research(rectification): reseal round7_integration_2026_11_09 on staging b453505d (BUG-1809)

Product owner approved 2026-11-09 ("agree new result"). Same protocol as BUG-1720 / BUG-1759: c1a69445^ replay tree + float_split + round5 overlays + b453505d scripts/references + a small replay overlay (new tag paths, retired scorer zero-row adapter, RuntimeError guards). Product default strength scope seed_only_v0, comparability policy common_evidence_coverage_v1. - sealed 20: top-1 0.35 -> 0.30, top-3 0.45 -> 0.45, mean error 6.80 -> 6.75 - offset 900: top-1 0.0156 -> 0.0100, coverage 0.900 -> 0.883 (164/900 vs round5 tag; 57/900 over 4 cases vs pre-integration staging) - live v5 231: best = truth 8 -> 8; 124 changes vs round5 tag, 115 vs pre-integration staging - leaders: sealed 40/40 comparable; offset 863 comparable / 35 provisional / 2 no single leader; live 226 / 5 Trial-for-trial identical to the integration agent's seed_only runs. Sensitivity only, not accuracy (event scoring near chance, BUG-1644). Older tags unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DbBXoyn7Hj3LCW3BUs5bZ8
2026-11-09 07:28:04 +08:00
9 changed files with 63226 additions and 0 deletions
@@ -24397,3 +24397,22 @@
 - 判断：她的生成器修复（§12）在本仓没有可移植对象；快照不重生成、不改。若以后移植她的生成器，需要一起带她的 4 个新用例。
 - 关联记录：无
 - 修复版本：无
+
+## BUG-1809 | 第七轮整合上 staging 后，生时校正三组测试集仍封存在 round5 标签：重封 round7_integration_2026_11_09
+
+- 状态：fixed-pending-verify（本地分支 `codex/rect-reseal-r7-1109`，基于 staging b453505d，未推送；产品负责人 2026-11-09 批准「agree new result」）
+- 首次发现 / 最近更新：2026-11-09（第七轮整合 b453505d 合入 staging 后，封存记录仍是 `round5_pvr_narayana_2026_11_08`）/ 2026-11-09（产品负责人批准，重封）
+- 影响面：只加研究记录与测试，产品代码不改。新文件 `docs/research/sealed_holdout_rerun_round7_integration_2026_11_09.{freeze.json,json}`、`docs/research/reported_offset_round7_integration_2026_11_09.{freeze.json,json}`、`docs/research/live_astrologer_v5_round7_integration_2026_11_09.json`（只存相对真值的分钟数，不含日期）、`docs/research/round7_integration_2026_11_09_comparability.json`、`docs/research/round7_integration_2026_11_09_replay_overlay.patch`；`tests/test_rectification_reseal_round7_bug1809.py`。
+- 现象：b453505d 带来 R1（小运方向默认 p50）、R2 局部传播（默认强弱范围 `seed_only_v0`，BUG-1784）、可比领先（`common_evidence_coverage_v1`，BUG-1801）等改动，回放结果已和 round5 标签不同，但标签没有更新。
+- 重跑（与 BUG-1720 / BUG-1759 同一协议）：回放树 = `git archive c1a69445^` + `float_split_2026_11_08_replay_overlay.patch` + `round5_pvr_narayana_2026_11_08_replay_overlay.patch` + b453505d 的 `scripts/`、`references/` 整体覆盖（退役文件保留回放树版本）+ 新补丁 `round7_integration_2026_11_09_replay_overlay.patch`。新补丁只有三处：两份回放脚本的 FREEZE / REPORT 指向新标签；退役打分器 Narayana 无证据行时按现行引擎「不计分」补 0 分（偏移回放触发 3351 / 1316340 行，封存回放不加载该文件）；回放副本里 `birth_at` 守卫改成 `RuntimeError`。回放实际加载的模块里，仍在 b453505d 的 57 个除两个守卫文件外与 b453505d 逐字节相同。按补丁头部步骤从零重建，`scripts/`、`references/` 与运行用的树完全相同。脚本 `--freeze` 新建记录后回放（`PYTHONHASHSEED=0`）。现行阶段用 `scripts/research/live_astrologer_v5_sensitivity.py`（多记 `leader_coverage`、`duplicate_event_ids`），直接跑 b453505d 工作树。`round5_pvr_narayana_2026_11_08`、`float_split_2026_11_08`、`astrologer_166d_2026_11_05` 及更早标签原样保留。
