Verdict: benign (by construction) — screened 2026-08 on the public panel
(Accruals, AssetGrowth, BM, vintages 200401–202610, retrieved via
openassetpricing).
0 of 18 predictor-cycles flagged; max |ρ̂| = 0.0357 against a 0.10 threshold — an order of magnitude of headroom.
OSAP’s uniform annual update calendar (a 6-month datadate lag, refreshed for all firms at once) removes filing speed from arrival ordering: completeness steps from roughly 7% to roughly 88% in a single June update. When everyone arrives together, arrival order cannot select on the disturbance — the channel this screen measures is closed by the dataset’s own design, and the measurement confirms it.
The same protocol run on data with no such calendar — a panel rebuilt from as-filed SEC EDGAR filings — flags 7 of 14 standard signals. The protection is real, and it belongs to the publication calendar.
The protection does not travel. Rebuilding OSAP-style signals directly from raw EDGAR at as-filed dates loses the calendar and re-opens the channel. A verdict is a finding about a specific version on a specific date, not a property of the signal definitions.
pip install openassetpricing pit-release-gate numpy pandas scipy matplotlib jupyter
git clone https://github.com/MaxWellApexLab/pit-audit-registry
jupyter lab pit-audit-registry/audits/2026-08_osap/osap_predictor_completeness_screen.ipynb
Free public OSAP release; no WRDS, no licence. ~2 minutes, ~25 s of it the download. The report includes a planted-truth positive control (4/4): the same screen fires at selection strengths far below what it would take to matter here.