title: Project Changelogs date: 2025-08-24 summary: Ongoing updates and notable changes across projects.

Changelogs

Track notable updates across active projects.

Kicked off a running log of Codex Orchestrator work for September 27, 2025.
Added a dedicated project entry outlining orchestration goals, current focus, and roadmap.
Continued pushing toward sturdier planner heuristics and richer multi-step integration tests.

Reverse‑engineered Colonist.io payloads/APIs and mirrored engine enums/types.
Crawler: cookie‑auth CLI crawls leaderboards → profiles → replays with concurrency; mode‑aware out paths under data/raw/colonist/{1v1,4v4}/<ts>.
Event engine + parser: typed mapped events, state‑change applier, and events‑from‑replay covering dice/robber (knight detection via logs), builds, bank trades (2:1/3:1/4:1), discards, steals, Year of Plenty, Monopoly.
Validator: replays mapped vs raw, compares core state, and writes board PNGs + state snapshots under data/validation_output/.
Env/action space: starter head [op, edge, node] with 1‑based IDs and masking; expansion planned for trades/devs.
Next: indexing maps, structured obs + legal masks, BC dataset shards, PPO self‑play.

Crawler: interactive/click CLI; supports 1v1 and 4p queues; per-user and per-replay concurrency; rate limiting; mode-specific replay folders (e.g., 1v1/<ts>/replays).
Normalizer: extracts initial_state, raw state_changes, final_state; convenience board, players (optional ranked-stats enrichment + cache), settings, result, and compact events_mapped (roads, setup placements, bank trades, discard, dev plays knight/road_building/year_of_plenty/monopoly).
Validator + engine: validator replays raw vs mapped events and compares final board + winner; reports diffs. Engine scaffolding (raw/mapped appliers + state snapshot) to support validation.
Docs: AGENTS updated with Crawl & Normalize usage, cookie setup, and output layout; noted example runs under data/raw/colonist/....

1v1 env skeleton integrated with PufferLib (MultiDiscrete [op, edge, node], turn masking, placeholder 2048-d obs).
Replay parser mapping Colonist logs to actions (build settlement/city, move robber, buy dev, steal), plus dataset builder script.
Pygame board renderer and sample state; smoke-tested short rollout via train.py with pufferlib vector env + default policy.
Next: legal action masks, roads/trades/dev plays, BC data pipeline, then PPO self-play.

2025-08-24: Initial entry. Outlined latent CoT approach and planned experiments.