fix(rectification): narrate the credible range and stop scoring periods by length (BUG-569–570)

Choice copy was comparing the search window, so every answer said the range had not changed. Block-scan summed raw scores, so longer afternoon windows won before any evidence difference.

Co-authored-by: Cursor <cursoragent@cursor.com>

BLOCKED.md

@@ -83,3 +83,20 @@
 - 无法获取任务书要求的上游 `interview_playbook.md`、`evidence_thresholds.md`：任务书所指的 `~/.workbuddy/skills/jyotish-birth-time-rectification/` 在当前执行环境不存在，仓库内只有测试对该外部路径的引用；未伪造文件，也没有可验证的上游来源可供导入。
 - 无法获取一次真实本地校正会话完整记录：当前仓库没有可证明为真实线上会话的完整原始记录，执行环境也没有受控会话/上游维护者提供的记录。因此无法可靠回答轮数、最终区间宽度、`confidence` 与 `can_apply`。
 - 该信息收集缺口不阻塞任务 1–3，按 v2 任务书继续实现并在进度文件中标记为未验证；不得据此声称已验证“固定题数后停止”的上游机制。
+
+## BLK-001 · 长对话本地收窄后 VedAstro 调用窗口与断言不一致（2026-09-06）
+
+- 状态：blocked（基线即失败，非 BUG-565～567 引入）
+- 首次记录：2026-09-06
+- 测试：`tests/test_active_rectification_api.py::test_long_real_conversation_reaches_vedastro_after_local_range_is_narrow`
+- 复现：
+
+```bash
+.venv/bin/python -m pytest -q tests/test_active_rectification_api.py::test_long_real_conversation_reaches_vedastro_after_local_range_is_narrow --tb=line
+```
+
+- 失败断言：`result["winning_segment"]` 期望 `start_time=05:07`、`end_time=05:08`、`representative_time=05:07`、`width_minutes=2`；实测 `04:16` / `04:16` / `04:16` / `width_minutes=1`。
+- 已知同样失败的 SHA：`e2f4b55c`（父任务书验收段已复跑）；工作树 `43a26a3c`（文档头，Python 与 `origin/staging` 一致）同样失败。测试首次出现于 `3ca30ed7`。未做完整二分：父任务已证明早于 BUG-558～560。
+- 本单范围：只记录。不改引擎、不改门槛、不改断言来让它变绿。
+- 相关：BUG-560 根因升级（分钟级原始分区分力≈随机）；`docs/tasks/TASK-rectification-convergence-exit-20260906.md` 验收段。
+

CHANGELOG.md

@@ -1,5 +1,9 @@
 # 印度占星 Skill 更新日志
 
+## 2026-09-07 — 答完题后的范围跟顶部一致，不清楚出生时间时不再偏爱下午
+
+生时校正点完选择题后，旁白里的范围跟输入框上方那一行走，不会再把开工搜索窗口说成「范围没变」。完全不清楚出生时间时，清晨、上午、下午、傍晚、夜里五段按段内平均分来比，不会因为下午更长就先天占优。Skill 版本仍是 10.0.14。
+
 ## 2026-09-06 — 完全不知道出生时间时先比时段
 
 选「完全不清楚」后可以直接开始生时校正：先用记得的经历比出清晨、上午、下午、傍晚或夜里哪一段更像，选定后再按分钟收窄。不会再劝退成「以后再做」，也不会在全日窗口上给出分钟或采用卡。家人记得的时段只作旁白建议，改不了搜索窗口。Skill 版本仍是 10.0.14。

docs/BLOCKED.md

@@ -1,21 +0,0 @@
-# Blocked work
-
-本文件记录当前无法在本仓内闭环、也不得标成通过的验证。新条目追加在表后。不写姓名、出生资料或完整请求体。
-
-## 条目
-
-### BLK-001 · 长对话本地收窄后 VedAstro 调用窗口与断言不一致
-
-- 状态：blocked（基线即失败，非 BUG-565～567 引入）
-- 首次记录：2026-09-06
-- 测试：`tests/test_active_rectification_api.py::test_long_real_conversation_reaches_vedastro_after_local_range_is_narrow`
-- 复现：
-
-```bash
-.venv/bin/python -m pytest -q tests/test_active_rectification_api.py::test_long_real_conversation_reaches_vedastro_after_local_range_is_narrow --tb=line
-```
-
-- 失败断言：`result["winning_segment"]` 期望 `start_time=05:07`、`end_time=05:08`、`representative_time=05:07`、`width_minutes=2`；实测 `04:16` / `04:16` / `04:16` / `width_minutes=1`。
-- 已知同样失败的 SHA：`e2f4b55c`（父任务书验收段已复跑）；本工作树 `43a26a3c`（文档头，Python 与 `origin/staging` 一致）同样失败。测试首次出现于 `3ca30ed7`。未做完整二分：父任务已证明早于 BUG-558～560。
-- 本单范围：只记录。不改引擎、不改门槛、不改断言来让它变绿。
-- 相关：BUG-560 根因升级（分钟级原始分区分力≈随机）；`docs/tasks/TASK-rectification-convergence-exit-20260906.md` 验收段。

docs/BUG_HISTORY.md

@@ -8806,4 +8806,36 @@
 - 复发自：无
 - 修复版本：待合入 `origin/staging`（`codex/rectification-convergence-exit-fix-20260906`）
 
