fix(rectification): use candidate dates for cross-midnight dasha scoring

Add date-isolated caches and regression coverage, align scoring identity, and freeze full research reruns while preserving historical artifacts. Record unresolved cache/receipt identity and end-to-end acceptance gaps for branch review only. Co-Authored-By: Claude Code <noreply@anthropic.com>
2026-09-20 13:56:11 +08:00
parent 03cba4780a
commit aa46da1016
49 changed files with 62080 additions and 109 deletions
@@ -8,7 +8,9 @@ This file is the small index for the current engineering fronts that still drive
 - [固定口径重跑与契约](sealed_holdout_rerun_2026_09_20.md)：已曝光 v3 的当前实现成绩，不是新的独立盲测；确认门关闭。
 - [v5 独立封存采集协议](sealed_holdout_v5_protocol_2026_09_20.md)：本轮只出协议，不采事件。
 - 前置结论：[分钟分辨率两轮研究](rectification_minute_resolution_closure_2026_09_14.md)。不重走已证伪的加权重与簇合并假设。
- 执行状态：[本轮进度](../tasks/PROGRESS-rectification-validation-20260920.md)。
+- 验证补缺轮：[进度](../tasks/PROGRESS-rectification-validation-20260920.md)。
+- 后续候选级跨午夜日期修复：[进度与独立验收](../tasks/PROGRESS-rectification-cross-midnight-20260920.md)。当前评测转至 `reported_offset_cross_midnight_2026_09_20.json` / `sealed_holdout_rerun_cross_midnight_2026_09_20.json`；旧 JSON / freeze 原字节留存，旧文档及评测器在 `history/rectification_pre_cross_midnight_2026_09_20/`。
+- 尚未闭环：BUG-982 跨午夜簇跨度、BUG-983 凌晨日期锚点、BUG-984 时段缓存/回执身份；[验收补单](../tasks/TASK-rectification-cross-midnight-dasha-fix-20260920.md) 不代表批准实施。

