fix(web): rank remaining-minute probes by split and match choice kind
Independent Staging Quality Gate / validate (push) Failing after 22m30s
Independent Staging Quality Gate / publish (push) Has been skipped

Choice cards used a hardcoded domain menu and always asked existence.
Rank scoring layers by remaining-minute entropy, keep finance and health
volunteer-only, and ask D9/D10 style or exam quality so taps match outcomes.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Jesse_Chen
2026-08-25 20:44:55 +08:00
parent 7a4360d848
commit f5e73ef326
21 changed files with 921 additions and 114 deletions
+16
View File
@@ -5563,6 +5563,22 @@
- 复发自:BUG-376(关掉思考正文后,进度条只剩工具名,技法句不再出现;回到最新仍按咨询页浮层)
- 修复版本:待发布
## BUG-382 | 点选题与剩余切开不是同一题型,选题不是按剩余最大切开
- 状态:resolved
- 首次发现:2026-08-25
- 最近更新:2026-08-25
- 影响面:`discriminating_event_probes``candidate_contrast_packet``choice_card``method_followup_plan`、window_scan transitions
- 用户现象:生时纠正点选卡一律问「某年前后有没有这类事」。口语在问相处或职责时,卡仍问存在性。剩余分钟其实还能被家人盘切开时,下一问仍按写死的学业/事业/感情/搬家顺序出。
- 触发条件:覆盖已齐、候选仍并列;剩余 D9/D10 换升,或 D12/D7 切开而 D9 已齐;或大运年界与剩余切开同时存在。
- 根因:(1) 选择题走 `eventHypothesis` 存在性模板,D10 内部职责三分仍显示成有没有这件事。(2) Python 探针按整窗 `*_candidates_differ` 和精度阶段优先选题,每个领域取第一个能分开的年。(3) TS 剩余层写死 `D24→D5→D10→D9→D4`,并用假 IG 让学业压过事业/感情。
- 修复:计分领域按当前剩余分钟仍切开的层进池,财务/健康仍要用户先提过。Python 对剩余分钟打分,按分组熵加置信度取最高年。TS 按分组熵选题。年界差问存在性;D9/D10 换升问类型表相处/职责;学业问考试质量。两路 style 的「都不像」计 `unsure`,不把另一组分钟当「没有这件事」。不改 Skill `10.0.11`
- 验证:`tests/test_rectification_event_probes.py``tests/test_rectification_refinement_packet.py``frontend/tests/rectification-candidate-contrast-packet.test.ts``frontend/tests/rectification-choice-card.test.ts``frontend/tests/rectification-inference-machine.test.ts`
- 防复发:剩余分钟选题不得写死学业优先。点选卡题型必须跟 `choice_kind``expected_outcomes` 一致。两路 style 的 C 不得淘汰另一组分钟。不得改已哈希 Skill `10.0.11`
- 相关记录:BUG-375、BUG-379、BUG-380
- 复发自:BUG-375(区分卡已落到剩余分钟,但题型仍是存在性,选题顺序仍写死)
- 修复版本:待发布
## BUG-379 | 生时纠正已记入学后仍编造高考年并再问入学
- 状态:resolved