+## BUG-569 | 答后旁白把搜索窗口当成可信区间，每题都说「范围没变」
+
+- 状态：resolved
+- 首次发现：2026-09-07
+- 最近更新：2026-09-07
+- 影响面：`composeChoiceNarration`、`applyRectificationChoice`、`probe-explain.explainRangeChange`
+- 用户现象：顶部只读范围已经随答题收窄，助手旁白仍说「范围没变」。
+- 触发条件：点选区分卡 A/B/C，搜索窗口 `range_start/range_end` 不变，而 `credible_range` 变了。
+- 根因：`applyRectificationChoice` 把 `InferenceState.range_start/range_end`（开工搜索窗口）传给旁白。答题改的是 `credible_range`。直喂 `composeChoiceNarration` 的单测没走这条路径。
+- 修复：旁白改比 `previous.credible_range` 与 `applied.state.credible_range`。入参改名 `credibleBefore/credibleAfter`。任一为空则不说范围句。
+- 验证：`frontend/tests/rectification-answer-choice.test.ts`：搜索窗口固定、`credible_range` 收窄时旁白含「范围从 X 收到 Y」；`credible_range` 不变时仍说「范围没变」。源扫描锁定不再传 `range_start`。
+- 防复发：不得把 `range_start/range_end` 再传给答后范围句。新增端到端用例必须走 `applyRectificationChoice`，不能只直喂纯函数。
+- 相关记录：BUG-568
+- 复发自：过程解释层（`814c924e`）把搜索窗口误当可信区间
+- 修复版本：待合入 `origin/staging`（`codex/rectification-capability-fix-20260907`）
+
+## BUG-570 | 时段支持度按段内原始分求和，长时段先天占优
+
+- 状态：resolved
+- 首次发现：2026-09-07
+- 最近更新：2026-09-07
+- 影响面：`scripts/rectification/api_service.py::block_scan`
+- 用户现象：完全不清楚出生时间时，第一张时段卡下午段支持度偏高，即使各分钟分数相同。
+- 触发条件：`stage=block_scan`、10 分钟步长扫 24 小时。五段候选数 24/24/36/30/30。
+- 根因：`relative_support` 用段内 `sum(score)`。下午 36 个候选，清晨 24 个；引擎原始分底座远大于事件差，于是在没有证据差异时也是在比时段长度。
+- 修复：`raw_support = max(mean(score in block) − min(all scores), 0)`，再归一。全部为零时五段各 20%。不进采用门，不改分钟级 `_relative_support`。
+- 验证：`tests/test_rectification_v5_services.py`：144 行同分 → 五段各 20.0；只抬高清晨 → 清晨最高，下午不再因长度领先。`minute_step=1` 指纹用例未改。
+- 防复发：`block_scan` 不得再对段内原始分求和。分钟级 prior 仍走原来的正比例归一（BUG-560）。
+- 相关记录：BUG-560
+- 复发自：unknown-time `block_scan`（`814c924e`）按时段长度偏置
+- 修复版本：待合入 `origin/staging`（`codex/rectification-capability-fix-20260907`）
+
 

