[nosam1998]

🤖 I have created a release beep boop

1.0.0 (2026-10-04)

Features

• add atomic PostgreSQL match storage (e6312f8)
• add bounded built-in search players (4ce7184)
• add bounded observation-only random minimax and MCTS players (9149afb)
• add bounded simulation with recorded assist policy (0df5dc4)
• add durable resignation and draw controls (c96579b)
• add engine-authored play guides (16b9e88)
• add engine-authored play guides and conformance checks (864643f)
• add event-sourced match lifecycle (#9) (da84e99)
• add exact-cover Sudoku solver and seeded unique puzzle generation (c9940a7)
• add match API and storage contracts (#8) (9e08dea)
• add match metadata and paginated history (a68eba6)
• add opponent seats with atomic replies and fork overrides (cc8bb3c)
• add PostgreSQL persistence and database-backed CI (828c71c)
• add single-player Sudoku engine with notes and mistake policies (8613d48)
• add Sudoku solver and unique puzzle generation (141fce0)
• add Tic-Tac-Toe and engine conformance tests (7864b31)
• add Tic-Tac-Toe and engine conformance tests (#4) (7864b31)
• add Tic-Tac-Toe and reusable engine conformance checks (6632eee)
• add validated Connect Four engine (b13bf0a)
• add validated Connect Four engine (#5) (b13bf0a)
• add validated Connect Four game plugin (d70843e)
• add viewer-scoped event history with stable pagination (f85dde6)
• add viewer-scoped match briefings (86154ca)
• analysis: expose ranked MultiPV recommendations (6216878)
• analysis: return bounded ranked recommendations through shared workers (66ca46b)
• bundle interactive API documentation (ff4785a)
• bundle interactive API documentation and browser tests (f5c4fd6)
• check SQLite queries at compile time against migrations (f87d7d4)
• chess: add standard rules engine with draw history (6bd68c5)
• chess: enable chess through the shared API (29d098e)
• chess: implement standard rules and replayable draw claims (aef76ab)
• chess: register engine and verify API lifecycle (7d26e92)
• cli: evaluate model snapshots and export tournament data (b8590ce)
• cli: evaluate own models from YAML configurations (dfb4e1f)
• cli: host MCP over stdio with durable games and graceful shutdown (33f94e0)
• cli: launch local MCP over stdio (c60e87e)
• connect4: expose board and turn tensors for training (a625a64)
• connect4: expose numeric training observations (a74ee67)
• core: add native training environments and checkpoints (f54ab01)
• core: add native training environments and exact checkpoints (eb92251)
• core: construct native training environments by game id (d96564c)
• core: register object-safe native training environments (6fb0653)
• define match API and event storage contracts (9e08dea)
• define match API types and atomic storage contracts (269478c)
• define pure engine contracts and crate boundaries (b7b770b)
• define pure engine contracts and crate boundaries (#3) (b7b770b)
• define pure engine contracts and enforce crate boundaries (7adb636)
• define simulation requests assist policy and usage accounting (5706573)
• expose bounded match history and public metadata (1b7b042)
• expose bounded opponent analysis and game-specific catalogs (f5fb57a)
• expose local REST match lifecycle (c0d6bf0)
• expose local REST routes for match lifecycle (1f8f49f)
• expose observation-scoped engine analysis and opponent catalogs (0df0ad1)
• fork matches with reproducible variation history (4e214c1)
• generate OpenAPI contracts from shared Rust types (ba85435)
• generate OpenAPI from shared Rust types (e087353)
• generate seeded Sudoku puzzles at verified technique grades (31f957e)
• generate Sudoku puzzles by verified difficulty (3ed0ca1)
• grade Sudoku by logical techniques and explain hints (b1b944f)
• grade Sudoku difficulty and explain logical hints (3dd7ddb)
• implement event-sourced match lifecycle and replay (0e8a6a5)
• implement match lifecycle with deterministic replay (da84e99)
• include events and authorized reconstruction inputs in replay bundles (12985e7)
• include recorded events and authorized inputs in replay bundles (a22caa5)
• include viewer-scoped match briefings in creation and info routes (67c54dc)
• llm: add bounded Anthropic Messages transport (#103) (4398aac)
• llm: add bounded direct and tool turn runner (#107) (6025922)
• llm: add provider-neutral player contracts and accounting (46e445a)
• llm: define provider contracts and bounded player configuration (5966a44)
• llm: persist provider calls and shared spending reservations (#108) (b104347)
• make Sudoku playable through the shared game API (948804c)
• mcp: add bounded live resource subscriptions (d687599)
• mcp: add compact match controls and actionable move errors (2e2c850)
• mcp: add compact match controls and automatic replies (4a925aa)
• mcp: add game resources and play prompt (e71775c)
• mcp: add hypothetical simulation and engine analysis (30cba60)
• mcp: add paginated match history and replay (ce5139f)
• mcp: add typed discovery tools over the shared service (75bd92c)
• mcp: expose bounded simulation and analysis with assist policies (2908c78)
• mcp: expose game resources and play prompts (ebb82b4)
• mcp: expose paginated history and viewer-scoped replay (ed8e3e5)
• mcp: expose typed game discovery through the official Rust SDK (1ef4f19)
• mcp: notify clients when watched match state changes (7148b41)
• persist authorized match forks and immutable replay ancestry (a6437da)
• persist match events and idempotency receipts in SQLite (7d32760)
• persist match events and retry receipts in SQLite (6ff41bc)
• persist match events in SQLite (#10) (7d32760)
• persist opponent seats and commit automatic replies atomically (b4396a5)
• persist resignation and draw agreement controls (4e885f7)
• players: distinguish provider failures from illegal moves (a0ba77a)