 ## Shortest-Path Closure Order (2026-06-29)

@@ -0,0 +1,64 @@
+# 申报偏差敏感性评测（2026-09-20）
+
+## 结论与边界
+
+**默认半径 ±15 分钟时，申报误差绝对值超过 15 分钟，记录真值就不在候选窗内；按整数分钟，第 16 分钟开始出窗。** 这是搜索窗几何边界，不需要知道真实用户偏差分布。
+
+本轮完整评测预先固定 15 个偏移 × 3 档半径 × 20 例，共 900 组合。修正版全部运行完成，validator 排除 0 例；实测结果以下列最终 JSON 为准，不能用中途旧版本初扫代替。
+
+**区间覆盖不等于分钟命中，模拟偏差敏感性不等于真实用户准确率。** 不运行六题真值方向回放，不以用户认可作真值，不调参、不改生产打分、不打开确认门。
+
+## 口径
+
+| 项目 | 口径 |
+| --- | --- |
+| 数据集 | `references/real_case_calibration/minute_rectification_holdout_v3.json`，20 例公开 AA，每例 3 件事件，已曝光 |
+| ayanamsa / node mode | `raman` / `mean` |
+| 申报偏移 | 0、±3、±5、±8、±10、±15、±20、±30 分钟 |
+| 搜索半径 | ±15 / ±30 / ±60 分钟；±120 仅说明几何边界，未评分 |
+| 候选步长 | 1 分钟，含两端；不是生产 2 分钟抽样 |
+| 12 文件历史身份 | `b15d9ea15227cd58`（完整值见 JSON） |
+| 实际评分路径 | 原生 event contribution matrix；不是 T2 的 shadow fact ranker |
+| 额外身份 | JSON 的 `production_scoring_sha256` 与 `research_implementation_sha256` 分别绑定扩展评分文件和研究适配文件；显式文件清单不是完整传递依赖锁 |
+| 头名与排名 | 分数降序，再用既有 `_opaque_winner` 同源 SHA-256 规则排同分；不选择最接近真值的候选 |
+| 交付区间 | 初始未答题：signature clusters 中落后头名不足 8 分的仍有效簇外包范围；无新回答、无淘汰，不是完整问答链交付效果 |
+| 跨午夜 | 候选、误差及区间使用完整日期；离线矩阵按候选日期分组，合并后全窗排序/聚类 |
+| 独立性 | `is_blind_evaluation=false`，已曝光集；评分/区间定稿后才揭示标签，仅证明程序标签隔离 |
+
+## 实测表
+
+每格依次为 **真值在窗比例 / 哈希头名命中 / 初始交付区间覆盖**，分母均为 20 例。列为申报偏移分钟，行为搜索半径。统一口径：v3、raman/mean、1 分钟步长；历史打分身份 `b15d9ea15227cd58`，扩展原生评分身份 `115c3fbcffdaff49`，研究适配身份 `8d0c613dc0900adb`。
+
+| 半径 | -30 | -20 | -15 | -10 | -8 | -5 | -3 | 0 | +3 | +5 | +8 | +10 | +15 | +20 | +30 |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
+| ±15 | 0/0/0 | 0/0/0 | 100/10/95 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/95 | 100/15/95 | 0/0/0 | 0/0/0 |
+| ±30 | 100/10/95 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/95 | 100/15/95 |
+| ±60 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 | 100/5/100 |
+
+表内全部数字单位为 %。这只是 20 例公开已曝光案例的条件敏感性，不附会总体置信度。即使真值在候选窗中，±15 的 -15/+10/+15 档及 ±30 的 -30/+20/+30 档仍各有一例未被初始交付范围覆盖；窗口包含与交付覆盖必须分开。±60 的所有采样偏移虽保持覆盖，但真正选中的头名只有 5%，不能用宽区间覆盖声称分钟准确。
+
+修正版与错误初版的汇总比例恰好相同，不表示初版评分正确；身份、逐候选计算与跨日回归以修正版为准。
+
+## 放宽能救回什么
+
+- 真值在候选窗的条件始终是 `abs(offset) <= radius`。本次偏移最大绝对值 30，因此半径 30 / 60 / 120 在几何上都能包含本次所有偏移标签。
+- 对 ±15 已出窗的 ±20 / ±30 两组偏移，放宽到 ±30 可把**候选窗包含率**从 0 恢复到 100%；±60 / ±120 对这些特定偏移不再增加几何包含率。
+- 半径 ±30 自第 31 个整数分钟偏移出窗，±60 自第 61 个、±120 自第 121 个。放宽不是无条件保证，更不代表排名命中或交付覆盖同样恢复。
+- ±120 未跑评分，不报告其头名命中率或交付覆盖率。真实用户申报偏差分布未知，不估算现实用户中有多少人能被放宽救回。
+
+## 初扫作废与独立复核
+
+首版正确枚举了跨日候选，但矩阵内 transition proximity 仍共用窗口起日，独立审查发现跨日候选得分偏移。因此旧初扫作废，修正版标记 `replay_revision=candidate_date_grouped_v2`，并登记 `supersedes=initial_sweep_invalidated_cross_midnight_transition_anchor`。
+
+修复仅在本次新脚本 `score_window()` 做按日期分组适配，未改冻结打分文件。实引擎回归验证跨日窗口的分组计算与逐候选独立重算相同；聚类和交付范围在合并后全窗运行。**生产日期处理未在本单修复，不能称为生产端到端回放。**
+
+## 可复算产物与验收
+
+```bash
+python scripts/research/reported_offset_sweep.py --json
+python -m pytest tests/test_reported_offset_research.py tests/test_rectification_validation_integrity_gate.py -q
+```
+
+`reported_offset_2026_09_20.json` 保留逐例四项必要输出：真值是否在窗、真值排名、头名分钟误差、交付区间覆盖，以及每个格子的分母、全部参数和实现哈希。仅保存公开集序号，不保存出生分钟、坐标或事件正文。
+
+排名不能与 T2 的竞争排名 top-1/top-3 混比：T2 按旧协议允许并列多个分钟同时算第一；本表把同分完全按哈希打破。最终测试数字与已知基线失败见 [进度](../tasks/PROGRESS-rectification-validation-20260920.md)。真实分布采集与新未曝光样本见 [v5 协议](sealed_holdout_v5_protocol_2026_09_20.md)。
@@ -0,0 +1,43 @@
+{
+  "record_version": "exposed-v3-fixed-protocol-rerun-v1",
+  "frozen_at_utc": "2026-09-20T03:27:26.670869+00:00",
+  "dataset_path": "references/real_case_calibration/minute_rectification_holdout_v3.json",
+  "dataset_sha256": "45aa76ab79dfc6c6404b6252083e1aa5c7e2e2b7de811fc11042a8ad2da3f0ea",
+  "algorithm_version": "birth-time-event-fact-ranker-v4-shadow",
+  "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+  "files": [
+    "scripts/active_rectification_event_engine.py",
+    "scripts/active_rectification_events.py",
+    "scripts/ashtakavarga.py",
+    "scripts/dasha_analyzer.py",
+    "scripts/divisional_charts_extended.py",
+    "scripts/domain_calculation_service.py",
+    "scripts/jaimini.py",
+    "scripts/minute_rectification_fact_ranker_v4.py",
+    "scripts/minute_rectification_feature_facts_v4.py",
+    "scripts/narayana_dasha.py",
+    "scripts/shadbala.py",
+    "scripts/varga.py"
+  ],
+  "historical_frozen_sha256": "f41c298dd6cdcebe7a93e632f7191954be7987f6012ca2a34cecb6e447fbf196",
+  "evaluator_sha256": "3bf35de5f8a0b9588808178d47ba2457c398cf2fd2d2fcb03c5c71420f998da5",
+  "ayanamsa": "raman",
+  "node_mode": "mean",
+  "candidate_radius_minutes": [
+    10
+  ],
+  "minute_step": 1,
+  "release_metrics": {
+    "top_1_rate_minimum": 0.6,
+    "top_3_rate_minimum": 0.85,
+    "mean_absolute_minute_error_maximum": 2.0,
+    "false_confirmation_rate_maximum": 0.05,
+    "correct_insufficient_evidence_rejection_rate_minimum": 0.9
+  },
+  "results_previously_seen": true,
+  "official_valid_independent_blind": false,
+  "must_not_use_for_tuning": true,
+  "tie_breaker": "sha256(benchmark_id:case_id:candidate_time); published minute is never passed to the ranker",
+  "metric_rank_definition": "competition_rank_1_plus_strictly_higher_scores_legacy_protocol",
+  "extra_metric_rank_definition": "score_desc_then_existing_opaque_sha256_total_order"
+}
@@ -0,0 +1,626 @@
+{
+  "scope": "fixed_protocol_previously_exposed_v3_rerun",
+  "evaluated_on": "2026-09-20",
+  "frozen_record": {
+    "record_version": "exposed-v3-fixed-protocol-rerun-v1",
+    "frozen_at_utc": "2026-09-20T03:27:26.670869+00:00",
+    "dataset_path": "references/real_case_calibration/minute_rectification_holdout_v3.json",
+    "dataset_sha256": "45aa76ab79dfc6c6404b6252083e1aa5c7e2e2b7de811fc11042a8ad2da3f0ea",
+    "algorithm_version": "birth-time-event-fact-ranker-v4-shadow",
+    "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+    "files": [
+      "scripts/active_rectification_event_engine.py",
+      "scripts/active_rectification_events.py",
+      "scripts/ashtakavarga.py",
+      "scripts/dasha_analyzer.py",
+      "scripts/divisional_charts_extended.py",
+      "scripts/domain_calculation_service.py",
+      "scripts/jaimini.py",
+      "scripts/minute_rectification_fact_ranker_v4.py",
+      "scripts/minute_rectification_feature_facts_v4.py",
+      "scripts/narayana_dasha.py",
+      "scripts/shadbala.py",
+      "scripts/varga.py"
+    ],
+    "historical_frozen_sha256": "f41c298dd6cdcebe7a93e632f7191954be7987f6012ca2a34cecb6e447fbf196",
+    "evaluator_sha256": "3bf35de5f8a0b9588808178d47ba2457c398cf2fd2d2fcb03c5c71420f998da5",
+    "ayanamsa": "raman",
+    "node_mode": "mean",
+    "candidate_radius_minutes": [
+      10
+    ],
+    "minute_step": 1,
+    "release_metrics": {
+      "top_1_rate_minimum": 0.6,
+      "top_3_rate_minimum": 0.85,
+      "mean_absolute_minute_error_maximum": 2.0,
+      "false_confirmation_rate_maximum": 0.05,
+      "correct_insufficient_evidence_rejection_rate_minimum": 0.9
+    },
+    "results_previously_seen": true,
+    "official_valid_independent_blind": false,
+    "must_not_use_for_tuning": true,
+    "tie_breaker": "sha256(benchmark_id:case_id:candidate_time); published minute is never passed to the ranker",
+    "metric_rank_definition": "competition_rank_1_plus_strictly_higher_scores_legacy_protocol",
+    "extra_metric_rank_definition": "score_desc_then_existing_opaque_sha256_total_order"
+  },
+  "implementation_hash_matches_at_replay": true,
+  "dataset_hash_matches_at_replay": true,
+  "source_audit_status": "corrected_known_date_errors",
+  "validation_status": "blocked_awaiting_public_aa_cases",
+  "valid_public_aa_cases": 20,
+  "excluded_cases": [],
+  "trial_count": 20,
+  "trials": [
+    {
+      "case_ordinal": 1,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 5,
+      "opaque_true_rank": 8,
+      "minute_error": 10,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 2,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 2,
+      "opaque_true_rank": 8,
+      "minute_error": 3,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "insufficient_events",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 3,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 12,
+      "minute_error": 5,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 4,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 1,
+      "minute_error": 0,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "insufficient_events",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 5,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 10,
+      "minute_error": 1,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 6,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 7,
+      "opaque_true_rank": 12,
+      "minute_error": 6,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 7,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 5,
+      "opaque_true_rank": 7,
+      "minute_error": 9,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 8,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 5,
+      "opaque_true_rank": 8,
+      "minute_error": 10,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "insufficient_events",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 9,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 3,
+      "minute_error": 1,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "missing_mandatory_layers",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "missing_mandatory_layers",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 10,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 8,
+      "minute_error": 5,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 11,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 6,
+      "opaque_true_rank": 17,
+      "minute_error": 5,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 12,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 4,
+      "minute_error": 5,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 13,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 5,
+      "opaque_true_rank": 7,
+      "minute_error": 10,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 14,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 17,
+      "opaque_true_rank": 19,
+      "minute_error": 10,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 15,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 7,
+      "minute_error": 9,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "missing_mandatory_layers",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "missing_mandatory_layers",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 16,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 12,
+      "opaque_true_rank": 12,
+      "minute_error": 10,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 17,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 6,
+      "opaque_true_rank": 6,
+      "minute_error": 6,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 18,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 11,
+      "opaque_true_rank": 14,
+      "minute_error": 10,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "insufficient_events",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 19,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 6,
+      "minute_error": 4,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    },
+    {
+      "case_ordinal": 20,
+      "candidate_count": 21,
+      "event_count": 3,
+      "true_rank": 1,
+      "opaque_true_rank": 7,
+      "minute_error": 10,
+      "would_confirm": false,
+      "false_confirmation": false,
+      "insufficient_evidence_rejected": true,
+      "full_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ],
+      "sparse_trial_reasons": [
+        "unique_minute_not_found",
+        "insufficient_events",
+        "insufficient_domains",
+        "neighbor_stability_not_passed",
+        "leave_one_event_out_not_passed",
+        "top_candidate_feature_not_unique",
+        "fact_ranker_v4_holdout_not_ready"
+      ]
+    }
+  ],
+  "metrics": {
+    "top_1_rate": 0.45,
+    "top_3_rate": 0.5,
+    "mean_absolute_minute_error": 6.45,
+    "false_confirmation_rate": 0.0,
+    "correct_insufficient_evidence_rejection_rate": 1.0,
+    "confirmation_coverage_rate": 0.0
+  },
+  "metric_gates_passed": false,
+  "opaque_exact_top_1_rate": 0.05,
+  "opaque_exact_top_3_rate": 0.1,
+  "official_valid_independent_blind": false,
+  "official_blind_trial_count": 0,
+  "is_blind_evaluation": false,
+  "truth_hidden_from_ranker": true,
+  "results_previously_seen": true,
+  "verified_minute_claim_allowed": false,
+  "status": "blocked_independent_blind_evidence",
+  "boundary": "Three-event low-information protocol, not a mathematical accuracy lower bound and not representative of real sessions. Historical v3/v4 exposure cannot be undone by freezing today's scorer. No tuning or release claims."
+}
@@ -0,0 +1,72 @@
+# 当前冻结实现的 v3 固定口径重跑（2026-09-20）
+
+## 结论
+
+**已补上当前实现可归属、可复算的成绩；没有补成新的独立官方盲测。** 既有 v3/v4 人物与成绩已经曝光，重新冻结今天的打分文件不能消除曝光。任务书 T2 中「首次口径干净的官方盲测」在既有 BUG-427 / BUG-428 红线下仍为 **blocked**，不得改名通过。确认门保持关闭。
+
+这是每例只有 3 件事的低信息量（任务书称「下限」）口径，不代表真实会话；事件少不等于数学上保证成绩更低，因此也不是准确率下界。
+
+## 固定口径与可审计文件
+
+| 项目 | 本次实际值 |
+| --- | --- |
+| 数据集 | `references/real_case_calibration/minute_rectification_holdout_v3.json` |
+| 数据集 SHA-256 | `45aa76ab79dfc6c6404b6252083e1aa5c7e2e2b7de811fc11042a8ad2da3f0ea` |
+| 计算规格 | ayanamsa `raman` / node mode `mean` |
+| 半径 / 步长 | ±10 分钟 / 1 分钟，含两端；真值为窗心 |
+| 算法 | `birth-time-event-fact-ranker-v4-shadow`，不是生产 V9 问答链 |
+| 当前打分身份 | `b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18`（前缀 `b15d9ea15227cd58`） |
+| 样本 / 事件 | 20 例 / 每例 3 件；validator 排除 0 例 |
+| 资料审计状态 | `corrected_known_date_errors`，不等同全新独立人工重审 |
+| 排名标签隔离 | 排序和稀疏证据判定完成后才读取真值标签 |
+| 独立官方盲测 | `official_valid_independent_blind=false`；有效官方盲测试次仍为 0 |
+
+冻结记录：`sealed_holdout_rerun_2026_09_20.freeze.json`。先以独占创建模式记录 12 个文件、数据集与评测器字节哈希，再执行评分；评分前验证全部身份，任何漂移即拒绝。历史 v3 `frozen_scoring` 原值 `f41c298dd6cdcebe…` 未改。
+
+脱敏逐例结果：`sealed_holdout_rerun_2026_09_20.json`，只含公开集序号、排名、误差、门禁原因，不含出生分钟、坐标或事件正文。复算：
+
+```bash
+python scripts/research/sealed_holdout_rerun.py --json
+```
+
+## 五项预注册指标与门槛
+
+本表统一口径：v3 / `raman` / `mean` / ±10 / 1 分钟 / `b15d9ea15227cd58`。
+