+- 与整合代理 seed_only 列核对：封存 20/20、偏移 900/900、现行 231/231 逐条相同，覆盖统计两份逐字节相同，没有差异。整合代理报的「偏移 57 次 / 4 例」「现行 115 次」都是对整合前的 staging（未封存）比的，不是对 round5 标签比的；对 round5 标签见下。整合前的 staging 本来就和 round5 标签不完全相同：偏移头名平均误差 20.65 对 20.49，现行 9/231 次不同，来自两次重封之间合入 staging 的其他改动。
+- 结果（旧 `round5_pvr_narayana_2026_11_08` → 新 `round7_integration_2026_11_09`）：
+  - 封存 20 例：top-1 0.35→0.30、top-3 0.45→0.45、平均分钟误差 6.80→6.75；2/20 例变：1 例失去头名（true_rank 1→11，分钟误差 9→8），1 例 true_rank 10→3；误确认 0→0。
+  - 报时偏移 900 次：top-1 0.0156→0.0100、top-3 0.113→0.106、头名平均误差 20.49→20.63 分、交付覆盖率 0.900→0.883；164/900 次变（7/20 例、44/45 格）：true_rank 变好 107 / 变差 39，头名误差变 57 次，头名命中变 5 次，交付覆盖变 15 次，交付宽度变 18 次。对整合前的 staging：57/900 次变（4 例），true_rank 变差 39 / 变好 0。
+  - 现行占星师阶段（holdout v5 77 × 3 = 231）：最佳 = 真值 8→8、真值在前三 19→18、平均最佳误差 10.38→10.29 分、最佳 = 报时 17→16；124/231 次变（49/77 例）：前十 105、最佳分钟 31、匹配档 13、报告置信度 13、引擎置信度 77、头部簇 38。对整合前的 staging：115/231 次变（45 例）。
+  - 可比性（`common_evidence_coverage_v1`）：封存 40 次评分（20 全量 + 20 稀疏）全部 comparable；偏移 900 次 comparable 863、暂列（`provisional_unequal_coverage`）35（全在 1 例）、无单一领先 2（覆盖相同但并列，无可比领先），不可拆 0；现行 231 次 comparable 226、暂列 5（3 例）；重复事件 0。偏移样本同一例重复，不按 900 次独立计。
+- 判断：这些只是敏感度，不是准确率。事件打分接近随机水平（BUG-1644），三组结果早被看过，升降不能当变好或变差，也不能当产品指标。重封理由是记录与产品口径一致。
+- 验证：新测试 `tests/test_rectification_reseal_round7_bug1809.py`：新标签封存 / 偏移 / 现行数字、可比性策略与领先计数、旧标签数字不变（带原值 / 新值 / 原因）。之前没有测试按标签钉封存结果，所以没有要改写的旧测试。`test_active_rectification_api.py::test_long_real_conversation_reaches_vedastro_after_local_range_is_narrow` 已知失败（见 BUG-1720），与本单无关。
+- 防复发：上述测试；以后重封必须新建标签，不覆盖旧标签。
+- 关联记录：BUG-1720、BUG-1759、BUG-1784、BUG-1801、BUG-1644
+- 修复版本：本地分支 `codex/rect-reseal-r7-1109`
@@ -0,0 +1,118 @@
+{
+  "record_version": "reported-offset-native-candidate-date-v3",
+  "frozen_at_utc": "2026-11-08T23:15:54.895716+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": "b32140f91dd8c65a4380ad31a29d77902b2ec2cd06495b9ce387eb7e60b68825",
+  "implementation_sha256_prefix": "b32140f91dd8c65a",
+  "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/functional_benefics.py",
+    "scripts/jaimini.py",
+    "scripts/minute_rectification_fact_ranker_v4.py",
+    "scripts/minute_rectification_feature_facts_v4.py",
+    "scripts/narayana_dasha.py",
+    "scripts/rectification/api_service.py",
+    "scripts/rectification/candidate_contrast.py",
+    "scripts/rectification/candidate_window.py",
+    "scripts/rectification/case_holdout.py",
+    "scripts/rectification/contracts.py",
+    "scripts/rectification/dasha_transition_proximity.py",
+    "scripts/rectification/decision_policy.py",
+    "scripts/rectification/event_probes.py",
+    "scripts/rectification/refinement_packet.py",
+    "scripts/rectification/scoring_service.py",
+    "scripts/shadbala.py",
+    "scripts/varga.py"
+  ],
+  "production_scoring_sha256": "9575adabfc825fe515d571ea2a8e8550b85ccbf089f9e4b185d330821d37ace8",