docs/tasks/PROGRESS-rectification-capability-fix-20260907.md

@@ -0,0 +1,42 @@
+# PROGRESS · 校正能力补齐修复单（2026-09-07）
+
+工作树：`.worktrees/rectification-capability-fix-20260907`
+分支：`codex/rectification-capability-fix-20260907`（跟踪 `origin/staging`）
+基线：`origin/staging` @ `d6ee8cbb`（含 `814c924e` BUG-565～568）
+
+未改 `SKILL.md`（仍 10.0.14）、未改采用门 / 确认门 / `MIN_SEPARATION_LEAD` / 分钟级 `_relative_support`、未改 `minute_step=1` 路径、未改 `page.tsx`。未把 `.venv` 加入提交。
+
+本单状态：**待验收**。BUG-569 / BUG-570 已实现；P2 文案与 P3 BLOCKED 同提交。
+
+| 任务 | 状态 | BUG |
+| --- | --- | --- |
+| 4.1 旁白比 credible_range | 完成 | BUG-569 |
+| 4.2 时段支持度按段内均值 | 完成 | BUG-570 |
+| 4.3 TRACK_LABEL + BLOCKED 归位 | 完成 | P2 / P3 |
+| 4.4 记录 | 完成 | BUG_HISTORY / CHANGELOG / 本文件 / testing 第 2 条 |
+
+## 实现要点
+
+- `composeChoiceNarration` 入参改为 `credibleBefore/credibleAfter`。`applyRectificationChoice` 传入 `previous.credible_range` 与 `applied.state.credible_range`。搜索窗口 `range_start/range_end` 不再进旁白。
+- `block_scan`：`raw_support = max(mean(score in block) − min(all scores), 0)`，全零时五段各 20%。`candidate_count` 保留。
+- `probe-explain.ts` 的大运标签改走 `publicRectificationMethodLabel`（`rectification-varga-sentence.ts`），不再维护第二份 TRACK_LABEL。
+- `docs/BLOCKED.md` 的 BLK-001 搬回根目录 `BLOCKED.md` 并删除前者。
+
+### 既有断言改动
+
+| 文件 | 原值 | 新值 | 原因 |
+| --- | --- | --- | --- |
+| `rectification-answer-choice` 纯函数旁白 | `rangeBefore/rangeAfter` 直喂 | 入参改名 `credibleBefore/credibleAfter`，仍测拼句；新增 `applyRectificationChoice` 端到端 | 直喂拦不住把搜索窗口当可信区间 |
+| `rectification-choice-card` why_user | 只锁「分成两组」 | 并锁 Vimshottari / Narayana，禁止音译 | P2 |
+
+## 测试
+
+| 命令 | 结果 |
+| --- | --- |
+| 非 DB `rectification-*` + `consultation-*` + `report-*` + `personal-report-*` + `agent-voice-copy-contract` | **1521 pass / 0 fail**（≥ 1518） |
+| `.venv/bin/python -m pytest tests/test_rectification_v5_services.py -k 'not seven_events'` | **31 pass / 0 fail**（含 BUG-570 等分网格与清晨抬高） |
+| `test_block_scan_seven_events_finishes_within_fifteen_seconds` 单独跑 | 通过。整文件连跑时偶发 16.0s / 24s / 34.8s，与本次均值公式无关；机器有负载时 15s 预算偏紧 |
+| `npx tsc --noEmit` | 0 错 |
+| `./node_modules/.bin/eslint --quiet src tests` | 0 error |
+
+排除 `rectification-*-database.test.ts`。ERR-011：验收命令用 `.venv/bin/python -m pytest`。未改 `page.tsx`。未改 `minute_step=1` 指纹。

docs/tasks/PROGRESS-rectification-convergence-exit-fix-20260906.md

@@ -13,7 +13,7 @@
 | 5.1 单一写入方 | 完成 | BUG-565 |
 | 5.2 穷尽态视为已交付 | 完成 | BUG-565 |
 | 5.3 穷尽分支不吞题 | 完成 | BUG-566 |
-| 5.4 范围小字还原 + 记录 | 完成 | BUG-567；BUG-560 根因升级；`docs/BLOCKED.md` BLK-001 |
+| 5.4 范围小字还原 + 记录 | 完成 | BUG-567；BUG-560 根因升级；`BLOCKED.md` BLK-001 |
 
 ## 实现要点
 
@@ -38,7 +38,7 @@
 | 非 DB `tests/rectification-*.test.ts` + `agent-voice-copy-contract.test.ts` | **910 pass / 0 fail**（解释层叠上后 917） |
 | `npx tsc --noEmit` | 0 错 |
 | `npm run lint` | 0 error / 81 warning（既有 warning，未清） |
-| `.venv/bin/python -m pytest tests/test_active_rectification_api.py::test_long_real_conversation_reaches_vedastro_after_local_range_is_narrow` | 失败：`winning_segment` 04:16 / 宽 1 vs 期望 05:07–05:08 / 宽 2。基线 `e2f4b55c` 同样失败。见 `docs/BLOCKED.md` BLK-001。 |
+| `.venv/bin/python -m pytest tests/test_active_rectification_api.py::test_long_real_conversation_reaches_vedastro_after_local_range_is_narrow` | 失败：`winning_segment` 04:16 / 宽 1 vs 期望 05:07–05:08 / 宽 2。基线 `e2f4b55c` 同样失败。见 `BLOCKED.md` BLK-001。 |
 