+| 指标 | 实测 | 预注册门槛 | 数值对照 |
+| --- | ---: | ---: | --- |
+| top-1（旧协议竞争排名，含并列） | 0.45 | ≥0.60 | 未通过 |
+| top-3（旧协议竞争排名，含并列） | 0.50 | ≥0.85 | 未通过 |
+| 头名平均绝对分钟误差（哈希打破并列） | 6.45 分钟 | ≤2.00 分钟 | 未通过 |
+| 误确认率（全样本分母） | 0.00 | ≤0.05 | 通过，但无确认覆盖，不能证明确认精度 |
+| 信息不足正确拒绝率（每例仅首件事件） | 1.00 | ≥0.90 | 通过 |
+| 确认覆盖率（补充） | 0.00 | 本单禁止抬高 | 保持关闭 |
+| 作为独立盲测发布成绩的资格 | 不适用 | 新封存、未曝光、独立审计 | blocked，不可当发布指标 |
+
+**重要口径纠正：旧 top-1/top-3 不是哈希选中那个分钟的命中率。** 现有评测的 `true_rank = 1 + 严格高于真值分数的候选数` 会把同分的多个分钟全算第一；本次为保证与旧协议可比保留它，同时增加哈希全序的真实选中指标，不暗改预注册定义。
+
+| 附加指标（同一上述口径） | 实测 |
+| --- | ---: |
+| `_opaque_winner` 真正选中的头名命中 | 0.05 |
+| 同一哈希全序中的 top-3 命中 | 0.10 |
+
+这两个数字不能与旧 0.45/0.50 混称同一种准确率。没有据此改权重、换半径、换案例或剔除失败例。
+
+## 契约改了什么、没有改什么
+
+- `current_tree_scorer` 绑定本次真实哈希及新报告，新增固定口径试次数与成绩；旧 `current_tree_unfrozen_diagnostic` 历史块保留。
+- `source_audit_status` 对齐数据集；`evaluated_on` 表示最新固定口径重跑日期，并新增作用域说明。历史顶层指标仍有 `metrics_produced_by` 的旧日期与来源，未伪装成本次结果。
+- `status`、`valid_public_aa_cases`、`required_cases`、`top_1_rate`、`confirmation_coverage_rate`、`sealed_benchmark_id` 六个 runtime 键的值与类型全不变；尤其 `status=not_ready`、确认覆盖率为零。
+- `official_eval_implementation_hash_matches=false`、`official_eval_trial_count=0` 保留；新增重跑记录不冒充已重新获得独立性。
+- 未改任何冻结打分文件、生产判定、现有 `_candidate_moments()` 或 `request_from_case()`。
+
+## 防复发与既有断言变更
+
+新增 `tests/test_sealed_holdout_contract_freshness.py` 钉死当前树身份、审计状态、冻结记录、报告汇总与独立性声明；冻结身份不符必须在计算前拒绝。`tests/test_rectification_validation_integrity_gate.py` 将其和偏差评测回归纳入现有 `test_rectification_*.py` 快速门，不改 workflow。
+
+| 原断言 | 新断言 | 原因 |
+| --- | --- | --- |
+| Python / 前端：`current_tree_scorer.implementation_sha256` 等于旧 post-audit sidecar 的哈希 | 等于新固定口径报告 `frozen_record.implementation_sha256`，另验证当前树与新报告 | 契约当前树块的含义就是当前实现；继续绑定旧报告会强制过期。新增身份断言更严格，不弱化门禁 |
+| 官方标志 false、官方试次 0、六键等于历史报告、确认门全套行为断言 | 原值保留，另加固定口径重跑身份/试次和独立盲测 false | 固定口径重跑不是新的独立盲测，不借元数据迁移打开确认门 |
+
+本机 Python 3.11.7 可跑；`python3` Windows 别名退出 49。该任务定向命令的最终测试清单与基线失败对照由 `PROGRESS-rectification-validation-20260920.md` 统一记录。特别是既有 v2 冻结身份测试不属本次刷新范围，不得为了通过而改其历史封存或断言。
@@ -0,0 +1,55 @@
+{
+  "scope": "immutable_pre_fix_artifacts_not_current_results",
+  "baseline_commit": "932f2fff",
+  "reason": "BUG-981 production candidate-date fix; prior T1 used offline date grouping and T2 shadow scorer; numerical changes not assumed",
+  "files": [
+    {
+      "source_path": "docs/research/reported_offset_2026_09_20.md",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/docs/research/reported_offset_2026_09_20.md",
+      "sha256": "1033fe68ff4b2802ab321ef2f76e558dc21c3c534486954bf03af8b7f114f6df",
+      "size_bytes": 6018
+    },
+    {
+      "source_path": "docs/research/reported_offset_2026_09_20.json",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/docs/research/reported_offset_2026_09_20.json",
+      "sha256": "9878f2b50c957a470fafcb2ed0a9eb16405c3f6fa422ea50f467e84b0189ace1",
+      "size_bytes": 324672
+    },
+    {
+      "source_path": "docs/research/sealed_holdout_rerun_2026_09_20.md",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/docs/research/sealed_holdout_rerun_2026_09_20.md",
+      "sha256": "dab1bf59748efab481aee394ab25822ee51b0665bf66c769afdfb4d53a492d29",
+      "size_bytes": 5818
+    },
+    {
+      "source_path": "docs/research/sealed_holdout_rerun_2026_09_20.json",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/docs/research/sealed_holdout_rerun_2026_09_20.json",
+      "sha256": "40df220bfcb51a683b31fbd626896e47a5aed7ee4f30ff669543691f7ce53c1e",
+      "size_bytes": 20052
+    },
+    {
+      "source_path": "docs/research/sealed_holdout_rerun_2026_09_20.freeze.json",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/docs/research/sealed_holdout_rerun_2026_09_20.freeze.json",
+      "sha256": "d17651cb50224acdd1af7a4692c777ed169b216a962a0675371254ce561d0921",
+      "size_bytes": 1898
+    },
+    {
+      "source_path": "references/rectification_sealed_holdout.v1.json",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/references/rectification_sealed_holdout.v1.json",
+      "sha256": "863806992ee7162e959bb319fd82bd75f13fadea6218e197dd4e78fd2ffd0279",
+      "size_bytes": 3705
+    },
+    {
+      "source_path": "scripts/research/reported_offset_sweep.py",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/scripts/research/reported_offset_sweep.py.txt",
+      "sha256": "87f87859bb399da1e860ab5e750e68088401a5a82fbba10ba709c17e5266167c",
+      "size_bytes": 11001
+    },
+    {
+      "source_path": "scripts/research/sealed_holdout_rerun.py",
+      "archive_path": "docs/research/history/rectification_pre_cross_midnight_2026_09_20/scripts/research/sealed_holdout_rerun.py.txt",
+      "sha256": "3bf35de5f8a0b9588808178d47ba2457c398cf2fd2d2fcb03c5c71420f998da5",
+      "size_bytes": 7482
+    }
+  ]
+}
@@ -0,0 +1,21 @@
+{
+  "scope": "superseded_preparation_audit",
+  "reason": "Existing report overwrite denied; retained initial freeze and completed 20-case preparation replay unchanged. Stopped 900 sweep before result write. Final report paths changed; evaluator identity re-frozen before new final replay.",
+  "files": [
+    {
+      "path": "docs/research/sealed_holdout_rerun_cross_midnight_2026_09_20.freeze.json",
+      "sha256": "6638e52191fed63ed3a2e99da149ba3269ea4367d8e60b8408f2b05165ca280e",
+      "status": "superseded_preparation_not_final_report"
+    },
+    {
+      "path": "docs/research/reported_offset_cross_midnight_2026_09_20.freeze.json",
+      "sha256": "80004557170595e7ae222ecf538482e2f1f279f3858896e2553ba0d565a45eea",
+      "status": "superseded_preparation_not_final_report"
+    },
+    {
+      "path": "docs/research/sealed_holdout_rerun_2026_09_20.cross_midnight.rerun.json",
+      "sha256": "f4b0ed8a8ad17604940f9022339854571501d95af863820f58501a4981d4e4e1",
+      "status": "superseded_preparation_not_final_report"
+    }
+  ]
+}
@@ -0,0 +1,79 @@
+{
+  "sealed_benchmark_id": "minute_rectification_fact_ranker_v4_holdout_v3",
+  "status": "not_ready",
+  "valid_public_aa_cases": 20,
+  "required_cases": 20,
+  "top_1_rate": 0.15,
+  "confirmation_coverage_rate": 0.0,
+  "previous_pilot_id": "minute_rectification_holdout_v2",
+  "dataset_path": "references/real_case_calibration/minute_rectification_holdout_v3.json",
+  "report_path": "references/real_case_calibration/minute_rectification_holdout_v3_report.json",
+  "source_audit_status": "corrected_known_date_errors",
+  "evaluated_on": "2026-09-20",
+  "evaluated_on_scope": "latest_fixed_protocol_rerun_not_historical_runtime_metrics",
+  "historical_source_audit_status": "invalidated_after_replay",
+  "report_status": "invalidated_after_source_audit",
+  "trial_count": 20,
+  "top_3_rate": 0.25,
+  "mean_absolute_minute_error": 6.95,
+  "metrics_produced_by": {
+    "implementation_sha256": "f41c298dd6cdcebe7a93e632f7191954be7987f6012ca2a34cecb6e447fbf196",
+    "algorithm_version": "birth-time-event-fact-ranker-v4-shadow",
+    "implementation_hash_matches_at_replay": true,
+    "source_report": "references/real_case_calibration/minute_rectification_holdout_v3_report.json",
+    "evaluated_on": "2026-07-21"
+  },
+  "current_tree_scorer": {
+    "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+    "matches_metrics_scorer": false,
+    "official_eval_implementation_hash_matches": false,
+    "official_eval_trial_count": 0,
+    "source_report": "docs/research/sealed_holdout_rerun_2026_09_20.json",
+    "fixed_protocol_rerun_trial_count": 20,
+    "fixed_protocol_rerun_hash_matches": true,
+    "metrics": {
+      "top_1_rate": 0.45,
+      "top_3_rate": 0.5,
+      "mean_absolute_minute_error": 6.45,
+      "false_confirmation_rate": 0.0,
+      "correct_insufficient_evidence_rejection_rate": 1.0,
+      "confirmation_coverage_rate": 0.0
+    }
+  },
+  "current_tree_fixed_protocol_rerun": {
+    "evaluated_on": "2026-09-20",
+    "report_path": "docs/research/sealed_holdout_rerun_2026_09_20.json",
+    "freeze_record_path": "docs/research/sealed_holdout_rerun_2026_09_20.freeze.json",
+    "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+    "scorer_frozen_before_rerun": true,
+    "source_audit_status": "corrected_known_date_errors",
+    "trial_count": 20,
+    "official_valid_independent_blind": false,
+    "is_blind_evaluation": false,
+    "truth_hidden_from_ranker": true,
+    "results_previously_seen": true,
+    "must_not_claim_as_release_metrics": true,
+    "must_not_use_for_tuning": true,
+    "verified_minute_claim_allowed": false,
+    "metric_gates_passed": false,
+    "events_per_case": 3,
+    "boundary": "First current-hash fixed-protocol rerun in this task, not a first independent official blind evaluation. Three-event low-information protocol is not representative of real sessions and is not a mathematical accuracy lower bound. Historical v3/v4 score exposure requires a fresh sealed set."
+  },
+  "current_tree_unfrozen_diagnostic": {
+    "is_blind_evaluation": false,
+    "results_already_seen": true,
+    "scorer_frozen": false,
+    "source_audit_passed": false,
+    "must_not_claim_as_release_metrics": true,
+    "must_not_use_for_tuning": true,
+    "report_path": "references/real_case_calibration/minute_rectification_holdout_v3_post_audit_diagnostic_report.json",
+    "implementation_sha256": "99730c84c6434f52669a05e4d9a4a87df0a218e3237b5436b933252f2028384e",
+    "trial_count": 20,
+    "top_1_rate": 0.45,
+    "top_3_rate": 0.5,
+    "mean_absolute_minute_error": 5.8,
+    "confirmation_coverage_rate": 0.0,
+    "metric_gates_passed": false,
+    "verified_minute_claim_allowed": false
+  }
+}
@@ -0,0 +1,211 @@
+#!/usr/bin/env python3
+"""Reported-centre sensitivity on exposed public cases, without oracle answers."""
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from datetime import datetime, timedelta
+from pathlib import Path
+from typing import Any
+
+ROOT = Path(__file__).resolve().parents[2]
+if str(ROOT) not in sys.path:
+    sys.path.insert(0, str(ROOT))
+
+from scripts.active_rectification_event_engine import (
+    AYANAMSA, NODE_MODE, compute_candidate_static_contexts,
+)
+from scripts.minute_rectification_blind_eval import _opaque_winner, implementation_sha256
+from scripts.minute_rectification_holdout_validator import validate
+from scripts.rectification.candidate_contrast import select_signature_representatives
+from scripts.rectification.scoring_service import build_event_contribution_matrix, score_from_matrix
+from scripts.research.cluster_width_lib import SEPARATION_LEAD, still_valid_public
+from scripts.research.probe_supply_after_six import request_from_case
+from scripts.research.sealed_holdout_rerun import DATASET, file_sha256, opaque_order
+
+OFFSETS = (-30, -20, -15, -10, -8, -5, -3, 0, 3, 5, 8, 10, 15, 20, 30)
+RADII = (15, 30, 60)
+MINUTE_STEP = 1
+PRODUCTION_FILES = [
+    "scripts/rectification/scoring_service.py",
+    "scripts/rectification/dasha_transition_proximity.py",
+    "scripts/rectification/candidate_contrast.py",
+    "scripts/rectification/case_holdout.py",
+    "scripts/rectification/contracts.py",
+    "scripts/rectification/event_probes.py",
+]
+RESEARCH_FILES = [
+    "scripts/research/reported_offset_sweep.py",
+    "scripts/research/cluster_width_lib.py",
+    "scripts/research/probe_supply_after_six.py",
+    "scripts/research/sealed_holdout_rerun.py",
+    "scripts/minute_rectification_blind_eval.py",
+]
+
+
+def shifted_window(case: dict[str, Any], offset: int, radius: int) -> tuple[dict[str, Any], list[datetime]]:
+    if radius not in (*RADII, 120):
+        raise ValueError("unsupported_radius")
+    true_at = datetime.fromisoformat(f"{case['birth']['date']}T{case['birth']['time']}:00")
+    reported_at = true_at + timedelta(minutes=offset)
+    start = reported_at - timedelta(minutes=radius)
+    end = reported_at + timedelta(minutes=radius)
+    candidates = [start + timedelta(minutes=i) for i in range(0, radius * 2 + 1, MINUTE_STEP)]
+    # Reuse only the established event normalization, then replace the window.
+    # The ranker sees neither the truth label nor the simulated offset.
+    request = request_from_case(case)
+    request.update({
+        "birth_date": start.date().isoformat(),
+        "start_time": start.strftime("%H:%M"),
+        "end_time": end.strftime("%H:%M"),
+        "minute_step": MINUTE_STEP,
+        "ayanamsa": AYANAMSA,
+        "node_mode": NODE_MODE,
+    })
+    return request, candidates
+
+
+def score_window(request: dict[str, Any], contexts: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    """Keep auxiliary transition anchors on each candidate's actual date.
+
+    The production matrix's transition-proximity helper accepts one birth date
+    per call, unlike the static chart layer which reads candidate_at. Grouping
+    is only an offline adapter; it does not change the production scorer.
+    """
+    by_date: dict[str, list[dict[str, Any]]] = {}
+    for context in contexts:
+        by_date.setdefault(context["candidate_at"].date().isoformat(), []).append(context)
+    by_time = {}
+    for candidate_date, group in by_date.items():
+        dated_request = {**request, "birth_date": candidate_date}
+        built = build_event_contribution_matrix(dated_request, static_contexts=group)
+        by_time.update({row["time"]: row for row in score_from_matrix(dated_request, built)})
+    return [by_time[context["candidate_at"].strftime("%H:%M")] for context in contexts]
+
+
+def delivery_moments(public: list[dict[str, Any]], candidates: list[datetime]) -> list[datetime]:
+    """Initial delivery envelope in date-aware order; no truth-based narrowing."""
+    valid = still_valid_public(public, {}, lead=SEPARATION_LEAD)
+    clocks = {str(value)[:5] for row in valid for value in (row.get("cluster_times") or [row["time"]])}
+    return [candidate for candidate in candidates if candidate.strftime("%H:%M") in clocks]
+
+
+def reveal_metrics(
+    rows: list[dict[str, Any]], candidates: list[datetime], delivery: list[datetime],
+    truth: datetime, benchmark_id: str, case_id: str,
+) -> dict[str, Any]:
+    ordered = opaque_order(benchmark_id, case_id, rows)
+    predicted_clock = _opaque_winner(benchmark_id, case_id, rows)
