Anchor Forest, an archive of failed quant hypotheses

Why it surfaced

Anchor Forest makes a case for documenting what fails: it pre-registers evaluation criteria, records reversals, and publishes concrete postmortems. Recent entries include a benchmark choice that flipped reported alpha from −31.9% to +7.6%, a trading-day bug that silently erased 22.8% of signals, and a three-stage LLM classifier that processed 104,659 items for $125.66.

A Korean research journal about testing stock-market signals, data pipelines, automation, and local-LLM systems.