Coil open data
Two datasets with a property almost no market data has: they were committed before the outcomes existed, and they cannot be revised.
1. The committed daily stock-score archive
Every trading day, Coil's scanner scores ~560 names — the S&P 500, the Nasdaq-100, and a macro book — on three axes: opp_pct (opportunity, 0–100), entry_q (entry-window quality), and hold_q (hold strength). At publish time, that day's complete payload is committed as a sha256 hash in an append-only public log — before the session's outcome is known. Nothing is ever rewritten; corrections append and disclose what they supersede.
That makes this archive the rare thing in market data: provably not back-fitted. The verification recipe (canonical JSON: keys sorted, no whitespace, integral floats without a decimal point) reproduces every committed hash byte-identically in Python and JavaScript.
- Free sample day, complete:
/api/board/asof?sample=1— 563 names with full scores and the committed hash to verify against. Diff it against your own scanner. - The commitment chain:
/api/board/proof— every hash ever committed, the recipe, and a free inline verification sample. - Any other date:
/api/board/asof?date=YYYY-MM-DD— $0.02 per read over x402, no account.
2. The signal forward-return audit
The audit answers the question the archive alone cannot: do the scores predict anything? For every committed day, it computes forward 1- and 5-session returns of the top-decile and entry-ready cohorts against two controls — the board median, and a basket of the decile's own sector ETFs weighted by its sector mix. The methodology was frozen before the first number was computed, and it publishes daily even when the answer is unflattering — right now the sample is thin and roughly flat, and the dataset says so with the sample size in the same sentence.
- The full audit, free:
/api/board/signal-audit— per-day cohort returns, controls, excess, hit rate, and the methodology text.
Access and terms
Both datasets are JSON over plain HTTPS. The free tiers above need no key, wallet, or account. Live and historical paid reads run over the x402 agent lane from $0.005. Data is provided for research under our terms; it is scores and states only — never stop prices, targets, or investment advice. Cite as "Coil (coil.trade), committed daily stock-score archive," and verify any date you cite against /api/board/proof — that is what it is for.
Why give this away? Because the archive's value is its verifiability. A dataset nobody can check is marketing; one that anyone can hash-verify, sample free, and audit for predictive power is infrastructure. The paid products sit on top of the same data; the proof layer is free forever.