• players: preserve operational failures across the service boundary (952a367)
• python: add atomic parallel environment batches (0f539a0)
• python: add atomic remote vector training (3b55ea1)
• python: add native search opponents for training (0b75c79)
• python: add parallel vector environments with NumPy batches (739dcd1)
• python: add PettingZoo AEC self-play (0d1ac99)
• python: batch remote vector environments atomically (bcc90a1)
• python: build native training wheels with NumPy observations (b92d524)
• python: checkpoint AEC turns rewards and curriculum RNG (9a75843)
• python: connect Gym and AEC training to stateless REST (87fc3ac)
• python: connect local policies to remote environments (dbaf58a)
• python: connect self-hosted chat models as policies (4d78d2b)
• python: expose native environments and owned NumPy frames (fa4098d)
• python: expose PettingZoo AEC self-play environments (d7c3b5e)
• python: expose readable legal actions for model policies (dc0f480)
• python: expose readable legal actions to model policies (a3c5364)
• python: integrate self-hosted chat model policies (dda6817)
• python: support Gymnasium training with user policies (e91095a)
• python: train against seeded native search opponents (50707f4)
• python: wrap native games as Gymnasium environments (57a8be5)
• ratings: add Glicko-2 periods and fixed anchors (3c0932e)
• ratings: implement bounded Glicko-2 period updates (f14cc37)
• resume automatic matches through a bounded background runner (9f30f16)
• run and resume automatic matches with bounded background workers (7cece7d)
• run bounded planning lines with assist policy checks (9d16955)
• run local games through the CLI and SQLite server (d64d463)
• scaffold modular React and Phaser workspace (5847368)
• scaffold modular React and Phaser workspace (ba1af3d)
• scaffold modular React and Phaser workspace (#6) (5847368)
• serve deterministic game briefings with executable examples (1a2e08d)
• serve game briefings with executable examples (598e93c)
• serve paginated event history with explicit viewer projection (5796e7a)
• server: expose installed Stockfish through opponents and analysis (4f2c290)
• server: install sandboxed Stockfish for opponent seats and analysis (5b467bf)
• server: mount streamable MCP at /mcp (c58f89f)
• server: select PostgreSQL through an environment-backed CLI option (9bd2ef9)
• server: serve streamable MCP with shared local access policy (d832c48)
• server: support PostgreSQL-backed local matches (e1b19a2)
• stream committed match states over local WebSockets (80c439a)
• stream committed match states over WebSockets (8eb0d90)
• sudoku: add explained reference-solver recommendations (da1fec2)
• sudoku: expose explained reference hints through the generic analysis API (d6ecea4)
• training: add executable masked PPO training and evaluation (ef31b2e)
• training: add hash-pinned validated position datasets (7a09d30)
• training: add reproducible vector curricula (9887bbd)
• training: add stateless batch REST environments (a4beaef)
• training: add Stockfish Gym opponents (be77957)
• training: add validated versioned position sets (0131e1b)
• training: avoid text and JSON in numeric batch observations (7977bcb)
• training: compare model snapshots in balanced round robins (5b508dd)
• training: execute bounded stateless environment batches (6ea0fef)
• training: export typed Parquet trajectories (91e03b1)
• training: expose sandboxed Stockfish to Python (939da5c)
• training: expose sandboxed UCI pool to Python (890e903)
• training: expose shared Glicko-2 ratings to Python (814c02c)
• training: project numeric tensors without text or JSON (0df4715)
• training: rate local model snapshots with Glicko-2 (bd91f98)
• training: record local model trajectories as JSONL (743e4ad)
• training: record seat-visible trajectories and JSONL exports (18defc8)
• training: resume AEC self-play checkpoints (f276ed4)
• training: run balanced model snapshot tournaments (63d097c)
• training: run selected model and engine matchups (42b7c0c)
• training: sample pinned curricula in vector batches (ce09756)
• training: sample pinned position curricula on reset (8c6e65b)
• training: sample reproducible position curricula in Gymnasium and AEC (8b72c97)
• training: select opponent matchups and include Stockfish pools (6d23432)
• training: stream typed Parquet trajectory datasets (bb93b17)
• training: support Stockfish Gym opponents and pinned identities (3d87140)
• training: train and evaluate a masked PPO policy (c5e6e03)
• uci: add validated commands and bounded response parsing (b90ff39)
• uci: pool sandboxed engines with bounded pipes and hard deadlines (9276269)
• uci: supervise a bounded pool of sandboxed engine processes (71cc414)
• uci: validate engine decisions against authorized game observations (a63507a)
• uci: validate observation-scoped engine decisions (d37897b)
• uci: validate protocol input and parse bounded engine output (c467132)
• validate and reseed standalone game positions (8ea6119)
• validate and reseed standalone positions (7b4941b)
• web: add interactive boards for all four MVP games (#99) (b7b9474)
• web: add local evaluation report viewer (#105) (1eaa50e)
• web: add match history and live spectator arena (#100) (4c86871)
• web: add match setup and accessible live play (#98) (ea115cf)
• web: add replay timeline and branching (#102) (9b0b489)
• web: add saved models and evaluation setup (#104) (7508406)
• web: add the game library and play guides (99d5869)
• web: build the live game library and play guides (707f563)
• web: compare evaluation runs by opponent (#106) (6ba524b)
• web: generate API DTOs and add match client (5052ef1)
• web: generate browser contracts and add match API client (1c2a630)
• wire local server and CLI to SQLite and game registry (4aa8d6a)

