[eaitbrahim]

Tests an outside claim that the last two digits of PAXG's price on the starting candle predict the next 4 hours' direction.

• Pre-registered: the driver was pushed in 30dbe5b before it ran on real data.
• Result: no arm shows dependence (Bonferroni alpha 0.0167).
  • In-sample permutation p: 0.097 (Coinbase dollar digits), 0.233 (Coinbase cents), 0.737 (Binance, 5 years).
  • Out-of-sample hit rate: 48.9–49.2%, no better than shuffled-label nulls (OOS p 0.39–0.80).
• Economics: the median 4-hour move is 0.20%, against a 1.8% taker round trip. The net return per "up" call is −1.80% on every arm.
• Ledger: three diagnostic_only rows appended to the trials ledger; verify_chain is clean.

Files: driver .py, record .md, data .jsonl, and the README index entry.

🤖 Generated with Claude Code