# RIDGE Open Benchmark

This directory holds the public evaluation dataset for Amazon FBA niche
outcome prediction.

## Files

  * `niches.jsonl` — 169 niches, each with keyword, category, top ASINs,
    and observed outcome (ALIVE / THRIVING / DEAD). Use this as your
    test set.
  * `v7_predictions.jsonl` — RIDGE v7 predictions on the same set, with
    out-of-fold score, 90%-coverage conformal prediction set, and binary
    correctness flag.
  * `leaderboard.md` — current industry leaderboard (RIDGE is the only
    entry until a competitor submits).

## How we labeled the 169 niches

  * 2022-2023: top-2 ASINs per niche recorded at entry window.
  * 2024-2026: periodic probes measure whether ASINs remain in-stock,
    gain reviews, hold BSR, and whether competing listings in the niche
    survive. Classification rules published at
    https://ridgeworldwide.com/pages/research/calibration-audit.html

## Reproducibility

  * Raw scoring code: `backtest/phase22_benchmark_export.py` in the
    RIDGE repo (private for now; extract upon request).
  * Feature definitions: `backtest/phase3_ml.py` in the same repo.