claude_dev 89a8dd1487 feat(factor): 向量化评估指标——RankIC逐行秩相关/ICIR/t/胜率/多空年化/换手/十分组/月度IC/结论分级 [vps]

2026-08-25 00:12:35 +08:00

..

api

test(api): 修channel_test随启随验过时断言——实盘=单probe_all定时(启动+90s随墙钟),四时点改由无env直测钉死 [nas]

2026-08-24 13:49:21 +08:00

backtest

fix(tests): Mac本地9个既有测试失败根治——test_cta_engine的sys.modules mock注入从import期挪进module-scoped fixture(pytest全量collection先于执行,import期注入污染后收集的metrics/datareader/factor)+venv310按lock补装empyrical-reloaded==0.5.12;影子柜台设计文档头更新(P1已上线/P3前半已实现) [nas]

2026-08-15 00:02:12 +08:00

data

…

data_platform

test(ci): CI扩跑tests/portfolio+钉死测试归一——issue#35方案B,440组合引擎题进门禁 [nas]

2026-08-22 22:51:57 +08:00

factor

feat(factor): 向量化评估指标——RankIC逐行秩相关/ICIR/t/胜率/多空年化/换手/十分组/月度IC/结论分级 [vps]

2026-08-25 00:12:35 +08:00

orchestrator

…

portfolio

perf(data): min/max改逐对双seek——INDEXED BY批IN形态VPS实测仍不达标(30只冷8.1s/热0.25s,大池冷态分钟级),ORDER BY ±datetime LIMIT 1三列全等前缀每股恰2次寻位冷热皆毫秒级——上一发(80fc517)生产探针续测定形态:①原IN+GROUP BY劣化计划101s冷;②钉复合索引后计划翻转(TEMP B-TREE消失)但仍逐股扫全部interval条目(d+15m+5m),30只冷8.1s→momentum池~2000只外推冷态≈9min过不了<60s验收;③行值IN变体在本SQLite对(row-value IN+INDEXED BY)走全索引扫描7min不归(弃);终态=逐对双seek,symbol/exchange/interval三列全等直落d子区间两端,连15m/5m条目都不碰;无日线对不落minmax(下游.get默认(None,None)同旧GROUP BY语义);测试5个重写到seek API(双索引名/回退/EXPLAIN计划签名interval=?直落+无TEMP B-TREE/batch==逐只等价);628绿;待VPS探针终验双seek计时 [vps]

2026-08-24 14:14:11 +08:00

shadow_desk

feat(shadow-desk): P0对账spike(runner加initial_positions通道+final_portfolio产出): BacktestEngine续跑注入初始持仓(引擎原生支持,不消耗现金); 期末现金/持仓/总值从engine.context.portfolio直取; run_backtest_json透传initial_positions+final_portfolio; scripts/shadow_desk/spike_p0全量vs 3段续跑对账(持仓逐只/现金总值/净值曲线<1e-6,自动月初分段或--split压除权边界); 6测试 [vps]

2026-08-14 10:10:51 +08:00

spikes

…

trader

fix(shadow): 双轨对账count_match改按票+方向聚合总量——QMT部分成交拆行致27vs5恒False纯计数噪音 [vps]

2026-08-24 23:07:12 +08:00

.gitkeep

…

test_live_api.py

…

test_live_engine.py

…

websocket_test.html

…