Bug Fixes

• api: retain the standardized automatic-action response (0ba7921)
• api: use the PRD opponent_actions response field (d996b83)
• bound append result size and use standard test waker (f296759)
• bound compact examples and protect puzzle solution metadata (d6c9d2d)
• box stored responses in append results (352c334)
• briefings: bound numeric previews and provide concise chess rules (d962502)
• briefings: keep chess rules and examples within prompt budgets (73e997c)
• chess: align tests with shakmaty and lock dependencies (8eb1e8f)
• chess: document license and reject out-of-range encodings (5ae3000)
• chess: match JSON wire projections and fork response shape (b1593ce)
• chess: use bounded briefing projections (61ac60d)
• cli: compose stdio service without an HTTP listener (0764116)
• cli: share the database selection guard across both backends (0b1457a)
• commit checked SQLite history metadata and format (e66bbd3)
• constrain player seat schemas and lock OpenAPI dependencies (2881789)
• format the runner and inherit updated fork fixtures (2ddf697)
• info: derive opponent availability and assists from installed host (76e0473)
• info: report installed opponents and match assist permissions (c1be539)
• info: retain compact briefing prompt budgets (4e19345)
• initialize SQLite before starting its connection pool (f1ff7e8)
• keep engine-private state out of match briefings (e103060)
• keep Sudoku answers out of serialized state and verify complete API lifecycle (1e4171c)
• limit oversized compact action lists to one labelled example (22cd397)
• llm: lock the workspace hash version and apply formatting (0bfeee2)
• lock bundled documentation and browser test dependencies (d326eca)
• mcp: commit SDK lockfile and own schema validation errors (07140cc)
• mcp: compact live match briefings consistently (1cdedc9)
• mcp: keep live compact briefings free of tensors and masks (063090c)
• mcp: negotiate modern subscriptions and verify watch limits (a297e8b)
• migrate SQLite before expanding the connection pool (169b574)
• opponents: stop search at truncated game states (2a709c4)
• opponents: treat length-capped states as search leaves (ad8ddc0)
• preserve match endings during simulation and pin hash dependency (d83426b)
• python: accept PettingZoo framework reset options (41fcd8b)
• python: integrate complete native observation support (a1dc4c7)
• python: preserve typed model failures and integrate policy views (fc2538c)
• python: return seat lists and include Connect Four tensors (c9fdcc7)
• require engine opt-in before exposing position notation (0e684c1)
• resolve runner paths within workflow steps (30a16db)
• retain tensor metadata in compact briefings (9d5ed32)
• satisfy runner formatting and fixture lint checks (eb1612c)
• scope HTTP test macro imports (7e60347)
• server: back off failed opponent jobs and release tracking after task failures (f0507f4)
• server: lock streamable MCP dependencies and format host (c3d65e5)
• store: check in PostgreSQL query metadata and locked dependencies (175ac27)
• training: bind checkpoints to opponent configurations (17e96d2)
• training: bind Gym checkpoints to opponent settings (b10e3ea)
• training: box batch frames to keep row variants compact (4bcea94)
• training: flush Parquet output with a writable Windows handle (19b387a)
• training: include UCI move-cap validation repair (8bf360d)
• training: include UCI move-cap validation repair (7e70e13)
• training: integrate shared ratings and validate anchor uncertainty (9fe9472)
• training: preserve compliant Parquet list field names (efed3c3)
• training: preserve dataset hashes across Windows checkouts (630da06)
• training: preserve reset provenance and platform-independent hashes (a333745)
• training: reserve final return storage and validate agent sequences (e8b4e69)
• training: retain position difficulty and rating metadata (c9ce7c6)
• type Sudoku pattern coordinates and format grading (0984581)
• uci: preserve protocol errors when the reader disconnects (0abb9bf)
• uci: require a user namespace before disabling nested namespaces (ba381e2)
• uci: retain legal advice at training move caps (0a157a2)
• uci: truncate validated analysis at episode caps (2ed9de8)
• web: make serve loads the browser app (#101) (2161ffe)
• web: reject foreign origins at the local proxy boundary (e7a27db)
• web: use the event history cursor contract (4f9099b)

---

This PR was generated with Release Please. See documentation.