Repository navigation
Conversation
Co-Authored-By: DeepSeek V4.1 Flash <noreply@deepseek.com>
Co-Authored-By: DeepSeek V4.1 Flash <noreply@deepseek.com>
Co-Authored-By: DeepSeek V4.1 Flash <noreply@deepseek.com>
Co-Authored-By: DeepSeek V4.1 Flash <noreply@deepseek.com>
Co-Authored-By: DeepSeek V4.1 Flash <noreply@deepseek.com>
Co-Authored-By: DeepSeek V4.1 Flash <noreply@deepseek.com>
Co-Authored-By: DeepSeek V4.1 Flash <noreply@deepseek.com>
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
概要
赛马被系统判负(
lost_futility、lost_exam)且相对母体改了两处及以上的挑战者,会在开发窗口上逐项消融:先测母体基线,再把每一处改动单独加到母体上回测,得出每项改动是有害、有益还是中性;结果作为单因素观察回流研究记忆,下一轮 AVO 可以据此提出「只留有益改动」的收窄候选。结论是开发窗口内的单因素效应,不含改动之间的交互,不是因果结论。依赖与合并顺序
codex/race-outcome-memory。请先合并 feat(arc): 赛马输赢回流研究记忆——前向结果表、裁判记账与沿接班链回流 #79;feat(arc): 赛马输赢回流研究记忆——前向结果表、裁判记账与沿接班链回流 #79 合并并删除分支后,本 PR 会自动改到main。0051_arc_ablation_runs,会改变生产研究环。做了什么
arc_ablation_runs(每个 target/挑战者一行,唯一键)与arc_ablation_steps(每项改动一步,唯一键(run_id, step_key));迁移幂等可重跑,SQLite 与 PostgreSQL 都测过。arc/ablation.py:从arc_forward_outcomes挑已对接血缘的输家;冻结开发窗口(as_of为前一天);走与 AVO develop 同一条ARCSelfTestService.run(purpose="development")路径(同成本、资金、标的口径);self_test.run新增可选参数data_reference,让候选直接用基线回测的封存数据,避免为带结构算子的候选再隐式建一行基线,从而每个输家恰好「改动数 + 1」次回测(不传时 AVO 行为不变)。BacktestActionRunner(读失败有界重试、中断有界续跑、单作业等待上限 24 小时、版本变化原样记失败);计费点是作业身份落库,崩溃后重放同一幂等键不重复建作业、不重复计费。EvolutionConfig新增ablation_enabled(默认开)、ablation_max_backtests_per_day(默认 6,按 UTC 日)、ablation_max_backtests_per_loser(默认 6,范围 3–6);不占每日研究名额;一次只跑一个回测。旧的已存配置照常加载。关闭:把ablation_enabled设为 false。_memory在前向结果之后、开发记忆之前插入type=ablation条目(memory_id形如ablation:<run_id>),只投影本轮已展示的前向输家的已完成消融;与前向结果合计不超过 8 条,模型可见的前 20 条里开发记忆至少 12 条;消融窗口不得晚于本轮开发窗口结束日;出错只记运行错误,研究照常。AVO 系统提示说明含义与引用方式;研究证据视图单列ablation_references。GET /api/v1/arc/evolution的budget.ablation(当日用量与上限、各状态任务数、原因计数、输家计数、下次检查时间;独立事务,失败只降级为 unavailable);效果报告新增ablation_*字段;只读列表GET /api/v1/arc/evolution/ablations(不含原始回测指标)。失败都写具体原因码(如parent_changed_since_race、parent_relay_hold、daily_budget_exhausted、too_many_changes),不补零。安全边界
auto_enroll_uncontested_variants只纳管has_formal_review_binding为真的候选,relay_variants也跳过未审核的候选,所以消融行不会被当成挑战者,也不会进入tried_variants。不需要新增 purpose,BitPro 未改。budget_status与_memory的 GitNexus 影响分析为 HIGH(tick扫描依赖),两处新增逻辑都在独立事务或 try/except 内,异常只降级自身;相对 feat(arc): 赛马输赢回流研究记忆——前向结果表、裁判记账与沿接班链回流 #79 分支的detect_changes为 LOW,无受影响执行流程。验证
tests/test_arc_ablation.py、test_arc_ablation_schema.py、test_arc_ablation_postgres.py(真 PostgreSQL)、test_arc_ablation_observability.py、test_arc_ablation_memory.py,并在test_arc_self_test.py、test_arc_evolution.py、test_arc_external_surface.py、test_avo_evidence.py补充用例。LC_ALL=en_US.UTF-8):前端 lint、15 项测试与构建,ruff、mypy(319 个文件)、pytest2325 项全部通过。前端复用主工作区依赖、未重新安装。未线上实测与已知限制
BacktestActionRunner相同的边界)。docs/agent_architecture_and_interview_handbook.pdf未重新生成,其中迁移数仍为旧值。实现由 DeepSeek V4.1 Flash 按 Claude 的任务书执行,每个提交都经 Claude 审查。
🤖 Generated with Claude Code