 排除 `rectification-*-database.test.ts`。ERR-011：验收命令用 `.venv/bin/python -m pytest`。未改 `page.tsx`。
 

docs/tasks/README.md

@@ -66,7 +66,7 @@
 | `TASK-rectification-explain-layer-20260906.md` | `PROGRESS-rectification-explain-layer-20260906.md` | 过程解释层（对照上游 yinduzhanxing 旧工作台唯一领先的业务层）：每张卡服务端生成「为什么问这题」与 A/B/C/D「答了会怎样」、答后旁白改成「哪段升降 + 范围从 X 收到 Y」、每轮步骤条「第 N 步 / 为什么 / 下一步」；不做双视图 | 已验收（带修复单：答后旁白比的是搜索窗口，永远「范围没变」，P1） | `814c924e`；修复单 `TASK-rectification-capability-fix-20260907.md`（BUG-569） |
 | `TASK-rectification-range-reading-20260906.md` | `PROGRESS-rectification-range-reading-20260906.md` | 可信区间成为一等公民：采用时落库 `adopted_credible_range`，报告 `read_report_candidate_range` 与聊天 `verified_chart(accepted)` 都改读它并接同一份 `birth_time_sensitivity`（现在报告读的是开工窗口，BUG-568）；引擎 >15 分钟只取 3 样本改为 ≤31 逐分钟；采用旁白加「稳定 / 随分钟变」两句；`declared_birth_window` 复用 | 已验收通过（test:db 37/0 为执行方数字，本机无 Docker） | `814c924e`（BUG-568）；两条迁移待应用 |
 | `TASK-rectification-unknown-time-20260906.md` | `PROGRESS-rectification-unknown-time-20260906.md` | 完全不知道出生时间的两段式路线：`stage=block_scan` 以 10 分钟步长扫 24 小时只做事件计分、出五时段四选卡（不写账本不采用），选定后进现有分钟流程；引擎加 `minute_step`；开场读 `birth_time_clue`；删 intake 劝退文案 | 已验收（带修复单：时段支持度按段长偏置，下午段先天 25%，P1） | `814c924e`；修复单 `TASK-rectification-capability-fix-20260907.md`（BUG-570） |
-| `TASK-rectification-capability-fix-20260907.md` | `PROGRESS-rectification-capability-fix-20260907.md` | 能力补齐修复单：答后旁白把 `range_start/range_end`（搜索窗口）当范围比较，每题都说「范围没变」；`block_scan` 五段支持度按段内原始分求和，长时段先天占优（24/24/36/30/30 个候选）；`TRACK_LABEL` 音译与产品 Vimshottari/Narayana 口径不一；BLK-001 写到 `docs/BLOCKED.md` 应回根目录 | 待领取 | `codex/rectification-capability-fix-20260907`（BUG-569～570） |
+| `TASK-rectification-capability-fix-20260907.md` | `PROGRESS-rectification-capability-fix-20260907.md` | 能力补齐修复单：答后旁白把 `range_start/range_end`（搜索窗口）当范围比较，每题都说「范围没变」；`block_scan` 五段支持度按段内原始分求和，长时段先天占优（24/24/36/30/30 个候选）；`TRACK_LABEL` 音译与产品 Vimshottari/Narayana 口径不一；BLK-001 写到 `docs/BLOCKED.md` 应回根目录 | 待验收 | `codex/rectification-capability-fix-20260907`（BUG-569～570） |
 | `TASK-api-not-configured-mislabel-20260904.md` | `PROGRESS-api-not-configured-mislabel-20260904.md` | 16 处路由把数据库瞬断（部署切换窗口）兜底翻译成 503「服务尚未配置」；改为仅配置错误用该文案，其余 `service_unavailable`，收敛为共享 helper | 已验收 | `5483649b`（BUG-542）；2 条子进程测试留 CI Node 22 复核 |
 | `TASK-rectification-ux-20260902.md` | `PROGRESS-rectification-ux-20260903.md` | 会话面空白假死与交互摩擦 | 已验收 | `d159f08e`（09-03 在新基线重做后合入，BUG-505～509） |
 

docs/testing/rectification-explain-layer-20260906.md

@@ -15,12 +15,12 @@
 - ✅ 预期：题干下方有可点开的「为什么问这题」。展开后能看到年月或 D9/D10，以及「分成两组」或「类型差异」。没有「概率 / 置信度 / 确定」。
 - ❌ 失败：没有折叠说明；或说明仍是给模型看的「时间范围锁定…语义目标是…」。
 
-## 2. 答完一题能看见哪段升降和范围变化
+## 2. 答完一题旁白里的范围与顶部范围一致
 
 点 A 或 B。
 
-- ✅ 预期：助手旁白仍以「已记录你的选择」开头，并写哪段领先/落后；范围变了就写「范围从 … 收到 …」，没变就写「范围没变」。点 D 仍是「这题先不计分，换一件事问」。
-- ❌ 失败：仍只有「并更新了候选比较」；或范围没变却写成收窄。
+- ✅ 预期：助手旁白仍以「已记录你的选择」开头，并写哪段领先/落后；旁白里的范围与输入框上方那一行相同，变了就写「范围从 … 收到 …」，没变就写「范围没变」。点 D 仍是「这题先不计分，换一件事问」。
+- ❌ 失败：仍只有「并更新了候选比较」；或顶部范围已经收窄，旁白却写「范围没变」。
 