+  "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": "b7130fd8a9dd0a734d1c10aa0a4900a19702ecf75044ad1740e596f20cada0b5",
+  "file_sha256": {
+    "scripts/active_rectification_event_engine.py": "3bda6fcf8855bec90dd62e64fae167ce3e6afe5956549bbd26587ea59eba3d94",
+    "scripts/active_rectification_events.py": "3d8a625d24aac6f7f05a3265a23b488153daa03b5b81c6b993160011cc07eb1c",
+    "scripts/ashtakavarga.py": "c524da3f54b2ed9c080b78b33854854a0ca3d5d79b7e536bb1c3bc2be3e02852",
+    "scripts/dasha_analyzer.py": "3949b96813217fe624899856afec9f0b787c4cb0fea3a9a0d961990dbd11943a",
+    "scripts/divisional_charts_extended.py": "b13ad68c45685568d6eb17f87017624a9288e31fad74662735cd6d6564f7755d",
+    "scripts/domain_calculation_service.py": "66923a2bc69daf65c965d2973c4b9234edbec5147909eaf3609b6b377ce3c7ca",
+    "scripts/functional_benefics.py": "d34dd7745ba9db0a61e3f7a2aa1e6acd1f819961442cb3412b5dbe22c02ef032",
+    "scripts/jaimini.py": "e474b3a1d5c50de87fafdbdcc3df9bd724de004aa2495463763f5b2688b8f5e2",
+    "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": "4661d725d4f04935169d9d8a86ce8d0ef4061915a3a985c4912b93a6e035689e",
+    "scripts/minute_rectification_holdout_validator.py": "b60b1c752e58672beaee8b36eb3b2348169726cda070ec969db0f13e7ed715af",
+    "scripts/narayana_dasha.py": "5a924f06f76426ab587163fd8c0b5e32a6a8c6917e8670029bd84a2eb6a6172e",
+    "scripts/rectification/api_service.py": "089730097d3872a4f7b5d7dd5574bb7e9ab4d2e246d4a380800bc824d7979b95",
+    "scripts/rectification/candidate_contrast.py": "eee7da71c0574a08876404b78ca111153fa95fb83ace82b9fdca141d1992e273",
+    "scripts/rectification/candidate_window.py": "2fbc5058b12049b15fca375cdc125b8283e19f9f3e4bedcc291e344501891bef",
+    "scripts/rectification/case_holdout.py": "fe0e699b617d9e216acdfeff87af4221c1d6f4c101bf33e1dbf14e2bcd75a8b5",
+    "scripts/rectification/contracts.py": "c6dc6bd230e38f9f2126c224aae9ddbd6415304c15340f8c7cd3f06f6ef3160f",
+    "scripts/rectification/dasha_transition_proximity.py": "66d29c0b182309d17264dd2645b5cb426e28b4066d6c7004fa66f3334286753f",
+    "scripts/rectification/decision_policy.py": "80e8de4680d19a052213bb821d2a9c94143a2962235aeb27f6ad81374d2e60d5",
+    "scripts/rectification/event_probes.py": "283ac525efa5ba5b12e804d9b87e6725741ede3365659002224fffe61722bc3f",
+    "scripts/rectification/refinement_packet.py": "dbeea5aa780889785ce850f3881e8b7f67bb44594e9bc382c65b2e730884a2f4",
+    "scripts/rectification/scoring_service.py": "d1530ea3569f86747c4e6d7bccf18639e50c870526c06702bc1d9b3fec07ed9b",
+    "scripts/research/cluster_width_lib.py": "1d8f2ffa7795d62069deebb2de8e16749317037d4ed3f24a2c0092676e5cfcc6",
+    "scripts/research/probe_supply_after_six.py": "098a5b4398d6f8d996917d809c2fc922ad64031d3b70660f8ef7c876aba92438",
+    "scripts/research/reported_offset_sweep.py": "1ef4645a954da5d8374efc807aeeb2daf6e3431db771ed1e346b46ebbe83de62",
+    "scripts/research/sealed_holdout_rerun.py": "388b8af89fb89da9c3f2ce0e2bb6351ed20fd54fa19b8be34438540eed128617",