+    by_clock = {candidate.strftime("%H:%M"): candidate for candidate in candidates}
+    if len(by_clock) != len(candidates):
+        raise ValueError("ambiguous_candidate_clock")
+    predicted = by_clock[predicted_clock]
+    within = min(candidates) <= truth <= max(candidates)
+    in_candidates = truth in candidates
+    rank = next((i for i, row in enumerate(ordered, 1) if by_clock[row["time"]] == truth), None)
+    covered = bool(delivery) and min(delivery) <= truth <= max(delivery)
+    return {
+        "truth_in_window": within,
+        "truth_in_candidates": in_candidates,
+        "true_rank": rank,
+        "top_1_hit": predicted == truth,
+        "top_1_minute_error": abs((predicted - truth).total_seconds()) / 60,
+        "delivery_covers_truth": covered,
+        "candidate_count": len(candidates),
+        "delivery_width_minutes": int((max(delivery) - min(delivery)).total_seconds() / 60) + 1 if delivery else 0,
+    }
+
+
+def summarize(trials: list[dict[str, Any]], radii: tuple[int, ...], offsets: tuple[int, ...]) -> list[dict[str, Any]]:
+    result = []
+    for radius in radii:
+        for offset in offsets:
+            group = [row for row in trials if row["radius_minutes"] == radius and row["offset_minutes"] == offset]
+            count = len(group)
+            result.append({
+                "radius_minutes": radius, "offset_minutes": offset, "trial_count": count,
+                **{key: round(sum(row[source] for row in group) / count, 4) if count else None for key, source in (
+                    ("truth_in_window_rate", "truth_in_window"),
+                    ("top_1_rate", "top_1_hit"),
+                    ("delivery_coverage_rate", "delivery_covers_truth"),
+                    ("mean_absolute_minute_error", "top_1_minute_error"),
+                )},
+            })
+    return result
+
+
+def run(dataset: Path = DATASET, radii: tuple[int, ...] = RADII, offsets: tuple[int, ...] = OFFSETS) -> dict[str, Any]:
+    manifest = json.loads(dataset.read_text(encoding="utf-8"))
+    frozen_files = manifest["frozen_scoring"]["files"]
+    scoring_files = sorted(set(frozen_files + PRODUCTION_FILES))
+    starting_hash = implementation_sha256(scoring_files)
+    validation = validate(dataset)
+    invalid = validation["invalid_cases"]
+    trials = []
+    for index, case in enumerate(manifest["cases"], 1):
+        if case["case_id"] in invalid:
+            continue
+        # Reuse static chart calculations, not window-dependent scores or ranks.
+        contexts_by_moment: dict[datetime, dict[str, Any]] = {}
+        for radius in radii:
+            for offset in offsets:
+                request, candidates = shifted_window(case, offset, radius)
+                missing = [moment for moment in candidates if moment not in contexts_by_moment]
+                if missing:
+                    contexts_by_moment.update(zip(missing, compute_candidate_static_contexts(request, candidates=missing), strict=True))
+                contexts = [contexts_by_moment[moment] for moment in candidates]
+                rows = score_window(request, contexts)
+                public = select_signature_representatives(rows, contexts)
+                delivery = delivery_moments(public, candidates)
+                # All scores and the delivery envelope are locked before reveal.
+                truth = datetime.fromisoformat(f"{case['birth']['date']}T{case['birth']['time']}:00")
+                trials.append({
+                    "case_ordinal": index, "offset_minutes": offset, "radius_minutes": radius,
+                    **reveal_metrics(rows, candidates, delivery, truth, manifest["benchmark_id"], case["case_id"]),
+                })
+    if implementation_sha256(scoring_files) != starting_hash:
+        raise ValueError("scorer_changed_during_sweep")
+    return {
+        "scope": "reported_offset_sensitivity_not_product_accuracy",
+        "specification": {
+            "ayanamsa": AYANAMSA, "node_mode": NODE_MODE,
+            "radii_minutes": list(radii), "offsets_minutes": list(offsets), "minute_step": MINUTE_STEP,
+            "dataset": dataset.relative_to(ROOT).as_posix(), "dataset_sha256": file_sha256(dataset),
+            "dataset_benchmark_id": manifest["benchmark_id"],
+            "implementation_sha256": implementation_sha256(frozen_files),
+            "implementation_sha256_prefix": implementation_sha256(frozen_files)[:16],
+            "production_scoring_files": scoring_files, "production_scoring_sha256": starting_hash,
+            "research_files": RESEARCH_FILES,
+            "research_implementation_sha256": implementation_sha256(RESEARCH_FILES),
+            "hash_scope": "explicit_identity_file_sets_not_a_transitive_dependency_lock",
+            "scorer": "native_event_contribution_matrix_not_shadow_fact_ranker",
+            "delivery": "initial_signature_clusters_peak_gap_lt_8_envelope_no_answers_no_elimination",
+            "rank": "score_desc_then_sha256(benchmark_id:case_id:candidate_time)",
+            "grid": "every_minute_inclusive_not_production_two_minute_sampling",
+            "cross_midnight": "date_aware_candidates_distance_and_matrix_grouped_by_candidate_date",
+            "evaluator_sha256": file_sha256(Path(__file__)),
+            "replay_revision": "candidate_date_grouped_v2",
+            "supersedes": "initial_sweep_invalidated_cross_midnight_transition_anchor",
+            "is_blind_evaluation": False, "truth_hidden_from_ranker": True,
+            "results_previously_seen": True, "must_not_use_for_tuning": True,
+        },
+        "excluded_cases": invalid, "case_count": validation["valid_public_aa_cases"],
+        "trial_count": len(trials), "trials": trials,
+        "summary": summarize(trials, radii, offsets),
+        "widening_geometry": {
+            "scored_radii": list(radii),
+            "radius_120_scored": 120 in radii,
+            "truth_in_window_condition": "abs(reported_offset_minutes) <= radius_minutes",
+            "radius_15_first_integer_minute_outside": 16,
+            "all_tested_offsets_within_radius_30_60_120": max(map(abs, offsets)) <= 30,
+            "boundary": "Geometry rescues candidate inclusion only, not ranking or delivery coverage; no population frequency without a reported-offset distribution.",
+        },
+    }
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--json", action="store_true")
+    args = parser.parse_args()
+    print(json.dumps(run(), ensure_ascii=False, indent=2))
@@ -0,0 +1,153 @@
+#!/usr/bin/env python3
+"""Freeze and replay the exposed v3 corpus; never claim a fresh blind holdout."""
+from __future__ import annotations
+
+import argparse
+import hashlib
+import json
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+ROOT = Path(__file__).resolve().parents[2]
+if str(ROOT) not in sys.path:
+    sys.path.insert(0, str(ROOT))
+
+from scripts.active_rectification_event_engine import AYANAMSA, NODE_MODE
+from scripts.minute_rectification_blind_eval import (
+    _candidate_moments, _clock_distance, _opaque_winner, _request,
+    implementation_sha256, summarize_trials,
+)
+from scripts.minute_rectification_fact_blind_eval_v4 import _would_confirm
+from scripts.minute_rectification_fact_ranker_v4 import (
+    ALGORITHM_VERSION, rank_fact_rows, score_fact_ranker_v4,
+)
+from scripts.minute_rectification_feature_facts_v4 import build_feature_fact_rows
+from scripts.minute_rectification_holdout_validator import validate
+
+DATASET = ROOT / "references/real_case_calibration/minute_rectification_holdout_v3.json"
+FREEZE = ROOT / "docs/research/sealed_holdout_rerun_2026_09_20.freeze.json"
+REPORT = ROOT / "docs/research/sealed_holdout_rerun_2026_09_20.json"
+
+
+def file_sha256(path: Path) -> str:
+    return hashlib.sha256(path.read_bytes()).hexdigest()
+
+
+def opaque_order(benchmark_id: str, case_id: str, rows: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    """Extend the existing opaque winner to a total, truth-independent ranking."""
+    return sorted(rows, key=lambda row: (
+        -row["score"],
+        hashlib.sha256(f"{benchmark_id}:{case_id}:{row['time']}".encode()).hexdigest(),
+    ))
+
+
+def freeze_record(dataset: Path = DATASET) -> dict[str, Any]:
+    manifest = json.loads(dataset.read_text(encoding="utf-8"))
+    files = manifest["frozen_scoring"]["files"]
+    return {
+        "record_version": "exposed-v3-fixed-protocol-rerun-v1",
+        "frozen_at_utc": datetime.now(timezone.utc).isoformat(),
+        "dataset_path": dataset.relative_to(ROOT).as_posix(),
+        "dataset_sha256": file_sha256(dataset),
+        "algorithm_version": ALGORITHM_VERSION,
+        "implementation_sha256": implementation_sha256(files),
+        "files": files,
+        "historical_frozen_sha256": manifest["frozen_scoring"]["implementation_sha256"],
+        "evaluator_sha256": file_sha256(Path(__file__)),
+        "ayanamsa": AYANAMSA,
+        "node_mode": NODE_MODE,
+        "candidate_radius_minutes": sorted({case["candidate_radius_minutes"] for case in manifest["cases"]}),
+        "minute_step": 1,
+        "release_metrics": manifest["release_metrics"],
+        "results_previously_seen": True,
+        "official_valid_independent_blind": False,
+        "must_not_use_for_tuning": True,
+        "tie_breaker": manifest["frozen_scoring"]["tie_breaker"],
+        "metric_rank_definition": "competition_rank_1_plus_strictly_higher_scores_legacy_protocol",
+        "extra_metric_rank_definition": "score_desc_then_existing_opaque_sha256_total_order",
+    }
+
+
+def run(freeze_path: Path = FREEZE, dataset: Path = DATASET) -> dict[str, Any]:
+    frozen = json.loads(freeze_path.read_text(encoding="utf-8"))
+    actual = freeze_record(dataset)
+    for key in actual:
+        if key != "frozen_at_utc" and actual[key] != frozen.get(key):
+            raise ValueError(f"frozen_record_mismatch:{key}")
+    manifest = json.loads(dataset.read_text(encoding="utf-8"))
+    validation = validate(dataset)
+    invalid = validation["invalid_cases"]
+    trials = []
+    for index, case in enumerate(manifest["cases"], 1):
+        if case["case_id"] in invalid:
+            continue
+        request = _request(case, case["events"])
+        candidates = _candidate_moments(case)
+        facts = build_feature_fact_rows(request, candidates=candidates)
+        rows, _ = rank_fact_rows(facts, request["events"])
+        result = score_fact_ranker_v4(facts, request["events"])
+        predicted = _opaque_winner(manifest["benchmark_id"], case["case_id"], rows)
+        ordered = opaque_order(manifest["benchmark_id"], case["case_id"], rows)
+        sparse_request = _request(case, case["events"][:1])
+        sparse_facts = build_feature_fact_rows(sparse_request, candidates=candidates)
+        sparse_result = score_fact_ranker_v4(sparse_facts, sparse_request["events"])
+        # Truth is revealed only after both full and sparse ranking/decisions.
+        truth = case["birth"]["time"]
+        truth_score = next(row["score"] for row in rows if row["time"] == truth)
+        would_confirm = _would_confirm(result)
+        trials.append({
+            "case_ordinal": index,
+            "candidate_count": len(rows),
+            "event_count": len(request["events"]),
+            "true_rank": 1 + sum(row["score"] > truth_score for row in rows),
+            "opaque_true_rank": next(i for i, row in enumerate(ordered, 1) if row["time"] == truth),
+            "minute_error": _clock_distance(predicted, truth),
+            "would_confirm": would_confirm,
+            "false_confirmation": would_confirm and predicted != truth,
+            "insufficient_evidence_rejected": not _would_confirm(sparse_result),
+            "full_trial_reasons": result["reasons"],
+            "sparse_trial_reasons": sparse_result["reasons"],
+        })
+    aggregate = summarize_trials(trials, manifest["release_metrics"])
+    count = len(trials)
+    return {
+        "scope": "fixed_protocol_previously_exposed_v3_rerun",
+        "evaluated_on": datetime.now(timezone.utc).date().isoformat(),
+        "frozen_record": frozen,
+        "implementation_hash_matches_at_replay": True,
+        "dataset_hash_matches_at_replay": True,
+        "source_audit_status": manifest["source_audit_status"],
+        "validation_status": validation["status"],
+        "valid_public_aa_cases": validation["valid_public_aa_cases"],
+        "excluded_cases": invalid,
+        "trial_count": count,
+        "trials": trials,
+        **aggregate,
+        "opaque_exact_top_1_rate": sum(row["opaque_true_rank"] == 1 for row in trials) / count if count else None,
+        "opaque_exact_top_3_rate": sum(row["opaque_true_rank"] <= 3 for row in trials) / count if count else None,
+        "official_valid_independent_blind": False,
+        "official_blind_trial_count": 0,
+        "is_blind_evaluation": False,
+        "truth_hidden_from_ranker": True,
+        "results_previously_seen": True,
+        "verified_minute_claim_allowed": False,
+        "status": "blocked_independent_blind_evidence",
+        "boundary": "Three-event low-information protocol, not a mathematical accuracy lower bound and not representative of real sessions. Historical v3/v4 exposure cannot be undone by freezing today's scorer. No tuning or release claims.",
+    }
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--freeze", action="store_true", help="Create a new record before replay; never overwrite an existing record")
+    parser.add_argument("--freeze-path", type=Path, default=FREEZE)
+    parser.add_argument("--json", action="store_true")
+    args = parser.parse_args()
+    if args.freeze:
+        with args.freeze_path.open("x", encoding="utf-8", newline="\n") as handle:
+            json.dump(freeze_record(), handle, ensure_ascii=False, indent=2)
+            handle.write("\n")
+        print(args.freeze_path.relative_to(ROOT).as_posix())
+    else:
+        print(json.dumps(run(args.freeze_path), ensure_ascii=False, indent=2))
@@ -2,6 +2,12 @@