 ## 3. 每轮顶部有第 N 步 / 原因 / 下一步
 

frontend/src/lib/rectification-agentic/v9/answer-choice.ts

@@ -592,8 +592,8 @@ export async function applyRectificationChoice(
     appliedInference: persistable && scoring,
     answerClass,
     deltasByCluster: clusterScoreDeltas(previous.candidates, appliedScoreDeltas),
-    rangeBefore: [previous.range_start, previous.range_end],
-    rangeAfter: [applied.state.range_start, applied.state.range_end],
+    credibleBefore: previous.credible_range,
+    credibleAfter: applied.state.credible_range,
   });
   const evidenceFp = dossier.latestResult?.evidenceLedgerFingerprint
     ?? evidenceLedgerFingerprint(dossier.evidence);

frontend/src/lib/rectification-agentic/v9/choice-action.ts

@@ -102,8 +102,8 @@ export function composeChoiceNarration(input: {
   appliedInference: boolean;
   answerClass?: AnswerClass | null;
   deltasByCluster?: readonly ClusterScoreDelta[];
-  rangeBefore?: readonly [string, string] | null;
-  rangeAfter?: readonly [string, string] | null;
+  credibleBefore?: readonly [string, string] | null;
+  credibleAfter?: readonly [string, string] | null;
 }): string {
   if (input.optionId === "stop") {
     return `已记录你的选择，并结束本次校正，交付当前可信区间和代表性工作时间。${RECTIFICATION_TERMINATION_COPY}`;
@@ -119,7 +119,7 @@ export function composeChoiceNarration(input: {
   }
   if (input.appliedInference) {
     const movement = explainScoreMovement(input.deltasByCluster ?? []);
-    const range = explainRangeChange(input.rangeBefore, input.rangeAfter);
+    const range = explainRangeChange(input.credibleBefore, input.credibleAfter);
     if (movement || range) {
       return ["已记录你的选择", movement, range].filter(Boolean).join("。") + "。";
     }

frontend/src/lib/rectification-agentic/v9/probe-explain.ts

@@ -3,6 +3,7 @@
  * The model writes the stem; this module writes "why" and "what answering does".
  */
 
+import { publicRectificationMethodLabel } from "../../rectification-varga-sentence.ts";
 import type { AnswerClass } from "../core/types.ts";
 import type { DiscriminatingEventProbe, ProbeExpectedOutcome } from "./refinement-packet.ts";
 
@@ -31,11 +32,6 @@ export type ProbeUserExplain = Readonly<{
   answer_impact: ProbeAnswerImpact;
 }>;
 
-const TRACK_LABEL: Readonly<Record<string, string>> = {
-  vimshottari: "毗湿奴多利",
-  narayana: "那罗延",
-};
-
 const CLOCK = /^(?:[01]\d|2[0-3]):[0-5]\d$/;
 
 export const PROBE_EXPLAIN_COPY = {
@@ -91,7 +87,10 @@ function rangesForTimes(
 }
 
 function trackLabels(tracks: readonly string[] | undefined): string {
-  const labels = [...new Set((tracks ?? []).map((track) => TRACK_LABEL[track]).filter(Boolean))];
+  const labels = [...new Set((tracks ?? []).flatMap((track) => {
+    const label = publicRectificationMethodLabel(track);
+    return label ? [label] : [];
+  }))];
   return labels.join("、");
 }
 

frontend/src/lib/rectification-varga-sentence.ts

@@ -22,11 +22,17 @@ export const PUBLIC_RECTIFICATION_METHOD_LABELS: Readonly<Record<PublicRectifica
   "functional-benefic-malefic": "本命功能吉凶星",
 };
 
+export function publicRectificationMethodLabel(methodOrTrack: string): string | null {
+  const direct = PUBLIC_RECTIFICATION_METHOD_LABELS[methodOrTrack as PublicRectificationMethod];
+  if (direct) return direct;
+  return PUBLIC_RECTIFICATION_METHOD_LABELS[`${methodOrTrack}-dasha` as PublicRectificationMethod] ?? null;
+}
+
 export function vargaSentenceFromMethods(
   methods: readonly string[] | null | undefined,
 ): string | null {
   const labels = [...new Set((methods ?? []).flatMap((method) => {
-    const label = PUBLIC_RECTIFICATION_METHOD_LABELS[method as PublicRectificationMethod];
+    const label = publicRectificationMethodLabel(method);
     return label ? [label] : [];
   }))];
   if (labels.length === 0) return null;

frontend/tests/agent-voice-copy-contract.test.ts

@@ -150,6 +150,13 @@ test("collect prompt skip no longer writes a second assistant turn", () => {
   assert.doesNotMatch(exit, /shouldPersistFocusPromptTurn/);
 });
 
