PIT Audit Registry
Reproducible findings on whether public datasets and pipelines are exposed to incomplete-cross-section leakage — the look-ahead risk that arises when a cross-sectional quantity is computed before the group has finished reporting, and reporting timing is related to the quantity being measured.
Every entry is a measurement with a command attached.
Verdicts
- Is the Open Source Asset Pricing (Chen–Zimmermann) dataset point-in-time safe?
—
benign (by construction): 0 of 18 predictor-cycles flagged, max |ρ̂| = 0.0357. - Does JKP Global Factor Data have look-ahead bias from staggered filings?
—
benign (by construction): a uniform four-month availability convention closes the channel structurally. - Is as-filed SEC EDGAR data safe for cross-sectional signals? — 7 of 14 standard signals flagged (28/84 signal-cycles). ROA is clean at 0/6.
- Do staggered SEC Form 13F filings bias cross-manager statistics? — flagged in 11 of 11 quarters, at ten times this panel’s measured noise floor, on filing dates that are observed rather than reconstructed.
The pattern across every entry so far: a publication calendar is doing real protective work, and the protection does not travel. Curated releases with a uniform update calendar come out clean; data taken at the dates it actually arrived — as-filed EDGAR filings, or 13F filings inside their 45-day window — does not.
Method, in one paragraph
The susceptibility statistic ρ̂ is the partial correlation between an entity’s filing latency and the complete-cross-section residual of the signal, given observables — fitted on prior completed periods only, frozen, then applied forward (threshold 0.10). Every published screen includes a planted-truth positive control: a report where the screen never fires is not published unless the screen has been shown to fire on data whose answer is known by construction. Method papers: measurement · release control.
Reproduce
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 data only; no accounts, no licences. ~2 minutes.
Reports and registry: github.com/MaxWellApexLab/pit-audit-registry · Screen implementation: pit-release-gate (MIT) · Maintained by Max Well Apex LLC · Reports CC BY 4.0