Honesty doctrine. Every candidate anomaly is an artifact until proven otherwise; in-sample results are never findings; past statistical regularity does not imply future returns. This is research on statistical properties of market data — not investment advice, not a trading system.

Code / benchmarks/synthetic/README.md

benchmarks/synthetic/README.md 18 lines
---
project: anomaly-atlas
document: Synthetic series with known properties
author: Simon-Pierre Boucher
contact: contact@spboucher.ai
data_source: hfmarketdata.io
created: 2026-08-12
status: draft
---

# Synthetic test series — "test the tests" (CLAUDE.md §8.1)

Before any detector touches real data it must pass, on series generated here:
  * pure random walks -> NO anomaly detected;
  * planted mean-reversion / lead-lag / seasonal effects -> recovered;
  * pure bid-ask bounce -> flagged as ARTIFACT, not anomaly.
A detector that finds anomalies in a random walk is broken.