fix(rectification): downgrade yearless personality probes to tie-breakers (BUG-629)
Independent Staging Quality Gate / validate (push) Successful in 11m54s
Independent Staging Quality Gate / publish (push) Successful in 8m14s

Ask dated dasha probes first; D9/D10 and nakshatra wait until that pool is empty, score at half weight, and never eliminate. Skill 10.0.21.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Jesse_Chen
2026-09-10 00:26:58 +08:00
co-authored by Cursor
parent eb8a3120e5
commit fef54b6e42
41 changed files with 1452 additions and 70 deletions
+16
View File
@@ -9732,6 +9732,22 @@
- 复发自:无
- 修复版本:待发布
## BUG-629 | 无年月性格题与带年月题同权,参与了淘汰
- 状态:resolved
- 首次发现:2026-09-09
- 最近更新:2026-09-09
- 影响面:`applyProbeOutcome``buildMethodFollowupPlan`、技法审计表、Skill 10.0.21
- 用户现象:真实校正里头两道题就是性格自评(相处方式 / 做事风格)。这些题没有年月锚点以外的分辨力验证,却按 ±2 计分并计入三次淘汰。
- 触发条件:账本已有感情或事业锚点(BUG-559),候选尚未分开,计划层把 `varga_style` / `nakshatra_boundary` 与带年月区分题排在同一池。
- 根因:`varga_style` 与带年月 `dasha_boundary` 共用 `SCORE_DELTA``strong_conflict_count`。分盘星座是确定计算,「星座 ↔ 性格」「用户自评 ↔ 类型」两层从未校准。
- 修复:性格题只在可问的带年月区分题为空、且仍有 ≥2 个未分开候选时才成为下一问,否则 `dropped_probes(reason="yearless_deferred")`。分值 ×0.5(±1),不递增冲突次数,`rounds.kind=tie_break`。报告 `D9 / D10 类型对照``月宿边界``reference`。离线脚本只对医院记录且不确定度 ≤2 分钟的匿名导出算命中率。
- 验证:`frontend/tests/rectification-yearless-probe-downgrade-20260909.test.ts``tests/test_varga_style_calibration_report.py`;既有 `rectification-probe-replay-loss-20260908.test.ts` 断言未改。
- 防复发:不得把 `varga_style` / `nakshatra_boundary` 加回带年月区分池;不得让性格冲突计入 `STRONG_CONFLICT_ELIMINATION_COUNT`;不得改 `SCORE_DELTA` 数值来代替权重。`asInferenceState``ROUND_KINDS` 必须含 `tie_break`,否则答过性格题的 `inference_state` 整份被丢掉。审计列 CHECK 仍只认 `informative` / `low_information``tie_break` 只活在 JSON,不得为迁列把 kind 改回 `informative`
- 相关记录:BUG-559、BUG-560、BUG-623
- 复发自:无
- 修复版本:待发布
## BUG-630 | 初始化后点首页「家庭」报运行合同未完成(runtime_contract_incomplete
- 状态:resolved