| name | reproduce-result |
| description | Reproduce (or refute) a published verdict from a clean clone at $0. Fetches the free public data, runs the per-domain harness, and compares to the recorded SUMMARY. Use when asked to reproduce, verify, or check a result. |
Skill: reproduce-result
Confirm — or refute — any verdict in output/edgehunt-*/SUMMARY.md from a clean clone, at $0.
Procedure
- Pick the target. Find the hypothesis ID and its scripts in
scripts/edgehunt-<domain>/ and its recorded
verdict in output/edgehunt-<domain>/SUMMARY.md.
- Regenerate the data. Raw caches under
output/ are gitignored. Run the domain fetcher (e.g.
npx tsx scripts/edgehunt-D2/fetch-data.ts, scripts/edgehunt-D5/fetch_extra.ts) to download the free
public data (Binance/Bybit/OKX REST, Coin Metrics Community, Deribit, FRED, DefiLlama, …). Some inputs (e.g.
long 15-minute history) note their source in docs/REPRODUCIBILITY.md.
- Run the harness.
npx tsx scripts/edgehunt-<domain>/<script>.ts. Inspect the printed VERDICT: line and
the JSON it writes under output/edgehunt-<domain>/.
- Compare. Does your net Sharpe / binding gate / surrogate p / verdict match the recorded SUMMARY within
noise? If it disagrees, that is exactly what this repo is for — open an issue with your code and numbers.
- Sanity-check the gates first.
npm test must be green (the gate primitives' own unit tests) before any
reproduction is trustworthy.
Notes
- Everything is $0 / free public data; no keys for the core path. If a fetch needs a paid source, the script
marks the result DEFERRED and tests the best free proxy instead — reproduce the proxy, not the paid metric.
- Costs (taker ~4 bps/side, financing on full notional) are baked into the harnesses; do not silently relax them.