clawock v0.4.1

Install from PyPI:

python -m pip install clawock==0.4.1

Added

• Intraday research plugins intraday_breakout, intraday_reversion and intraday_vwap, with strategy-lab replay/study workflows, account-aware net costs, walk-forward evidence and explicit research-only permissions (#3062).
• Intraday research evidence cards and traceable manual order recommendations, including expiry, historical net results and costs, stop-buyback semantics and audit fingerprints; recommendations do not place orders (#3070, #3077, #3080, #3104).
• Headless dsh task dispatch, a dedicated queue slot, per-run estimated costs, /dispatch-stats, task-chip statistics and patrol views in clawock-dsh (#3040, #3042, #3045, #3060, #3061).
• DeepSeek Search alongside Tavily for news retrieval (#3046).
• Shared-memory host toolkit and read-only OpenClaw corpus plugin, memctl doctor duplicate-memory diagnostics, and Chinese search with rarity-weighted words and partial-match credit (#3065, #3066, #3067, #3068, #3079, #3081).
• dispatch.sh adopt for handing Claude sessions to quota-aware runner continuation, plus automatic free-model catalog audits and patrol model selection (#3096, #2995, #2996).

Changed

• Chat cards render from structured slot data, with bold heading/today lines and tabular totals; brief packet summaries fold to fit their budget (#3092, #3103, #3113, #3163).
• config/models.toml owns model IDs, parameters and fallback chains. MiniMax uses the GA MiniMax-M3.1-Flash ID and documented adaptive thinking (#3091, #3087, #3097).
• Briefs, analyzers and dashboard views distinguish session price returns from position P&L, use the same input ledger for accounting checks, and render risk against the configured policy caps (#3035, #3036, #3104, #2987, #2992).

Fixed

• Quote integrity preserves HK dual-source divergence and exposes a warning, handles missing previous closes without fabricating flat returns, clears intraday ranges across sessions, and accepts floating-point rounding at existing tolerances (#3104, #3155, #3161).
• Decision coverage separates never-triggered episodes from unresolved ones; flat outcomes remain in win-rate denominators, reconstructed opening lots do not become real fills/T+1 verdicts, and legacy mind rows do not inflate pending plan counts (#3122, #3145, #3155, #3161).
• Trim projections jointly recalculate the remaining position and leg denominator, attribute concentration claims to the correct holding, and show current-book scenarios beside frozen plan rationales with explicit price bases (#3138, #3147, #3155, #3156, #3162).
• Public scorecards verify the published ledger's fingerprint and ancestry; stop, peer-divergence and leveraged-sibling checks use session returns rather than new-lot P&L (#3129, #3145).
• Brief revisions remain visible after delivery through either channel; collector crashes survive diagnostic retries, fallback-only delivery is marked degraded, and watchdog waiting and bar freshness follow their actual writers and schedules. Rate-limited reports can receive a deterministic Telegram backstop (#3106, #3130, #3149, #3162).
• Empty delivery windows no longer reuse an earlier card silently, stale queue/provider readings show their last successful timestamp, and intraday banners expire with their session (#3084, #3122, #3146, #3013).
• Dispatch inbox closure is atomic, terminal and continuation evidence is retained, and exhausted retries report rate limits; workflow errors retain stage distinctions, while missing shadow counts remain unknown instead of becoming zero (#3014, #3015, #3017, #3019, #3039, #3112).
• Additional evidence and rendering corrections preserve zero-price triggers and HK cards with missing underlying moves, report source failures instead of empty evidence, classify Eastmoney forums correctly, and use full DCA history for gold counterfactuals (#3104, #3108, #3047, #3003).

npm

clawock-dsh@0.4.1 published to npm.