 Purpose: read this file before substantial project work. It exists to stop repeat mistakes caused by multiple Codex windows, WorkBuddy mirrors, local drafts, backup folders, and partial cloud-git visibility.

+## 2026-09-20 · 跨午夜生产修复的身份与本机预检
+
+- 任务书的“V9 `engine_version` 全仓只写不比”不完整：成功回执来自实际 `algorithmVersion`，已有 score cache 通过版本接口/环境覆盖比较算法身份。只 bump 前端默认串不能保证成功结果与旧缓存可区分；本单同步 bump 既有后端算法身份，不新增历史会话打开门，不重标旧缓存。
+- 新工作树预检仍为远端 verified、适配器/碎片检查成功、focused 镜像路径断言失败；快速门基线与本轮均在 source inventory 因缺 `mcp` 失败。BUG-089 的 `mcp>=1.0,<2` 约束两处仍在，本机问题是未安装，不是再次升到 MCP 2。
+- 广域校正 Python 基线已有 4 个失败，必须逐条对照，不借机修改历史冻结值或削弱旧断言。验收记录见 `docs/tasks/PROGRESS-rectification-cross-midnight-20260920.md`。
+
 ## 2026-09-20 · 校正验证补缺的本机验收复现

 - 基线与实现工作树均使用 Python 3.11.7（无项目 `.venv`；`python3` launcher 退出 49）。开工预检远端 verified，但同一碎片镜像路径断言失败；不得把同步成功写成预检通过。
@@ -1,4 +1,6 @@
-# 申报偏差敏感性评测（2026-09-20）
+# 申报偏差敏感性评测（2026-09-20，BUG-981 后原生单矩阵重跑）
+
+> 当前正式结果：`reported_offset_cross_midnight_2026_09_20.json`，冻结：`reported_offset_cross_midnight_2026_09_20.final.freeze.json`。旧 `reported_offset_2026_09_20.json` 原字节保留，不再充当当前身份报告。修订前本 Markdown、旧 JSON/冻结/契约/评测器逐字节保留在 `history/rectification_pre_cross_midnight_2026_09_20/`，由 manifest 固定哈希。

 ## 结论与边界

@@ -6,7 +8,7 @@

 本轮完整评测预先固定 15 个偏移 × 3 档半径 × 20 例，共 900 组合。修正版全部运行完成，validator 排除 0 例；实测结果以下列最终 JSON 为准，不能用中途旧版本初扫代替。

-**区间覆盖不等于分钟命中，模拟偏差敏感性不等于真实用户准确率。** 不运行六题真值方向回放，不以用户认可作真值，不调参、不改生产打分、不打开确认门。
+**区间覆盖不等于分钟命中，模拟偏差敏感性不等于真实用户准确率。** 不运行六题真值方向回放，不以用户认可作真值，不调参、不换案例、不打开确认门。本轮生产代理单独修正候选日期与版本标记；本研究使用修复后的原生路径，不增加评分规则。

 ## 口径

@@ -22,12 +24,12 @@
 | 额外身份 | JSON 的 `production_scoring_sha256` 与 `research_implementation_sha256` 分别绑定扩展评分文件和研究适配文件；显式文件清单不是完整传递依赖锁 |
 | 头名与排名 | 分数降序，再用既有 `_opaque_winner` 同源 SHA-256 规则排同分；不选择最接近真值的候选 |
 | 交付区间 | 初始未答题：signature clusters 中落后头名不足 8 分的仍有效簇外包范围；无新回答、无淘汰，不是完整问答链交付效果 |
-| 跨午夜 | 候选、误差及区间使用完整日期；离线矩阵按候选日期分组，合并后全窗排序/聚类 |
+| 跨午夜 | 候选、误差及区间使用完整日期；修复后的原生单矩阵内按候选日期计分，全窗排序/聚类；不再运行日期分组适配 |
 | 独立性 | `is_blind_evaluation=false`，已曝光集；评分/区间定稿后才揭示标签，仅证明程序标签隔离 |

 ## 实测表

-每格依次为 **真值在窗比例 / 哈希头名命中 / 初始交付区间覆盖**，分母均为 20 例。列为申报偏移分钟，行为搜索半径。统一口径：v3、raman/mean、1 分钟步长；历史打分身份 `b15d9ea15227cd58`，扩展原生评分身份 `115c3fbcffdaff49`，研究适配身份 `8d0c613dc0900adb`。
+每格依次为 **真值在窗比例 / 哈希头名命中 / 初始交付区间覆盖**，分母均为 20 例。列为申报偏移分钟，行为搜索半径。统一口径：v3、raman/mean、1 分钟步长；历史 12 文件打分身份 `b15d9ea15227cd58`，扩展原生评分身份 `7fffd1db612af1f2`，研究评测身份 `2931d4661bf53dde`。

 | 半径 | -30 | -20 | -15 | -10 | -8 | -5 | -3 | 0 | +3 | +5 | +8 | +10 | +15 | +20 | +30 |
 | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
@@ -46,11 +48,25 @@
 - 半径 ±30 自第 31 个整数分钟偏移出窗，±60 自第 61 个、±120 自第 121 个。放宽不是无条件保证，更不代表排名命中或交付覆盖同样恢复。
 - ±120 未跑评分，不报告其头名命中率或交付覆盖率。真实用户申报偏差分布未知，不估算现实用户中有多少人能被放宽救回。

-## 初扫作废与独立复核
+## 历史初扫、日期分组版与本轮原生重跑

-首版正确枚举了跨日候选，但矩阵内 transition proximity 仍共用窗口起日，独立审查发现跨日候选得分偏移。因此旧初扫作废，修正版标记 `replay_revision=candidate_date_grouped_v2`，并登记 `supersedes=initial_sweep_invalidated_cross_midnight_transition_anchor`。
+历史初扫正确枚举跨日候选，但矩阵内 transition proximity 共用窗口起日，因此初扫作废。其后已修成 `candidate_date_grouped_v2` 离线日期分组版，旧同名 JSON 是这个修正版，不是错误初扫。修正版旧身份：扩展原生评分 `115c3fbcffdaff49`，研究适配 `8d0c613dc0900adb`，旧数值不能被误说成必然错误。

-修复仅在本次新脚本 `score_window()` 做按日期分组适配，未改冻结打分文件。实引擎回归验证跨日窗口的分组计算与逐候选独立重算相同；聚类和交付范围在合并后全窗运行。**生产日期处理未在本单修复，不能称为生产端到端回放。**
+本轮 `replay_revision=native_candidate_date_v3`：生产 helper 已修，`score_window()` 改为一次原生矩阵调用；回归同时锁定调用次数为 1、与逐候选独立真实引擎重算相等。不再保留运行中的日期分组替代路径。旧数字因实现/路径身份更新不再代表当前报告，是否变化只由完整比对决定，不能预设数字必变。
+
+**这里只保证 matrix scoring path 是修复后的原生单次调用，不是完整生产端到端回放。** `shifted_window()` 使用完整日期，`delivery_moments()` 按完整日期定义研究交付外包区间；它们仍是离线协议。协调独立审查另报 BUG-982/983：生产跨午夜簇/span 的钟点排序与前端早凌晨申报窗日期锚点仍有独立缺陷。本研究不覆盖这些路径，不能凭区间覆盖数字声称线上所有跨日问题已修。
+
+## 本轮预冻结身份
+
+| 项目 | 值 |
+| --- | --- |
+| 冻结时间（先于评分） | `2026-09-20T04:56:55.260470+00:00` |
+| 原 12 文件 | `b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18` |
+| 扩展生产 18 文件 | `7fffd1db612af1f244a8b5d5eac2f1a3f5a8e9e138c09ac6b3743b98f9a47a51` |
+| 研究/校验 7 文件 | `2931d4661bf53dde5c57f6492de65be4c00877293b7277353279a05eda5af215` |
+| 冻结文件 SHA-256 | `1f281b6c3e9b5a7dc2de91fb2ae1fd94da38f9b8a7bb1f930f670e9a39865074` |
+
+原 12 文件未变，但实际变化的 `scoring_service.py` 与 `dasha_transition_proximity.py` 都进入扩展身份。冻结亦绑定数据集、所有组合/参数、每文件字节哈希、归档 manifest 和评测器；评分前后核对，漂移即拒绝，不能事后刷新冻结。准备阶段曾因旧 JSON 覆盖被安全工具拒绝而迁移新报告路径，准备产物及弃用原因由历史目录 `preparation_manifest.json` 保留；最终路径评测器另行预冻结，没有追认旧冻结。

