Add Windows cross-platform support#1
Open
ivandobskygithub wants to merge 21 commits into
Open
Conversation
Rewrite task runner from bash to Python for Windows compatibility, add explicit UTF-8 encoding to all file I/O in the loop runner, update PROMPT.md to use `python .project/task` invocations, and add venv activation rule for cross-platform use. Co-Authored-By: Claude Opus 4.6 <[email protected]>
… dataclass Co-Authored-By: Claude Opus 4.6 <[email protected]>
Add schema.sql with all 6 tables (markets, tokens, price_snapshots, game_states, signals, orders), indexes, and WAL/FK pragmas. Implement async StateStore class with typed CRUD for all tables and is_fresh() data freshness helper. 11 passing tests cover round-trips and edge cases. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Add schema.sql with all 6 tables (markets, tokens, price_snapshots, game_states, signals, orders), indexes, and WAL/FK pragmas. Implement async StateStore class with typed CRUD for all tables and is_fresh() data freshness helper. 11 passing tests cover round-trips and edge cases. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-Authored-By: Claude Opus 4.6 <[email protected]>
…ities Gamma API client (httpx) for events/markets discovery, CLOB Level 0 wrappers for order books/midpoints via py-clob-client, parse_event() for normalizing stringified JSON fields, and odds.py with Kambi/American conversion functions. 32 new tests, all passing. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-Authored-By: Claude Opus 4.6 <[email protected]>
…alignment Co-Authored-By: Claude Opus 4.6 <[email protected]>
…eneration Co-Authored-By: Claude Opus 4.6 <[email protected]>
…ent integration Co-Authored-By: Claude Opus 4.6 <[email protected]>
…in application loop Wire all components into asyncio-based pipeline with bounded queues, WebSocket listener, REST pollers, signal/order consumers, and coordinated startup/shutdown sequence. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-Authored-By: Claude Opus 4.6 <[email protected]>
…t, and fixtures Co-Authored-By: Claude Opus 4.6 <[email protected]>
…ds conversion Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-Authored-By: Claude Opus 4.6 <[email protected]>
Parametrized aliasing tests covering all abbreviation conflicts (PHX/PHO, NOP/NOH, SAS/SA), fixture-based team extraction, market alignment, date normalization, composite key commutativity, and match_and_persist integration with in-memory SQLite. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Fix unused Signal import lint error in test_signal_engine.py. Tests were already implemented by a prior iteration (26 signal engine tests, 14 order manager tests). All 239 tests pass, lint clean. Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-Authored-By: Claude Opus 4.6 <[email protected]>
Co-Authored-By: Claude Opus 4.6 <[email protected]>
…stem) Reliability (P0/P1): - SQLite busy_timeout + retry-on-locked writes (was 120k lock errors) - Live-auth hard-fail at startup + consecutive-failure trading halt watchdog (read-only-client fallback previously failed 1,420 live orders silently) - Settlement grading validates game finality before booking PnL; losing totals get real negative PnL; total_h1 graded on first-half score; untrustworthy finals flagged 'unsettled' instead of written off - Exit lifecycle is place-and-confirm: positions become 'exiting' and only book exited_* when the fill tracker confirms the sell matched; hockey ML time/OT exits now actually place sells; stale exit sells cancel + reopen - ARB_DRY_RUN is a hard gate over every per-variant/engine is_live flag - Live orders gated on Kambi game-state freshness (no trading stale feeds) - Market WS best bid/ask fixed (was taking worst level), freshness-gated; reconnect when new tokens are discovered (resubscribe was a no-op) - User WS handles maker_orders fills (resting sells), accumulates partials volume-weighted, confirms position exits on sell fills - Restart-safe order dedup via DB; bust-ladder timers and ladder/resting sell order IDs persisted on positions; exact signal attribution - Risk limits in USD (was summing shares), per-game exposure cap, real category caps; rotating logs; tracked background tasks Strategy: - Per-league live gating: NBA unders disabled (-0.104 EV/$1, n=17,303), NCAAB unders stay (+0.125, n=19,886) - EV-checked taker entries: buy at the ask (FAK) only when ask <= Wilson-bound win prob - margin; never rest limits above market (median ML fill was +0.58 above signal mid = adverse selection) - Fractional Kelly stake sizing on Wilson lower bounds from per-game samples - Clinch sniper: totals lines mathematically clinched are bought immediately, sized to book liquidity - Complete-the-box: buy the opposite token cheap to lock profit while liquidity is present; live moneyline price stop-losses (previously none); systematic resting take-profit sells for all live positions - Order book captured at +0s/+2s/+5s to measure edge surviving Polymarket's ~2s taker matching delay Tests: 244 -> 649 passing, ruff clean. Co-Authored-By: Claude Fable 5 <[email protected]>
Correctness (real-money impact): - Partial sell fills now reduce the position and keep monitoring the remainder instead of closing the whole row (untracked shares had no stop-loss/settlement); cumulative vs incremental fill semantics made explicit across fill_tracker and user-WS paths - User WS creates the position on a live BUY fill — marking the order 'matched' removed it from fill_tracker's scan, leaving real shares untracked; box hedges are recognized and never become independent positions (a parked take-profit would have silently unwound the box) - _begin_live_exit cancels competing resting sells first (no dual sells against the same shares) and no longer fabricates an exit price on zero balance — it links the possibly-filled prior sell for the fill tracker to confirm at the real traded price - _expire_stale_exits does not reopen a position when the cancel fails (the sell may have just matched — a second sell would be unbacked) - Live order dedup split live/dry-run: a dry-run grid variant no longer blocks the later live clinch sniper on the same token - Exposure caps count only committed orders (live/pending/matched) — expired GTD entries were consuming the per-game and daily caps - Live bust ladder now covers total_h1 (the inner H1 guard was dead code); dry-run ML positions keep their stop_loss_price (was zeroed, skewing paper stats vs live) - Tick snapping rounds to nearest (float floor dropped a full tick for on-grid prices like 0.58); 'other' category cap can actually trigger - Live signals without win-prob estimates (hockey/football) now EV-check at the variant limit and cross the spread or skip — never rest above market; per-sport kill switches ARB_HOCKEY_LIVE/ARB_FOOTBALL_LIVE - main.py requires trading auth whenever not in dry-run (hockey/football engines go live outside the basketball variant lists) - settlement: unknown leagues are never graded with basketball math; cross-platform engine freshness tightened to 30s - unsettled positions surfaced with recurring warnings Performance / hygiene: - Index on signals(token_id, signal_type) — dedup lookups were full table scans of an 85k+ row table on every poll cycle - Removed dead ws_client.py (shipped the known-bad best-price logic), dead _CATEGORY_PREFIX map, duplicate _FOOTBALL_LEAGUES set; reuse per-game exposure within order_consumer; fixed stale ladder comment Co-Authored-By: Claude Fable 5 <[email protected]>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
.project/taskfrom bash to Python for Windows compatibility (bash builtins likesed,awk, process substitution don't work reliably on Windows)encoding="utf-8"to all file I/O in.project/loop(Windows defaults to system encoding, not UTF-8)PROMPT.mdto usepython .project/taskinvocations and add venv activation rule supporting both Windows and Linux/macOS pathsREADME.mdto reflect the Python task runnerTest plan
python .project/task nextreturns the correct next taskpython .project/task listdisplays all task states correctlypython .project/task <ID> progressandpython .project/task <ID> donemanage state transitionspython .project/loop --max 1to confirm the loop runner works with UTF-8 encoding🤖 Generated with Claude Code