+test("probe explain tracks use the shared Vimshottari / Narayana labels", () => {
+  const explain = readFileSync(new URL("../src/lib/rectification-agentic/v9/probe-explain.ts", import.meta.url), "utf8");
+  const visible = listUserVisibleCopy().join("\n");
+  assert.doesNotMatch(explain, /毗湿奴多利|那罗延/);
+  assert.doesNotMatch(visible, /毗湿奴多利|那罗延/);
+});
+
 test("E-2 display copy keeps engine meaning out of instruction tone", () => {
   const dasha = engineMeaningToDisplayCopy(
     "主限更偏向 05:00，分盘大运更偏向 04:54。冲突时不能按更高把握收口。",

frontend/tests/rectification-answer-choice.test.ts

@@ -28,6 +28,7 @@ import {
   isIncompleteRunBanner,
 } from "../src/lib/rectification-agentic/v9/run-diagnostic.ts";
 import { applyHoldoutAnswer, buildInferenceState } from "../src/lib/rectification-agentic/core/build-state.ts";
+import { applyChoiceWithoutEvidence } from "../src/lib/rectification-agentic/v9/inference-adapter.ts";
 import { RECTIFICATION_TERMINATION_COPY } from "../src/lib/rectification-agentic/core/rectification-decision.ts";
 import { containsBoundarySemantics, RECTIFICATION_USER_COPY } from "../src/lib/rectification-agentic/user-copy.ts";
 import { parseV9CaseDossier, RectificationToolServiceError } from "../src/lib/rectification-agentic/v9/tool-service.ts";
@@ -193,6 +194,68 @@ function inferenceState() {
   });
 }
 
+function rangeNarrationInference(leadSupport: number, trailSupport: number) {
+  return buildInferenceState({
+    range_start: "04:31",
+    range_end: "05:07",
+    candidates: [
+      { id: "04:31", time: "04:31", relative_support: leadSupport },
+      { id: "05:07", time: "05:07", relative_support: trailSupport },
+    ],
+    events: [
+      { id: "e1", domain: "education", year: 2016, precision: "month" },
+      { id: "e2", domain: "career", year: 2018, precision: "year" },
+      { id: "e3", domain: "relationship", year: 2021, precision: "year" },
+      { id: "e4", domain: "family", year: 2023, precision: "year" },
+    ],
+    probes: [{
+      id: "p-cd",
+      semantic_key: "career.2015",
+      candidate_split_hash: "04:31|05:07",
+      domain: "career",
+      year: 2015,
+      question: "2016 年前后有没有高考或重要考试发挥失常？",
+      candidate_ids: ["04:31", "05:07"],
+      expected_outcomes: [
+        { answer_class: "yes", supports: ["04:31"], conflicts: ["05:07"] },
+        { answer_class: "no", supports: ["05:07"], conflicts: ["04:31"] },
+        { answer_class: "unsure", supports: [], conflicts: [] },
+      ],
+      information_gain: 0.4,
+      source: "dasha_boundary",
+    }],
+  });
+}
+
+function rangeNarrationDossier(inference: ReturnType<typeof buildInferenceState>) {
+  return dossierFixture({
+    latestResult: candidateSnapshotFixture({
+      decisionReceipt: { inference_state: inference },
+    }),
+    conversationSummary: conversationSummaryFixture({
+      activeFocus: activeFocusFixture({
+        expectedAnswerSchema: {
+          choice: {
+            prompt: "2016 年前后，有没有明显高考或重要考试发挥失常？",
+            option_a: "是，大概就在那段时间",
+            option_b: "有类似，但年份不对或不够重大",
+            option_c: "没有明显发生",
+            option_d: "不记得 / 不确定",
+            options: [
+              { key: "A", label: "是，大概就在那段时间", answer_class: "yes" },
+              { key: "B", label: "有类似，但年份不对或不够重大", answer_class: "weak_yes" },
+              { key: "C", label: "没有明显发生", answer_class: "no" },
+              { key: "D", label: "不记得 / 不确定", answer_class: "unsure" },
+            ],
+          },
+          probe_id: "p-cd",
+          semantic_key: "career.2015",
+        },
+      }),
+    }),
+  });
+}
+
 function choiceDossier() {
   const inference = inferenceState();
   const snapshot = candidateSnapshotFixture({
@@ -1207,8 +1270,10 @@ test("a scoring choice narrates cluster movement and range change", () => {
       { range: ["04:31", "04:39"], delta: 2 },
       { range: ["05:00", "05:07"], delta: -2 },
     ],
-    rangeBefore: ["04:31", "05:07"],
-    rangeAfter: ["04:31", "04:39"],
+    // 旧：rangeBefore/rangeAfter 直喂两个不同搜索窗口 → 新：入参改名 credibleBefore/credibleAfter，仍测纯函数拼句
+    // 原因：直喂拦不住 applyRectificationChoice 把 range_start/range_end 当成范围
+    credibleBefore: ["04:31", "05:07"],
+    credibleAfter: ["04:31", "04:39"],
   });
   assert.match(narrowed, /领先/);
   assert.match(narrowed, /落后/);
@@ -1218,8 +1283,8 @@ test("a scoring choice narrates cluster movement and range change", () => {
     scoring: true,
     appliedInference: true,
     deltasByCluster: [{ range: ["04:31", "04:39"], delta: 1 }],
-    rangeBefore: ["04:31", "04:39"],
-    rangeAfter: ["04:31", "04:39"],
+    credibleBefore: ["04:31", "04:39"],
+    credibleAfter: ["04:31", "04:39"],
   });
   assert.match(unchanged, /范围没变/);
   const unsure = composeChoiceNarration({
@@ -1227,12 +1292,80 @@ test("a scoring choice narrates cluster movement and range change", () => {
     scoring: true,
     appliedInference: true,
     answerClass: "unsure",
-    rangeBefore: ["04:31", "04:39"],
-    rangeAfter: ["04:31", "04:39"],
+    credibleBefore: ["04:31", "04:39"],
+    credibleAfter: ["04:31", "04:39"],
   });
   assert.equal(unsure, "已记录。这题先不计分，换一件事问。");
 });
 