 ## 可复算产物与验收

@@ -59,6 +75,19 @@ python scripts/research/reported_offset_sweep.py --json
 python -m pytest tests/test_reported_offset_research.py tests/test_rectification_validation_integrity_gate.py -q
 ```

-`reported_offset_2026_09_20.json` 保留逐例四项必要输出：真值是否在窗、真值排名、头名分钟误差、交付区间覆盖，以及每个格子的分母、全部参数和实现哈希。仅保存公开集序号，不保存出生分钟、坐标或事件正文。
+`reported_offset_cross_midnight_2026_09_20.json` 保留逐例四项必要输出：真值是否在窗、真值排名、头名分钟误差、交付区间覆盖，以及每个格子的分母、全部参数和实现哈希。仅保存公开集序号，不保存出生分钟、坐标或事件正文。
+
+**最终全量 900/900 组合完成，用时 399.902 秒；排除 0 例。与旧日期分组版逐例全部字段相同，改变 0/900，45/45 汇总格一致。** JSON 的 `historical_comparison.trials` 保留所有 before/after 与 changed_fields（含未变化行）。这符合旧最终版已经做日期分组修正的事实，不表示生产旧 helper 没有 Bug。本轮没有调参、删失败样本、换半径或把已曝光集包装成新独立盲测。
+
+### 既有测试断言变更
+
+| 原值 / 原断言 | 新值 / 新断言 | 原因 |
+| --- | --- | --- |
+| 跨日实引擎 `grouped == expected` 且生产 `old_rows != expected` | 原生单矩阵 `native == expected` 且 `native_rows == expected`，并锁一次矩阵调用 | 原测试第二条锁住的是已知生产缺陷；修复后必须相等。独立逐候选真实引擎 oracle 原样保留并强化，不是删掉差异验证 |
+| `replay_revision=candidate_date_grouped_v2` | `native_candidate_date_v3` | 原生路径已修，不再离线分组 |
+| 记录测试读取旧 JSON | 读取脚本新 `REPORT` 路径，全部数据集/评测器/生产/研究哈希断言保留 | 旧产物原字节留档，新报告独立创建 |
+| 仅运行后记录参数身份 | 预冻结完整身份，评分前后核对；新增漂移提前拒绝、时间先后、完整旧新对比断言 | 不能事后伪造冻结或只比较有利样本 |
+
+本轮定向验证：`test_reported_offset_research.py`、`test_sealed_holdout_contract_freshness.py`、`test_rectification_validation_integrity_gate.py`、`test_rectification_confirmation_and.py` 合计 **52/52 通过**（含桥接重复收集）。独立逐行比较新旧 900 trials、45 summary 全相同；当前报告 SHA-256 为 `26c3d4a6e5e96ffaacdd00c64b5f3ed08fb9eacb18ef6fffc4b01c15159207b4`。旧产物与准备产物 manifest 的全部字节校验通过。预检因既有 Windows 镜像路径断言失败（23 通过 / 1 失败），不是预检通过。完整环境/广域验收由 `../tasks/PROGRESS-rectification-cross-midnight-20260920.md` 记录。

 排名不能与 T2 的竞争排名 top-1/top-3 混比：T2 按旧协议允许并列多个分钟同时算第一；本表把同分完全按哈希打破。最终测试数字与已知基线失败见 [进度](../tasks/PROGRESS-rectification-validation-20260920.md)。真实分布采集与新未曝光样本见 [v5 协议](sealed_holdout_v5_protocol_2026_09_20.md)。
@@ -0,0 +1,108 @@
+{
+  "record_version": "reported-offset-native-candidate-date-v3",
+  "frozen_at_utc": "2026-09-20T04:56:55.260470+00:00",
+  "ayanamsa": "raman",
+  "node_mode": "mean",
+  "radii_minutes": [
+    15,
+    30,
+    60
+  ],
+  "offsets_minutes": [
+    -30,
+    -20,
+    -15,
+    -10,
+    -8,
+    -5,
+    -3,
+    0,
+    3,
+    5,
+    8,
+    10,
+    15,
+    20,
+    30
+  ],
+  "minute_step": 1,
+  "dataset": "references/real_case_calibration/minute_rectification_holdout_v3.json",
+  "dataset_sha256": "45aa76ab79dfc6c6404b6252083e1aa5c7e2e2b7de811fc11042a8ad2da3f0ea",
+  "dataset_benchmark_id": "minute_rectification_fact_ranker_v4_holdout_v3",
+  "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+  "implementation_sha256_prefix": "b15d9ea15227cd58",
+  "historical_artifacts_manifest_sha256": "6102a26a840be207b5858b3a4c0509274468ae9071e86d308cbbac7371d6864e",
+  "production_scoring_files": [
+    "scripts/active_rectification_event_engine.py",
+    "scripts/active_rectification_events.py",
+    "scripts/ashtakavarga.py",
+    "scripts/dasha_analyzer.py",
+    "scripts/divisional_charts_extended.py",
+    "scripts/domain_calculation_service.py",
+    "scripts/jaimini.py",
+    "scripts/minute_rectification_fact_ranker_v4.py",
+    "scripts/minute_rectification_feature_facts_v4.py",
+    "scripts/narayana_dasha.py",
+    "scripts/rectification/candidate_contrast.py",
+    "scripts/rectification/case_holdout.py",
+    "scripts/rectification/contracts.py",
+    "scripts/rectification/dasha_transition_proximity.py",
+    "scripts/rectification/event_probes.py",
+    "scripts/rectification/scoring_service.py",
+    "scripts/shadbala.py",
+    "scripts/varga.py"
+  ],
+  "production_scoring_sha256": "7fffd1db612af1f244a8b5d5eac2f1a3f5a8e9e138c09ac6b3743b98f9a47a51",
+  "research_files": [
+    "scripts/research/reported_offset_sweep.py",
+    "scripts/research/cluster_width_lib.py",
+    "scripts/research/probe_supply_after_six.py",
+    "scripts/research/sealed_holdout_rerun.py",
+    "scripts/minute_rectification_blind_eval.py",
+    "scripts/minute_rectification_fact_blind_eval_v4.py",
+    "scripts/minute_rectification_holdout_validator.py"
+  ],
+  "research_implementation_sha256": "2931d4661bf53dde5c57f6492de65be4c00877293b7277353279a05eda5af215",
+  "file_sha256": {
+    "scripts/active_rectification_event_engine.py": "edbe93ef3c8bd65f986139dba6b95b96d9b0fa9920341fbe677cd6f53e4db84f",
+    "scripts/active_rectification_events.py": "03e333ccf1103caf685dfe50fe890ab1793e945037085f50ce834a4a3b232222",
+    "scripts/ashtakavarga.py": "cc32776674cf60713634a360a6f09774b3af0ec5d0f650687f52b79597b0dfbe",
+    "scripts/dasha_analyzer.py": "5e53b1b4d4c7414c08dd0e7716f120a63c70ec3428ec93547f64d55c580da0e1",
+    "scripts/divisional_charts_extended.py": "73bd50a83ac444830940184d0988cea621855f91db76a3f1a232014b992eacea",
+    "scripts/domain_calculation_service.py": "9ee955874eea806c7b56f4246d90009c02ee8318b8f3dfe1d7132b1c0b4262e1",
+    "scripts/jaimini.py": "b98bf965603dd2e3cc871fe40f3c8184731a701875ca50fef3ca68e9682eda81",
+    "scripts/minute_rectification_blind_eval.py": "07aa30d8bad3278669495d5fea8be937aaf6367c8af980ff4f2fe9aa84ccc6ff",
+    "scripts/minute_rectification_fact_blind_eval_v4.py": "bdf460e3467208e8b87345c5beb2128f3e22de7eeffd68f72b281b3ad9affc92",
+    "scripts/minute_rectification_fact_ranker_v4.py": "f1a0d8d2482cd754d2e7f253a9606b0373519435a2bf58cfc3753e2b9ae0fb76",
+    "scripts/minute_rectification_feature_facts_v4.py": "3d899a2cdd4825b8e32062443d487edcd2173a4ea55a36eb9f3abe587c0faf0f",
+    "scripts/minute_rectification_holdout_validator.py": "b60b1c752e58672beaee8b36eb3b2348169726cda070ec969db0f13e7ed715af",
+    "scripts/narayana_dasha.py": "7ff2c3238cd113b11b815e14b41967bd8e486ff62e36650e55613b9c79d1bfe3",
+    "scripts/rectification/candidate_contrast.py": "10a360586359e72846593ef6d7c15a2cea79358245a34c288851406fe9e47ca8",
+    "scripts/rectification/case_holdout.py": "fe0e699b617d9e216acdfeff87af4221c1d6f4c101bf33e1dbf14e2bcd75a8b5",
+    "scripts/rectification/contracts.py": "fdd1a47b2e3dac8579e870a282ed74c68518b39f6c0576b27df766a4b8021143",
+    "scripts/rectification/dasha_transition_proximity.py": "0afba494aa12899621173c3cc941c6c1d0be9cec0bdf4c4a1c9bcd004ab22de8",
+    "scripts/rectification/event_probes.py": "45478ee3fdc36dc36e1a602bee5dd0627a5464275e370e64adb5826fe2aa58f5",
+    "scripts/rectification/scoring_service.py": "0adf5700bb22dd8c4c3147027251210e21ec66a2c9f9a91217edba3ad9e5d1f5",
+    "scripts/research/cluster_width_lib.py": "1d8f2ffa7795d62069deebb2de8e16749317037d4ed3f24a2c0092676e5cfcc6",
+    "scripts/research/probe_supply_after_six.py": "098a5b4398d6f8d996917d809c2fc922ad64031d3b70660f8ef7c876aba92438",
+    "scripts/research/reported_offset_sweep.py": "f81c681fd0fe257e90a761dfab370f623e5f8a881b62cc6b1c596cbd58800f8f",
+    "scripts/research/sealed_holdout_rerun.py": "34a6724530fbecdecff9e5e07fa64980b8e8aff1fac2e000f4268debe1ee96e1",
+    "scripts/shadbala.py": "912e0e6d169c2172aab85f71e4347e3e39193020777b43825b775d906400956b",
+    "scripts/varga.py": "4331de5a25ea08729af91aae863943c4223183937f419d52dc63fd6fe7f27be6"
+  },
+  "hash_scope": "explicit_identity_file_sets_not_a_transitive_dependency_lock",
+  "scorer": "native_event_contribution_matrix_not_shadow_fact_ranker",
+  "delivery": "initial_signature_clusters_peak_gap_lt_8_envelope_no_answers_no_elimination",
+  "rank": "score_desc_then_sha256(benchmark_id:case_id:candidate_time)",
+  "grid": "every_minute_inclusive_not_production_two_minute_sampling",
+  "cross_midnight": "date_aware_candidates_distance_and_native_single_matrix",
+  "evaluator_sha256": "f81c681fd0fe257e90a761dfab370f623e5f8a881b62cc6b1c596cbd58800f8f",
+  "replay_revision": "native_candidate_date_v3",
+  "supersedes": "candidate_date_grouped_v2_identity_and_path_not_assumed_numerically_wrong",
+  "is_blind_evaluation": false,
+  "truth_hidden_from_ranker": true,
+  "official_valid_independent_blind": false,
+  "official_blind_trial_count": 0,
+  "results_previously_seen": true,
+  "must_not_use_for_tuning": true
+}
@@ -0,0 +1,108 @@
+{
+  "record_version": "reported-offset-native-candidate-date-v3",
+  "frozen_at_utc": "2026-09-20T04:50:20.160896+00:00",
+  "ayanamsa": "raman",
+  "node_mode": "mean",
+  "radii_minutes": [
+    15,
+    30,
+    60
+  ],
+  "offsets_minutes": [
+    -30,
+    -20,
+    -15,
+    -10,
+    -8,
+    -5,
+    -3,
+    0,
+    3,
+    5,
+    8,
+    10,
+    15,
+    20,
+    30
+  ],
+  "minute_step": 1,
+  "dataset": "references/real_case_calibration/minute_rectification_holdout_v3.json",
+  "dataset_sha256": "45aa76ab79dfc6c6404b6252083e1aa5c7e2e2b7de811fc11042a8ad2da3f0ea",
+  "dataset_benchmark_id": "minute_rectification_fact_ranker_v4_holdout_v3",
+  "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+  "implementation_sha256_prefix": "b15d9ea15227cd58",
+  "historical_artifacts_manifest_sha256": "6102a26a840be207b5858b3a4c0509274468ae9071e86d308cbbac7371d6864e",
+  "production_scoring_files": [
+    "scripts/active_rectification_event_engine.py",
+    "scripts/active_rectification_events.py",
+    "scripts/ashtakavarga.py",
+    "scripts/dasha_analyzer.py",
+    "scripts/divisional_charts_extended.py",
+    "scripts/domain_calculation_service.py",
+    "scripts/jaimini.py",
+    "scripts/minute_rectification_fact_ranker_v4.py",
+    "scripts/minute_rectification_feature_facts_v4.py",
+    "scripts/narayana_dasha.py",
+    "scripts/rectification/candidate_contrast.py",
+    "scripts/rectification/case_holdout.py",
+    "scripts/rectification/contracts.py",
+    "scripts/rectification/dasha_transition_proximity.py",
+    "scripts/rectification/event_probes.py",
+    "scripts/rectification/scoring_service.py",
+    "scripts/shadbala.py",
+    "scripts/varga.py"
+  ],
+  "production_scoring_sha256": "7fffd1db612af1f244a8b5d5eac2f1a3f5a8e9e138c09ac6b3743b98f9a47a51",
+  "research_files": [
+    "scripts/research/reported_offset_sweep.py",
+    "scripts/research/cluster_width_lib.py",
+    "scripts/research/probe_supply_after_six.py",
+    "scripts/research/sealed_holdout_rerun.py",
+    "scripts/minute_rectification_blind_eval.py",
+    "scripts/minute_rectification_fact_blind_eval_v4.py",
+    "scripts/minute_rectification_holdout_validator.py"
+  ],
+  "research_implementation_sha256": "61dbb47eb46f9a01c78096677ac6238e3a65e9674ce4f2514b18e1ca50c0b954",
+  "file_sha256": {
+    "scripts/active_rectification_event_engine.py": "edbe93ef3c8bd65f986139dba6b95b96d9b0fa9920341fbe677cd6f53e4db84f",
+    "scripts/active_rectification_events.py": "03e333ccf1103caf685dfe50fe890ab1793e945037085f50ce834a4a3b232222",
+    "scripts/ashtakavarga.py": "cc32776674cf60713634a360a6f09774b3af0ec5d0f650687f52b79597b0dfbe",
+    "scripts/dasha_analyzer.py": "5e53b1b4d4c7414c08dd0e7716f120a63c70ec3428ec93547f64d55c580da0e1",
+    "scripts/divisional_charts_extended.py": "73bd50a83ac444830940184d0988cea621855f91db76a3f1a232014b992eacea",
+    "scripts/domain_calculation_service.py": "9ee955874eea806c7b56f4246d90009c02ee8318b8f3dfe1d7132b1c0b4262e1",
+    "scripts/jaimini.py": "b98bf965603dd2e3cc871fe40f3c8184731a701875ca50fef3ca68e9682eda81",
+    "scripts/minute_rectification_blind_eval.py": "07aa30d8bad3278669495d5fea8be937aaf6367c8af980ff4f2fe9aa84ccc6ff",
+    "scripts/minute_rectification_fact_blind_eval_v4.py": "bdf460e3467208e8b87345c5beb2128f3e22de7eeffd68f72b281b3ad9affc92",
+    "scripts/minute_rectification_fact_ranker_v4.py": "f1a0d8d2482cd754d2e7f253a9606b0373519435a2bf58cfc3753e2b9ae0fb76",
+    "scripts/minute_rectification_feature_facts_v4.py": "3d899a2cdd4825b8e32062443d487edcd2173a4ea55a36eb9f3abe587c0faf0f",
+    "scripts/minute_rectification_holdout_validator.py": "b60b1c752e58672beaee8b36eb3b2348169726cda070ec969db0f13e7ed715af",
+    "scripts/narayana_dasha.py": "7ff2c3238cd113b11b815e14b41967bd8e486ff62e36650e55613b9c79d1bfe3",
+    "scripts/rectification/candidate_contrast.py": "10a360586359e72846593ef6d7c15a2cea79358245a34c288851406fe9e47ca8",
+    "scripts/rectification/case_holdout.py": "fe0e699b617d9e216acdfeff87af4221c1d6f4c101bf33e1dbf14e2bcd75a8b5",
+    "scripts/rectification/contracts.py": "fdd1a47b2e3dac8579e870a282ed74c68518b39f6c0576b27df766a4b8021143",
+    "scripts/rectification/dasha_transition_proximity.py": "0afba494aa12899621173c3cc941c6c1d0be9cec0bdf4c4a1c9bcd004ab22de8",
+    "scripts/rectification/event_probes.py": "45478ee3fdc36dc36e1a602bee5dd0627a5464275e370e64adb5826fe2aa58f5",
+    "scripts/rectification/scoring_service.py": "0adf5700bb22dd8c4c3147027251210e21ec66a2c9f9a91217edba3ad9e5d1f5",
+    "scripts/research/cluster_width_lib.py": "1d8f2ffa7795d62069deebb2de8e16749317037d4ed3f24a2c0092676e5cfcc6",
+    "scripts/research/probe_supply_after_six.py": "098a5b4398d6f8d996917d809c2fc922ad64031d3b70660f8ef7c876aba92438",
+    "scripts/research/reported_offset_sweep.py": "cdcd7b583d9ecee8a313c25cac795923da2f8af45875d572f0a3efbc25f5d071",
+    "scripts/research/sealed_holdout_rerun.py": "33c294b21db0631a68068dcf135b70db7aae6a1a3e66414bb72959a448883465",
+    "scripts/shadbala.py": "912e0e6d169c2172aab85f71e4347e3e39193020777b43825b775d906400956b",
+    "scripts/varga.py": "4331de5a25ea08729af91aae863943c4223183937f419d52dc63fd6fe7f27be6"
+  },
+  "hash_scope": "explicit_identity_file_sets_not_a_transitive_dependency_lock",
+  "scorer": "native_event_contribution_matrix_not_shadow_fact_ranker",
+  "delivery": "initial_signature_clusters_peak_gap_lt_8_envelope_no_answers_no_elimination",
+  "rank": "score_desc_then_sha256(benchmark_id:case_id:candidate_time)",
+  "grid": "every_minute_inclusive_not_production_two_minute_sampling",
+  "cross_midnight": "date_aware_candidates_distance_and_native_single_matrix",
+  "evaluator_sha256": "cdcd7b583d9ecee8a313c25cac795923da2f8af45875d572f0a3efbc25f5d071",
+  "replay_revision": "native_candidate_date_v3",
+  "supersedes": "candidate_date_grouped_v2_identity_and_path_not_assumed_numerically_wrong",
+  "is_blind_evaluation": false,
+  "truth_hidden_from_ranker": true,
+  "official_valid_independent_blind": false,
+  "official_blind_trial_count": 0,
+  "results_previously_seen": true,
+  "must_not_use_for_tuning": true
+}
@@ -1,4 +1,6 @@
-# 当前冻结实现的 v3 固定口径重跑（2026-09-20）
+# 当前冻结实现的 v3 固定口径重跑（2026-09-20，BUG-981 后重新冻结）
+
+> 当前正式结果已迁到 `sealed_holdout_rerun_cross_midnight_2026_09_20.json`；旧同名 JSON 与旧 `.freeze.json` 原字节保留，不代表本轮身份。本文修订前的 Markdown、JSON、冻结记录、契约和评测器均逐字节归档在 `history/rectification_pre_cross_midnight_2026_09_20/`，由 `manifest.json` 校验。历史冻结不刷新、不追认。

 ## 结论

@@ -21,9 +23,18 @@
 | 排名标签隔离 | 排序和稀疏证据判定完成后才读取真值标签 |
 | 独立官方盲测 | `official_valid_independent_blind=false`；有效官方盲测试次仍为 0 |

-冻结记录：`sealed_holdout_rerun_2026_09_20.freeze.json`。先以独占创建模式记录 12 个文件、数据集与评测器字节哈希，再执行评分；评分前验证全部身份，任何漂移即拒绝。历史 v3 `frozen_scoring` 原值 `f41c298dd6cdcebe…` 未改。
+本轮冻结记录：`sealed_holdout_rerun_cross_midnight_2026_09_20.final.freeze.json`。在 **2026-09-20T04:56:54.914584+00:00** 以独占创建模式预冻结，然后重跑。12 文件身份继续为 `b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18`；实际修复在这 12 文件之外，故增加以下身份，不能拿旧 12 文件哈希不变冒充生产未变：

-脱敏逐例结果：`sealed_holdout_rerun_2026_09_20.json`，只含公开集序号、排名、误差、门禁原因，不含出生分钟、坐标或事件正文。复算：
+| 身份 | SHA-256 |
+| --- | --- |
+| 扩展生产评分（18 文件，含 `scoring_service.py` 与 `dasha_transition_proximity.py`） | `7fffd1db612af1f244a8b5d5eac2f1a3f5a8e9e138c09ac6b3743b98f9a47a51` |
+| 研究/校验（7 文件） | `2931d4661bf53dde5c57f6492de65be4c00877293b7277353279a05eda5af215` |
+| 本轮冻结文件 | `b74f70d21011f98b78f141535b1bfa49b20701418812a3ad97f9350cc793c67d` |
+| 旧产物归档 manifest | `6102a26a840be207b5858b3a4c0509274468ae9071e86d308cbbac7371d6864e` |
+
+每个文件的字节哈希与有序文件集身份均在冻结记录内；评分前后都核对数据集、评测器与扩展实现。文件集为显式范围，不冒称完整传递依赖或环境锁。历史 v3 `frozen_scoring` 原值 `f41c298dd6cdcebe…` 未改。
+
+脱敏逐例结果：`sealed_holdout_rerun_cross_midnight_2026_09_20.json`，只含公开集序号、排名、误差、门禁原因，不含出生分钟、坐标或事件正文。复算：

