Failures stay in the record.
The public ledger tracks the progression from causal data infrastructure to preregistered action-side treatments. Closed studies remain visible; no failed lineage is rewritten as a success.
From market data to frozen selection rules.
Only results already inside the scientific record are shown. Current Fresh candidate performance is deliberately absent.
EXP-0003 — MarketGenome foundation
Multi-timeframe BTCUSDT causal dataset, dataset validation and normalized observation architecture. The foundation dataset established 475 columns with 468 numeric market features before later RL state extensions.
EXP-0004 — Discrete PPO validation
LONG / FLAT / SHORT PPO iteration. Validation did not produce an eligible candidate under its frozen constraints; Historical TEST remained locked.
EXP-0005 — Futures-native RL stack
Introduced futures-native execution semantics: mark-to-market equity, historical funding, leverage/margin mechanics, next-bar execution and deterministic risk ordering. This became part of the later parent stack.
TURNOVER-CONTROL-STUDY-0001 — cooldown study
Three preregistered cooldown arms tested whether reducing churn was sufficient. The frozen result was NO_ELIGIBLE_TURNOVER_CONTROL: 0 / 30 eligible, no arm winner, no champion, Historical TEST not accessed.
ENTRY-CONFIRMATION-STUDY-0001 — current lineage
Tests whether requiring 1, 2 or 3 consecutive identical directional requests before a new entry improves selectivity without changing economics, risk limits, PPO hyperparameters or the 30-minute policy cadence.
Cooldown reduced churn — not enough.
These figures are frozen development evidence from the predecessor and have no selection authority in the current ENTRY lineage.
| Arm | Treatment | Risk termination | Mean return | Median return | Mean turnover |
|---|---|---|---|---|---|
| EXP-0009 | 0m CONTROL | 58.333% | -8.468% | -9.120% | 122.8805x |
| EXP-0010 | 60m cooldown | 12.500% | -4.684% | -5.552% | 67.4022x |
| EXP-0011 | 120m cooldown | 7.500% | -4.548% | -4.633% | 60.0498x |