+test("answering A narrates the credible range even when the search window stays put", async () => {
+  const previous = rangeNarrationInference(10, 3);
+  assert.deepEqual([previous.range_start, previous.range_end], ["04:31", "05:07"]);
+  assert.deepEqual(previous.credible_range, ["04:31", "05:07"]);
+  const reduced = applyChoiceWithoutEvidence(previous, {
+    choiceKey: "A",
+    schema: {
+      probe_id: "p-cd",
+      semantic_key: "career.2015",
+      options: [
+        { key: "A", label: "是，大概就在那段时间", answer_class: "yes" },
+      ],
+    },
+  });
+  assert.equal(reduced.applied, true);
+  assert.deepEqual([reduced.state.range_start, reduced.state.range_end], ["04:31", "05:07"]);
+  assert.deepEqual(reduced.state.credible_range, ["04:31", "04:31"]);
+  const applied = await applyRectificationChoice(persistChoiceAccounting(rangeNarrationDossier(previous)).client, {
+    userId: USER_ID,
+    caseId: CASE_ID,
+    sessionId: SESSION_ID,
+    actionId: ACTION_ID,
+    action: CHOICE_ACTION,
+    focusId: FOCUS_ID,
+    questionId: QUESTION_ID,
+    probeId: "p-cd",
+    optionId: "A",
+    expectedRevision: previous.revision,
+  });
+  assert.match(applied.narration, /范围从 04:31–05:07 收到 04:31–04:31/);
+  assert.doesNotMatch(applied.narration, /范围没变/);
+  const source = readFileSync(new URL("../src/lib/rectification-agentic/v9/answer-choice.ts", import.meta.url), "utf8");
+  assert.match(source, /credibleBefore:\s*previous\.credible_range/);
+  assert.match(source, /credibleAfter:\s*applied\.state\.credible_range/);
+  assert.doesNotMatch(source, /rangeBefore:\s*\[previous\.range_start/);
+});
+
+test("answering A says the range did not change when the credible range stays put", async () => {
+  const previous = rangeNarrationInference(10, 10);
+  assert.deepEqual(previous.credible_range, ["04:31", "05:07"]);
+  const reduced = applyChoiceWithoutEvidence(previous, {
+    choiceKey: "A",
+    schema: {
+      probe_id: "p-cd",
+      semantic_key: "career.2015",
+      options: [
+        { key: "A", label: "是，大概就在那段时间", answer_class: "yes" },
+      ],
+    },
+  });
+  assert.deepEqual([reduced.state.range_start, reduced.state.range_end], ["04:31", "05:07"]);
+  assert.deepEqual(reduced.state.credible_range, ["04:31", "05:07"]);
+  const applied = await applyRectificationChoice(persistChoiceAccounting(rangeNarrationDossier(previous)).client, {
+    userId: USER_ID,
+    caseId: CASE_ID,
+    sessionId: SESSION_ID,
+    actionId: ACTION_ID,
+    action: CHOICE_ACTION,
+    focusId: FOCUS_ID,
+    questionId: QUESTION_ID,
+    probeId: "p-cd",
+    optionId: "A",
+    expectedRevision: previous.revision,
+  });
+  assert.match(applied.narration, /范围没变/);
+  assert.doesNotMatch(applied.narration, /范围从 /);
+});
+
 test("the public agent route treats structured choice as a non-model command", () => {
   const route = readFileSync(new URL("../src/app/api/rectification/agent/route.ts", import.meta.url), "utf8");
   const start = route.indexOf("if (isStructuredChoice)");

