and the stop fired 82% of the time — the geometry, not costs, killed it; ironically its entry beat random.

retail playbook — SMB scalp · retired 2026-07-01

NO-GO (smoke)

The advertised 3:1 reward-to-risk was measured from the signal cross, not the fill — realized 1.30:1; always compute R:R from the fill.

retail playbook — SMB scalp · retired 2026-07-01

NO-GO (smoke)

Matched-random won 12/12 — the 1-minute consolidation-break engine is dead in both its dressings; no filter can rescue an engine the control beats every time.

retail playbook — SMB scalp · retired 2026-07-01

NO-GO (smoke; selection re-test closed-null)

The batch's only real entry signal still lost money: the taught reversion completed only ~19% of the time, and the sheet's own market-selection gate barely moved that while gutting the two cost-positive names.

retail playbook — ICT/SMC · retired 2026-06-29

SUPPORTED=FALSE (sealed)

0/3 index futures beat the matched-random control and the playbook's headline news-day gate had negative marginal value — the one taught state element made it worse.

retail playbook — ICT/SMC · retired 2026-06-30

sealed FALSE

0/10 beat the matched-random control despite being the batch's most faithful construction — nothing material was missing to restore.

retail playbook — black-box indicator · retired 2026-07-02

NO-GO (smoke screen)

A real-but-trivial micro-edge an order of magnitude under its cost; the methodology prize: re-resolving brackets on finer bars collapsed a coarse-bar 'keeper' — coarse bars mask tight-stop stop-outs.

retail video — stochastic scalp · retired 2026-07-04

NO-GO (smoke + rescue exhausted)

A genuine control-beating micro-signal whose breakeven cost is roughly a third of real cost; the celebrated 'go-to' divergence setup was a pure null, and moving up timeframes killed the signal.

book — mined price patterns · retired 2026-07-02

NO-GO (0/10 out-of-sample)

Mined patterns with high in-sample win rates on small samples all lost to matched-random out-of-sample — the author himself waived significance testing.

code project — price patterns · retired 2026-06-28

not a deployable edge (clean-room eval)

Pooled Sharpe was diversification, not alpha; the celebrated look-ahead was mostly a red herring — the dominant distortion was per-instrument in-sample optimization.

Sherman overlay · retired 2026-06-26

REFUTED on all four axes — do not re-attempt (two sealed attempts consumed)

A real per-name median improvement that was non-load-bearing everywhere it mattered: the gated book lost to the raw book and to random gating — broad shallow tilts are not deployable edges.

v2-ORB filter overlay · retired 2026-07-05

NO-GO on all three gates (unsealed smoke)

The filter had negative selection value — random subsets matched or beat it, under every exit model tested; the fourth converging line that the ORB primary carries no filterable edge.

Velez classical · retired 2026-06-11

RETIRED (verification INSUFFICIENT_DATA per the sealed binding; cost-stress FAIL confirmed)

The baseline pass never survived verification: the frozen methodology yielded zero decidable days, and the diagnostic confirmed cost-fragility below the retail tier; the taught 72% win claim measured 47.5% net.

Velez classical · retired 2026-06-10

NO-GO (cost via turnover)

The literal construction scanned every bar into millions of trades, so a positive gross Sharpe drowned at any cost — the exemplar of a missing frequency-discipline layer.

Velez state-conditioning · retired 2026-07-08

GATE_ADDS_NOTHING + SUBTRACTS (sealed)

The gate deleted most of the breadth while selecting nothing — below unconditional, below persistence-matched placebo relabelings, and the taught state ordering came out inverted.

factor timing · retired 2026-07-08

FAIL on the placebo gate (sealed)

A great-looking book whose rotation was statistically indistinguishable from random shifts — the Sharpe was the zoo, not the timing.

Tranche G mechanism arm · retired 2026-07-08

REAL_BUT_DECAYED (sealed)

A deflation-surviving full-window premium that is dead forward — the non-decay gate refused a fossil exactly as designed.

Tranche G mechanism arm · retired 2026-07-08

FAIL — inverted (sealed)

Cheap keeps losing under this construction too — the futures value family is dead under both constructions tried.

Tranche G mechanism arm · retired 2026-07-08

FAIL (sealed)

Hedging-pressure flow was empty at 13 weeks just as at 4 — it is a level signal only; the flow family is closed at both horizons.

Tranche G mechanism arm · retired 2026-07-08

FAIL — inverted (sealed)

The other-month reversal inverts — the program's anti-reversion finding extends to seasonal-complement levels.

Tranche G mechanism arm · retired 2026-07-08

FAIL (sealed)

Replicated the earlier COT-contrarian refutation through the commercial/noncommercial mirror, exactly as the sidecar pre-declared.

Tranche G mechanism arm · retired 2026-07-08

REAL_BUT_TOO_WEAK (sealed) — the tranche's only forward-positive object, still not promotable

Non-decaying and directionally real but too weak to clear the frozen gates on ~32 pairs — Tranche G closed 0/6 and the mechanism-class-targeting thesis was not supported.

AI-enhancement arm (H-DRL-BOOKALLOC-1) · retired 2026-07-08

ALLOC_ADDS_NOTHING_SUBTRACTS (sealed)

Every seed lost to the static book out-of-sample; huge train-vs-OOS gap = memorization, and the policy was worse than nearly all random same-intensity allocations.

AI-enhancement arm (H-ML-CONVCAP-1) · retired 2026-07-08

CONVICTION_NULL (sealed)

Keeping the higher-conviction half did not beat count-matched random thinning and lost to the full book — the model's rank already exhausts its information.

AI-enhancement arm (H-ML-UNC-1) · retired 2026-07-08

UNCERTAINTY_NULL (sealed)

Quantile-interval widths carried zero assignment information and the tilt class itself subtracts — the sealed inverse-vol book already carries the sizing information.

AI-enhancement arm (H-ML-NN-1) · retired 2026-07-08

TREES_WIN (sealed)

A Charter-quality loser: the MLP ensemble is itself gates-clean yet loses to the deterministic GBT in every tier-window cell — complexity did not add value at this data size, and single inits were forward-unstable.