# project: anomaly-atlas document: F001 confidence author: Simon-Pierre Boucher contact: contact@spboucher.ai data_source: hfmarketdata.io created: 2026-08-12 status: reviewed

Level: 2

# Confidence — F001 (negative finding)

Level 2 — robust (corrected & OOS & robust). Evidence per layer:

• Corrected: the claim is made against the FULL searched universe (372 signed rules), with BH-FDR + White RC + Hansen SPA (500 stationary bootstraps) + DSR; the 22-hypothesis budget was declared before scanning (research_gaps.md) and never expanded.
• Artifact-adjusted: scans ran net of measured nulls (taxonomy T1-T7: bounce/EDGE variance-consistent MA(1) null, both-fresh synchronization, permuted calendar); detectors passed 29 synthetic gate tests first.
• Out-of-sample: the 2016-2021 validation split, frozen before any scan and opened exactly once (expH), REPLICATES the negative: zero net survivors at full half-spread; the lone 0.25x survivor fails its pre-registered sub-period clause; ES->SPY flips sign; the daily reversal family collapses (-0.143 -> +0.023 VR30 excess).
• Robust: holds across 4 timeframes, 2 train sub-periods, 2 validation sub-periods, liquidity buckets, futures splice variants, and both close conventions where applicable.

Why not Level 3: Level 3 is reserved for confirmation on the sealed holdout (2022->), which this project deliberately did not spend on an already-settled negative. Costs are estimated (EDGE), not quoted — the cost-real qualifier is model-based.