 ```bash
 python scripts/research/sealed_holdout_rerun.py --json
@@ -58,7 +69,15 @@ python scripts/research/sealed_holdout_rerun.py --json
 - `source_audit_status` 对齐数据集；`evaluated_on` 表示最新固定口径重跑日期，并新增作用域说明。历史顶层指标仍有 `metrics_produced_by` 的旧日期与来源，未伪装成本次结果。
 - `status`、`valid_public_aa_cases`、`required_cases`、`top_1_rate`、`confirmation_coverage_rate`、`sealed_benchmark_id` 六个 runtime 键的值与类型全不变；尤其 `status=not_ready`、确认覆盖率为零。
 - `official_eval_implementation_hash_matches=false`、`official_eval_trial_count=0` 保留；新增重跑记录不冒充已重新获得独立性。
- 未改任何冻结打分文件、生产判定、现有 `_candidate_moments()` 或 `request_from_case()`。
+- 本轮生产代理修复 transition-proximity 候选日期并将矩阵算法身份升为 `rectification-v5-matrix-scoring-8`，落在历史 12 文件之外。本研究未改权重、候选/事件/半径，未改 `_candidate_moments()` 或 `request_from_case()`。
+
+## BUG-981 前后逐例对比
+
+**20/20 例全部字段相同，改变 0 例；五项指标和附加 opaque 指标均不变。** 当前 JSON 的 `historical_comparison.trials` 保留全部 20 例 before/after 与 changed_fields，包括未变化行，并校验旧报告 SHA-256。不是只保留变化样本。
+
+原因边界：本协议是 shadow fact ranker，`build_feature_fact_rows()` 原先已使用 `candidate_at` 日期，不调用被修复的 transition-proximity helper。扩展生产身份用于标识同一工作树上下文，不得写成该 shadow 路径执行了生产 helper。旧成绩因身份更新不再充当当前报告，但不能把数值本身说成错误；重新冻结也不恢复独立盲测资格。
+
+准备阶段曾生成两份 `*.freeze.json` 和一个 `*.cross_midnight.rerun.json`。因安全工具拒绝覆盖旧 JSON，改用新的正式报告路径，停止未完成的 900 组合准备运行，评测器最终路径修改后重新冻结再完整重跑。所有准备产物保留并由历史目录的 `preparation_manifest.json` 标为 `superseded_preparation_not_final_report`；不得当最终结果。

 ## 防复发与既有断言变更

@@ -69,4 +88,16 @@ python scripts/research/sealed_holdout_rerun.py --json
 | Python / 前端：`current_tree_scorer.implementation_sha256` 等于旧 post-audit sidecar 的哈希 | 等于新固定口径报告 `frozen_record.implementation_sha256`，另验证当前树与新报告 | 契约当前树块的含义就是当前实现；继续绑定旧报告会强制过期。新增身份断言更严格，不弱化门禁 |
 | 官方标志 false、官方试次 0、六键等于历史报告、确认门全套行为断言 | 原值保留，另加固定口径重跑身份/试次和独立盲测 false | 固定口径重跑不是新的独立盲测，不借元数据迁移打开确认门 |

