Evidence

Six domains, one protocol, full ledgers.

All results are founder-reported, single-seed in most domains, on public datasets with contamination caveats, and reproducible from the repository — not independently verified. That framing is binding on every artifact we publish.A6

DomainResultTrials
EUR/USD signalsSharpe +6.52 — raw, pre-deflation265+
QQQ signalsComposite Sharpe +1.32 · PSR 0.997216+
Higgs classificationAUROC ≈ 0.8675+—
PathMNIST OODAUC 0.997—
Olivetti facesConditional ARI 0.874—
Fraud detectionAUC 0.6097 vs AutoGluon 0.522 / H2O 0.518—
ABLATION

The rigor layer pays for its overhead

Removing the citation gate spiked invalid experiments 42% and produced 3 leakage incidents the gated arm did not. One task, one seed, founder-reported.

SCOPE

What we do not claim

Attestation packs, BYO-endpoint, fingerprint refusal and the acceptance gate are roadmapROADMAP — named as such everywhere, never sold as shipped.

A dense grid of faint cells with a scattered few glowing — kept results among the discarded.
Trial 218 — 9 kept, 209 discarded. The denominator every honest statistic needs.
The gap

The demand is proven. The supply walked away.

DEMAND

94,800 stars in six months

13,400 forks on the minimal loop — and no active maintainer since 2026-03-26.A1

SUPPLY

0 of 16 forks added rigor

Fourteen are hardware ports or translations; roughly one is a domain extension.A2

FIELD

294 papers, 17 fields

Documented data leakage in published, peer-reviewed work — the failure the gates exist to prevent.D5