frontend/tests/rectification-choice-card.test.ts

@@ -130,6 +130,9 @@ test("choice frames ask one biographical event from a server probe, not competin
   assert.equal(mergeChoiceCard(frame, null), null);
   assert.match(frame.why_user, /2016 年前后/);
   assert.match(frame.why_user, /分成两组/);
+  assert.match(frame.why_user, /Vimshottari/);
+  assert.match(frame.why_user, /Narayana/);
+  assert.doesNotMatch(frame.why_user, /毗湿奴多利|那罗延/);
   assert.match(frame.answer_impact.A, /05:00/);
   assert.match(frame.answer_impact.A, /领先/);
   assert.equal(frame.answer_impact.D, "不计分，换一题");

scripts/rectification/api_service.py

@@ -427,6 +427,8 @@ def block_scan(request: RectificationRequest) -> dict[str, Any]:
         row for row in scored.get("candidate_scores") or []
         if isinstance(row, dict) and str(row.get("time") or "")[:5]
     ]
+    day_scores = [float(row.get("score") or 0) for row in rows]
+    min_day = min(day_scores) if day_scores else 0.0
     raw_support: list[float] = []
     blocks: list[dict[str, Any]] = []
     for period, start_time, end_time in BLOCK_SCAN_PERIODS:
@@ -448,7 +450,9 @@ def block_scan(request: RectificationRequest) -> dict[str, Any]:
                 "domain": event.get("domain"),
                 "summary": event.get("summary"),
             })
-        raw_support.append(sum(float(row.get("score") or 0) for row in members))
+        member_scores = [float(row.get("score") or 0) for row in members]
+        mean = (sum(member_scores) / len(member_scores)) if member_scores else 0.0
+        raw_support.append(max(mean - min_day, 0.0))
         blocks.append({
             "period": period,
             "start_time": start_time,

tests/test_rectification_v5_services.py

@@ -908,6 +908,46 @@ class RectificationV5ServicesTest(unittest.TestCase):
         self.assertEqual(result["candidate_count"], 144)
         self.assertLessEqual(elapsed, 15, f"block_scan took {elapsed:.1f}s")
 
+    def test_block_scan_equal_scores_share_evenly_and_length_does_not_win(self):
+        from scripts.rectification.api_service import block_scan
+
+        def grid(score_for):
+            return [
+                {"time": f"{hour:02d}:{minute:02d}", "score": score_for(hour), "supporting_event_ids": []}
+                for hour in range(24)
+                for minute in range(0, 60, 10)
+            ]
+
+        equal_rows = grid(lambda _hour: 12.0)
+        morning_rows = grid(lambda hour: 20.0 if 4 <= hour <= 7 else 10.0)
+        dummy = {
+            "result_id": "00000000-0000-4000-8000-000000000099",
+            "algorithm_version": "test",
+            "calculation_spec": {},
+            "calculation_spec_hash": "abc",
+            "decision_receipt": {},
+        }
+        request_body = {
+            "birth_date": "1998-03-15",
+            "start_time": "00:00",
+            "end_time": "23:59",
+            "lat": 39.9042,
+            "lon": 116.4074,
+            "tz": 8.0,
+            "minute_step": 10,
+            "events": [],
+        }
+        with patch("scripts.rectification.api_service.score_candidates", return_value={**dummy, "candidate_scores": equal_rows}):
+            equal = block_scan(request_body)
+        self.assertEqual([row["relative_support"] for row in equal["blocks"]], [20.0, 20.0, 20.0, 20.0, 20.0])
+        self.assertEqual([row["candidate_count"] for row in equal["blocks"]], [24, 24, 36, 30, 30])
+        with patch("scripts.rectification.api_service.score_candidates", return_value={**dummy, "candidate_scores": morning_rows}):
+            morning = block_scan(request_body)
+        shares = {row["period"]: row["relative_support"] for row in morning["blocks"]}
+        self.assertEqual(max(shares, key=shares.get), "early_morning")
+        self.assertGreater(shares["early_morning"], shares["afternoon"])
+        self.assertEqual(shares["afternoon"], 0.0)
+
 
 if __name__ == "__main__":
     unittest.main()