Breaking the Selection Link: A Pre-Registered Validation Audit of the Factor Zoo

This note implements the independent-validation remedy of López de Prado and Fabozzi (2026, SSRN 6450418) under internal pre-registration. A per-factor validation design (publication-year and sample-end gates, Newey-West t statistics, BH(0.10) certification, four-bucket ledger) was frozen on 2026-07-24, with quantitative predictions committed under three worlds calibrated from the published cross-section: Chen-Zimmermann (6.3% false, 88% retention), McLean-Pontiff (all true, 42% retention), and the López de Prado-Fabozzi search-adjusted calibration (80% false). On the 212 Chen-Zimmermann predictors (data through 2024-12), 64 of 210 tested factors certify post-publication; on the 142 JKP-USA capped-value-weight factors (through 2025-12), 3 certify. Every committed certified-count interval misses, at every certification level swept, on both datasets. The misses are asymmetric: certified counts fall far below the optimistic calibrations, while sign inversions fall far below the 80-percent-false calibration, implying a null share near 39% (CZ) and 22% (JKP). The validation data reject both poles of the debate. The supplementary file pilot_archive.zip contains the complete evidence base: the frozen committed-prediction JSONs and seeded simulators reproducing every committed cell exactly (seed 20260724), per-factor scorecards for both gates and both datasets, run logs, input hashes and source pointers (raw returns data are not redistributed; obtain them from openassetpricing.com and jkpfactors.com under their public research terms), environment stamp, figures, the exact code executed, and PROVENANCE.md describing the pre-registration chain and reproduction steps. Companion papers: - Certified Discovery on Real Grams: An Instrument, its Limits, and its Laws (10.5281/zenodo.21819895)- Dark-Mass Metrology: Pricing What a Certified Model Does Not Know (10.5281/zenodo.21819925). Instrument: The provable package v0.9.0 (10.5281/zenodo.21819947; PyPI: provable; github.com/provable-ml/provable).