-本机 Python 3.11.7 可跑；`python3` Windows 别名退出 49。该任务定向命令的最终测试清单与基线失败对照由 `PROGRESS-rectification-validation-20260920.md` 统一记录。特别是既有 v2 冻结身份测试不属本次刷新范围，不得为了通过而改其历史封存或断言。
+本轮既有测试调整三栏（此前 BUG-978 的三栏保留在上表）：
+
+| 原值 / 原断言 | 新值 / 新断言 | 原因 |
+| --- | --- | --- |
+| `FREEZE` 指向旧 `sealed_holdout_rerun_2026_09_20.freeze.json`，当前 `REPORT` 指向旧同名 JSON | 指向新 `sealed_holdout_rerun_cross_midnight_2026_09_20.final.freeze.json` 与新 JSON；保留原完整身份相等断言 | 历史冻结不可刷新，新实现/评测器必须重新预冻结；旧文件继续逐字节校验 |
+| 当前树身份只核对 12 文件 | 保留 12 文件相等与数量断言，另核对生产 18 文件、研究 7 文件、每文件哈希、归档 manifest、契约及报告 | 修复位于旧 12 文件之外；新增身份漂移拒绝，不弱化原断言 |
+| 当前报告与冻结身份一致 | 原断言全部保留，另核对冻结时间早于评分开始、评分前后身份一致、全部旧新逐例对比、六键类型与值不变 | 防止事后追认、选样与误开确认门 |
+| Python 确认合同仅比较 legacy 实现身份 | 原断言保留，新增扩展身份相等与 helper 在文件集内 | 原 12 文件不变不足以证明当前修复已被绑定 |
+| 前端确认测试读取旧固定口径 JSON | 由 T4 协调改为新正式 JSON，原门禁与身份断言保留 | 报告地址迁移，不改变确认行为 |
+
+本轮定向四文件合计 **52/52 通过**（含 quick 桥接重复收集），身份/历史字节/全部 before-after 独立复核通过。20 例重跑用时 **3.158 秒**；当前报告 SHA-256：`0122d3cd1545416e1de9029b912d665d130049e9ea768a8972d18bb917331d38`。
+
+本机 Python 3.11.7 可跑；`python3` Windows 别名退出 49。开工预检 remote verified，但 23 通过 / 1 既有 Windows 镜像路径断言失败，不冒写预检通过。本轮测试清单与环境/广域基线对照由 `../tasks/PROGRESS-rectification-cross-midnight-20260920.md` 统一记录；前轮记录仍在 `../tasks/PROGRESS-rectification-validation-20260920.md`。既有 v2 冻结身份测试不属刷新范围，不得为了通过而改历史封存或断言。
@@ -0,0 +1,106 @@
+{
+  "record_version": "exposed-v3-fixed-protocol-cross-midnight-rerun-v2",
+  "extended_identity": {
+    "historical_artifacts_manifest_sha256": "6102a26a840be207b5858b3a4c0509274468ae9071e86d308cbbac7371d6864e",
+    "production_scoring_files": [
+      "scripts/active_rectification_event_engine.py",
+      "scripts/active_rectification_events.py",
+      "scripts/ashtakavarga.py",
+      "scripts/dasha_analyzer.py",
+      "scripts/divisional_charts_extended.py",
+      "scripts/domain_calculation_service.py",
+      "scripts/jaimini.py",
+      "scripts/minute_rectification_fact_ranker_v4.py",
+      "scripts/minute_rectification_feature_facts_v4.py",
+      "scripts/narayana_dasha.py",
+      "scripts/rectification/candidate_contrast.py",
+      "scripts/rectification/case_holdout.py",
+      "scripts/rectification/contracts.py",
+      "scripts/rectification/dasha_transition_proximity.py",
+      "scripts/rectification/event_probes.py",
+      "scripts/rectification/scoring_service.py",
+      "scripts/shadbala.py",
+      "scripts/varga.py"
+    ],
+    "production_scoring_sha256": "7fffd1db612af1f244a8b5d5eac2f1a3f5a8e9e138c09ac6b3743b98f9a47a51",
+    "research_files": [
+      "scripts/research/reported_offset_sweep.py",
+      "scripts/research/cluster_width_lib.py",
+      "scripts/research/probe_supply_after_six.py",
+      "scripts/research/sealed_holdout_rerun.py",
+      "scripts/minute_rectification_blind_eval.py",
+      "scripts/minute_rectification_fact_blind_eval_v4.py",
+      "scripts/minute_rectification_holdout_validator.py"
+    ],
+    "research_implementation_sha256": "2931d4661bf53dde5c57f6492de65be4c00877293b7277353279a05eda5af215",
+    "file_sha256": {
+      "scripts/active_rectification_event_engine.py": "edbe93ef3c8bd65f986139dba6b95b96d9b0fa9920341fbe677cd6f53e4db84f",
+      "scripts/active_rectification_events.py": "03e333ccf1103caf685dfe50fe890ab1793e945037085f50ce834a4a3b232222",
+      "scripts/ashtakavarga.py": "cc32776674cf60713634a360a6f09774b3af0ec5d0f650687f52b79597b0dfbe",
+      "scripts/dasha_analyzer.py": "5e53b1b4d4c7414c08dd0e7716f120a63c70ec3428ec93547f64d55c580da0e1",
+      "scripts/divisional_charts_extended.py": "73bd50a83ac444830940184d0988cea621855f91db76a3f1a232014b992eacea",
+      "scripts/domain_calculation_service.py": "9ee955874eea806c7b56f4246d90009c02ee8318b8f3dfe1d7132b1c0b4262e1",
+      "scripts/jaimini.py": "b98bf965603dd2e3cc871fe40f3c8184731a701875ca50fef3ca68e9682eda81",
+      "scripts/minute_rectification_blind_eval.py": "07aa30d8bad3278669495d5fea8be937aaf6367c8af980ff4f2fe9aa84ccc6ff",
+      "scripts/minute_rectification_fact_blind_eval_v4.py": "bdf460e3467208e8b87345c5beb2128f3e22de7eeffd68f72b281b3ad9affc92",
+      "scripts/minute_rectification_fact_ranker_v4.py": "f1a0d8d2482cd754d2e7f253a9606b0373519435a2bf58cfc3753e2b9ae0fb76",
+      "scripts/minute_rectification_feature_facts_v4.py": "3d899a2cdd4825b8e32062443d487edcd2173a4ea55a36eb9f3abe587c0faf0f",
+      "scripts/minute_rectification_holdout_validator.py": "b60b1c752e58672beaee8b36eb3b2348169726cda070ec969db0f13e7ed715af",
+      "scripts/narayana_dasha.py": "7ff2c3238cd113b11b815e14b41967bd8e486ff62e36650e55613b9c79d1bfe3",
+      "scripts/rectification/candidate_contrast.py": "10a360586359e72846593ef6d7c15a2cea79358245a34c288851406fe9e47ca8",
+      "scripts/rectification/case_holdout.py": "fe0e699b617d9e216acdfeff87af4221c1d6f4c101bf33e1dbf14e2bcd75a8b5",
+      "scripts/rectification/contracts.py": "fdd1a47b2e3dac8579e870a282ed74c68518b39f6c0576b27df766a4b8021143",
+      "scripts/rectification/dasha_transition_proximity.py": "0afba494aa12899621173c3cc941c6c1d0be9cec0bdf4c4a1c9bcd004ab22de8",
+      "scripts/rectification/event_probes.py": "45478ee3fdc36dc36e1a602bee5dd0627a5464275e370e64adb5826fe2aa58f5",
+      "scripts/rectification/scoring_service.py": "0adf5700bb22dd8c4c3147027251210e21ec66a2c9f9a91217edba3ad9e5d1f5",
+      "scripts/research/cluster_width_lib.py": "1d8f2ffa7795d62069deebb2de8e16749317037d4ed3f24a2c0092676e5cfcc6",
+      "scripts/research/probe_supply_after_six.py": "098a5b4398d6f8d996917d809c2fc922ad64031d3b70660f8ef7c876aba92438",
+      "scripts/research/reported_offset_sweep.py": "f81c681fd0fe257e90a761dfab370f623e5f8a881b62cc6b1c596cbd58800f8f",
+      "scripts/research/sealed_holdout_rerun.py": "34a6724530fbecdecff9e5e07fa64980b8e8aff1fac2e000f4268debe1ee96e1",
+      "scripts/shadbala.py": "912e0e6d169c2172aab85f71e4347e3e39193020777b43825b775d906400956b",
+      "scripts/varga.py": "4331de5a25ea08729af91aae863943c4223183937f419d52dc63fd6fe7f27be6"
+    },
+    "hash_scope": "explicit_identity_file_sets_not_a_transitive_dependency_lock"
+  },
+  "production_identity_scope": "context_only_shadow_scorer_does_not_call_transition_proximity",
+  "frozen_at_utc": "2026-09-20T04:56:54.914584+00:00",
+  "dataset_path": "references/real_case_calibration/minute_rectification_holdout_v3.json",
+  "dataset_sha256": "45aa76ab79dfc6c6404b6252083e1aa5c7e2e2b7de811fc11042a8ad2da3f0ea",
+  "algorithm_version": "birth-time-event-fact-ranker-v4-shadow",
+  "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+  "files": [
+    "scripts/active_rectification_event_engine.py",
+    "scripts/active_rectification_events.py",
+    "scripts/ashtakavarga.py",
+    "scripts/dasha_analyzer.py",
+    "scripts/divisional_charts_extended.py",
+    "scripts/domain_calculation_service.py",
+    "scripts/jaimini.py",
+    "scripts/minute_rectification_fact_ranker_v4.py",
+    "scripts/minute_rectification_feature_facts_v4.py",
+    "scripts/narayana_dasha.py",
+    "scripts/shadbala.py",
+    "scripts/varga.py"
+  ],
+  "historical_frozen_sha256": "f41c298dd6cdcebe7a93e632f7191954be7987f6012ca2a34cecb6e447fbf196",
+  "evaluator_sha256": "34a6724530fbecdecff9e5e07fa64980b8e8aff1fac2e000f4268debe1ee96e1",
+  "ayanamsa": "raman",
+  "node_mode": "mean",
+  "candidate_radius_minutes": [
+    10
+  ],
+  "minute_step": 1,
+  "release_metrics": {
+    "top_1_rate_minimum": 0.6,
+    "top_3_rate_minimum": 0.85,
+    "mean_absolute_minute_error_maximum": 2.0,
+    "false_confirmation_rate_maximum": 0.05,
+    "correct_insufficient_evidence_rejection_rate_minimum": 0.9
+  },
+  "results_previously_seen": true,
+  "official_valid_independent_blind": false,
+  "must_not_use_for_tuning": true,
+  "tie_breaker": "sha256(benchmark_id:case_id:candidate_time); published minute is never passed to the ranker",
+  "metric_rank_definition": "competition_rank_1_plus_strictly_higher_scores_legacy_protocol",
+  "extra_metric_rank_definition": "score_desc_then_existing_opaque_sha256_total_order"
+}
@@ -0,0 +1,106 @@
+{
+  "record_version": "exposed-v3-fixed-protocol-cross-midnight-rerun-v2",
+  "extended_identity": {
+    "historical_artifacts_manifest_sha256": "6102a26a840be207b5858b3a4c0509274468ae9071e86d308cbbac7371d6864e",
+    "production_scoring_files": [
+      "scripts/active_rectification_event_engine.py",
+      "scripts/active_rectification_events.py",
+      "scripts/ashtakavarga.py",
+      "scripts/dasha_analyzer.py",
+      "scripts/divisional_charts_extended.py",
+      "scripts/domain_calculation_service.py",
+      "scripts/jaimini.py",
+      "scripts/minute_rectification_fact_ranker_v4.py",
+      "scripts/minute_rectification_feature_facts_v4.py",
+      "scripts/narayana_dasha.py",
+      "scripts/rectification/candidate_contrast.py",
+      "scripts/rectification/case_holdout.py",
+      "scripts/rectification/contracts.py",
+      "scripts/rectification/dasha_transition_proximity.py",
+      "scripts/rectification/event_probes.py",
+      "scripts/rectification/scoring_service.py",
+      "scripts/shadbala.py",
+      "scripts/varga.py"
+    ],
+    "production_scoring_sha256": "7fffd1db612af1f244a8b5d5eac2f1a3f5a8e9e138c09ac6b3743b98f9a47a51",
+    "research_files": [
+      "scripts/research/reported_offset_sweep.py",
+      "scripts/research/cluster_width_lib.py",
+      "scripts/research/probe_supply_after_six.py",
+      "scripts/research/sealed_holdout_rerun.py",
+      "scripts/minute_rectification_blind_eval.py",
+      "scripts/minute_rectification_fact_blind_eval_v4.py",
+      "scripts/minute_rectification_holdout_validator.py"
+    ],
+    "research_implementation_sha256": "61dbb47eb46f9a01c78096677ac6238e3a65e9674ce4f2514b18e1ca50c0b954",
+    "file_sha256": {
+      "scripts/active_rectification_event_engine.py": "edbe93ef3c8bd65f986139dba6b95b96d9b0fa9920341fbe677cd6f53e4db84f",
+      "scripts/active_rectification_events.py": "03e333ccf1103caf685dfe50fe890ab1793e945037085f50ce834a4a3b232222",
+      "scripts/ashtakavarga.py": "cc32776674cf60713634a360a6f09774b3af0ec5d0f650687f52b79597b0dfbe",
+      "scripts/dasha_analyzer.py": "5e53b1b4d4c7414c08dd0e7716f120a63c70ec3428ec93547f64d55c580da0e1",
+      "scripts/divisional_charts_extended.py": "73bd50a83ac444830940184d0988cea621855f91db76a3f1a232014b992eacea",
+      "scripts/domain_calculation_service.py": "9ee955874eea806c7b56f4246d90009c02ee8318b8f3dfe1d7132b1c0b4262e1",
+      "scripts/jaimini.py": "b98bf965603dd2e3cc871fe40f3c8184731a701875ca50fef3ca68e9682eda81",
+      "scripts/minute_rectification_blind_eval.py": "07aa30d8bad3278669495d5fea8be937aaf6367c8af980ff4f2fe9aa84ccc6ff",
+      "scripts/minute_rectification_fact_blind_eval_v4.py": "bdf460e3467208e8b87345c5beb2128f3e22de7eeffd68f72b281b3ad9affc92",
+      "scripts/minute_rectification_fact_ranker_v4.py": "f1a0d8d2482cd754d2e7f253a9606b0373519435a2bf58cfc3753e2b9ae0fb76",
+      "scripts/minute_rectification_feature_facts_v4.py": "3d899a2cdd4825b8e32062443d487edcd2173a4ea55a36eb9f3abe587c0faf0f",
+      "scripts/minute_rectification_holdout_validator.py": "b60b1c752e58672beaee8b36eb3b2348169726cda070ec969db0f13e7ed715af",
+      "scripts/narayana_dasha.py": "7ff2c3238cd113b11b815e14b41967bd8e486ff62e36650e55613b9c79d1bfe3",
+      "scripts/rectification/candidate_contrast.py": "10a360586359e72846593ef6d7c15a2cea79358245a34c288851406fe9e47ca8",
+      "scripts/rectification/case_holdout.py": "fe0e699b617d9e216acdfeff87af4221c1d6f4c101bf33e1dbf14e2bcd75a8b5",
+      "scripts/rectification/contracts.py": "fdd1a47b2e3dac8579e870a282ed74c68518b39f6c0576b27df766a4b8021143",
+      "scripts/rectification/dasha_transition_proximity.py": "0afba494aa12899621173c3cc941c6c1d0be9cec0bdf4c4a1c9bcd004ab22de8",
+      "scripts/rectification/event_probes.py": "45478ee3fdc36dc36e1a602bee5dd0627a5464275e370e64adb5826fe2aa58f5",
+      "scripts/rectification/scoring_service.py": "0adf5700bb22dd8c4c3147027251210e21ec66a2c9f9a91217edba3ad9e5d1f5",
+      "scripts/research/cluster_width_lib.py": "1d8f2ffa7795d62069deebb2de8e16749317037d4ed3f24a2c0092676e5cfcc6",
+      "scripts/research/probe_supply_after_six.py": "098a5b4398d6f8d996917d809c2fc922ad64031d3b70660f8ef7c876aba92438",
+      "scripts/research/reported_offset_sweep.py": "cdcd7b583d9ecee8a313c25cac795923da2f8af45875d572f0a3efbc25f5d071",
+      "scripts/research/sealed_holdout_rerun.py": "33c294b21db0631a68068dcf135b70db7aae6a1a3e66414bb72959a448883465",
+      "scripts/shadbala.py": "912e0e6d169c2172aab85f71e4347e3e39193020777b43825b775d906400956b",
+      "scripts/varga.py": "4331de5a25ea08729af91aae863943c4223183937f419d52dc63fd6fe7f27be6"
+    },
+    "hash_scope": "explicit_identity_file_sets_not_a_transitive_dependency_lock"
+  },
+  "production_identity_scope": "context_only_shadow_scorer_does_not_call_transition_proximity",
+  "frozen_at_utc": "2026-09-20T04:50:19.815357+00:00",
+  "dataset_path": "references/real_case_calibration/minute_rectification_holdout_v3.json",
+  "dataset_sha256": "45aa76ab79dfc6c6404b6252083e1aa5c7e2e2b7de811fc11042a8ad2da3f0ea",
+  "algorithm_version": "birth-time-event-fact-ranker-v4-shadow",
+  "implementation_sha256": "b15d9ea15227cd58b3097555537fa7c37fb5a993905ada4f5aa04f52de1f8f18",
+  "files": [
+    "scripts/active_rectification_event_engine.py",
+    "scripts/active_rectification_events.py",
+    "scripts/ashtakavarga.py",
+    "scripts/dasha_analyzer.py",
+    "scripts/divisional_charts_extended.py",
+    "scripts/domain_calculation_service.py",
+    "scripts/jaimini.py",
+    "scripts/minute_rectification_fact_ranker_v4.py",
+    "scripts/minute_rectification_feature_facts_v4.py",
+    "scripts/narayana_dasha.py",
+    "scripts/shadbala.py",
+    "scripts/varga.py"
+  ],
+  "historical_frozen_sha256": "f41c298dd6cdcebe7a93e632f7191954be7987f6012ca2a34cecb6e447fbf196",
+  "evaluator_sha256": "33c294b21db0631a68068dcf135b70db7aae6a1a3e66414bb72959a448883465",
+  "ayanamsa": "raman",
+  "node_mode": "mean",
+  "candidate_radius_minutes": [
+    10
+  ],
+  "minute_step": 1,
+  "release_metrics": {
+    "top_1_rate_minimum": 0.6,
+    "top_3_rate_minimum": 0.85,
+    "mean_absolute_minute_error_maximum": 2.0,
+    "false_confirmation_rate_maximum": 0.05,
+    "correct_insufficient_evidence_rejection_rate_minimum": 0.9
+  },
+  "results_previously_seen": true,
+  "official_valid_independent_blind": false,
+  "must_not_use_for_tuning": true,
+  "tie_breaker": "sha256(benchmark_id:case_id:candidate_time); published minute is never passed to the ranker",
+  "metric_rank_definition": "competition_rank_1_plus_strictly_higher_scores_legacy_protocol",
+  "extra_metric_rank_definition": "score_desc_then_existing_opaque_sha256_total_order"
+}