Each file is a standalone episode outline tied to a git branch/tag and commit hash.
After each checkpoint commit, update the commit: line with git rev-parse --short HEAD.
| Episode | Branch | Commit | Script |
|---|---|---|---|
| 01 | checkpoint/01-path-bfs |
43a1b93 |
01-path-bfs.md |
| 02 | checkpoint/02-legal-moves |
19864b8 |
02-legal-moves.md |
| 03 | checkpoint/03-perft |
5a4b0fc |
03-perft.md |
| 04 | checkpoint/04-bench |
90193b0 |
04-bench.md |
| 05 | checkpoint/05-perft-bugfix |
6b9e00d |
05-first-perft-bug.md |
| 06 | checkpoint/06-threading |
098477c |
06-threading-prep.md |
| 07 | checkpoint/07-gorisanson-ui |
7c85a20 |
07-ai-opponents.md |
| 08 | checkpoint/08-greedy-ui |
10bcb23 |
08-greedy-ui-lab.md |
| 09 | (future) | — | 09-pondering-prep.md — Stockfish-style ponder (prep only) |
| 10 | checkpoint/10-hybrid-search |
(pending) | 10-hybrid-search.md — MCTS↔minimax, LMR, Ka bench, terminal UI |
| 11 | (future) | — | 11-search-hardening.md — negamax bugs, CAT v3 gaps, qsearch, CAT UI |
Perft debug: node benchmark/perft_diff.mjs 2 — divide diff vs JS oracle.
Perft regression (Rust): cargo test --release perft_depth4 -- --ignored --nocapture — timed d1..d4, ~3.5s total on idle CPU.
Outdated: "perft 4 takes minutes" — release d4 is ~3.4s (see PERFT-BENCHMARKS.md).
CAT overlay: CAT-VIEW-UI.md · spec CAT-SPEC.md · wall shape CAT-v3-wall-shape.md
Future Elo ladder: TOURNAMENT-ROADMAP.md · machine list: benchmark/checkpoints.json
Series index: 00-SERIES-OVERVIEW.md · BUG-DIARY.md · PERFT-BENCHMARKS.md · 00-HOW-THE-ENGINE-WORKS.md
Checkout any episode: git checkout checkpoint/03-perft