+    "scripts/shadbala.py": "cf1127cb60ce9aedc01050e7e391257d1c354b9ca735c7a2293baefb7d57e343",
+    "scripts/varga.py": "c1c8521320176e31b1f897807e50322f147654251c323731666482fc98f6df0a"
+  },
+  "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": "1ef4645a954da5d8374efc807aeeb2daf6e3431db771ed1e346b46ebbe83de62",
+  "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 +1,42 @@
+{
+ "tag": "round7_integration_2026_11_09",
+ "source_commit": "b453505d50f8d9c5ff4a472eb225a0c60a68e607",
+ "policy": "common_evidence_coverage_v1",
+ "policy_summary": "evidence unit = (event, layer); equal coverage -> comparable (comparable_leader = provisional_leader = winning_segment); unequal -> provisional_unequal_coverage, re-rank on units every candidate has; merged layers that cannot be split -> not_reducible (BUG-1801)",
+ "narayana_strength_scope": "seed_only_v0",
+ "scope": "coverage counts only, not accuracy (BUG-1644)",
+ "probe": "product adjudicator compute_event_candidate_result on the replay script's own requests (sealed: full + sparse per case; offset: every shifted window); live: chart_choice astrologer leader_coverage",
+ "sealed_20": {
+  "evaluations": 40,
+  "leader_status": {
+   "comparable": 40
+  },
+  "coverage_comparison": {
+   "comparable/raw_totals": 40
+  }
+ },
+ "reported_offset_900": {
+  "evaluations": 900,
+  "leader_status": {
+   "comparable": 863,
+   "provisional_unequal_coverage": 35,
+   "no_single_leader": 2
+  },
+  "coverage_comparison": {
+   "comparable/raw_totals": 865,
+   "provisional_unequal_coverage/common_coverage": 35
+  },
+  "not_reducible": 0,
+  "provisional_cases": 1,
+  "no_comparable_leader": 2
+ },
+ "live_v5_231": {
+  "evaluations": 231,
+  "leader_status": {
+   "comparable": 226,
+   "provisional_unequal_coverage": 5
+  },
+  "provisional_cases": 3,
+  "duplicate_event_runs": 0
+ }
+}
@@ -0,0 +1,128 @@
+# BUG-1809 回放树：round7_integration_2026_11_09（staging b453505d）
+#
+# 构建步骤（与 BUG-1720 / BUG-1759 同一协议）：
+#   1. git archive c1a69445^ （两份回放脚本与退役打分器随 BUG-1644 删除，从这里取）
+#   2. patch -p1 < docs/research/float_split_2026_11_08_replay_overlay.patch
+#   3. patch -p1 < docs/research/round5_pvr_narayana_2026_11_08_replay_overlay.patch
+#   4. git archive b453505d scripts references | tar -x  （staging 的 scripts/、references/ 整体覆盖；
+#      仓库已删的退役文件保留回放树版本。两份回放实际加载的 67 个模块里，仍在 b453505d 的 57 个除 (c) 两个守卫文件外逐字节相同）
+#   5. 本补丁（三部分）：
+#      (a) 两份回放脚本 FREEZE / REPORT 指向新标签（r5tree → r7tree）；
+#      (b) 退役打分器 scoring_service.py 的回放适配：Narayana 定不出、该候选该事件无证据行时按现行引擎「不计分」补 0 分
+#          （BUG-1769 已记录的 IndexError 路径；本次偏移回放 3351 / 1316340 行触发，封存回放不加载该文件）；
+#      (c) 回放副本里 birth_at 守卫由 TypeError 改为 RuntimeError（REPLAY_GUARD），避免任何调用方 except TypeError 吞掉日期字符串；
+#          staging → r7guard。
+#   6. python scripts/research/sealed_holdout_rerun.py --freeze；reported_offset_sweep.py --freeze；再各自回放（PYTHONHASHSEED=0）。
+diff -ruN r5tree/scripts/rectification/scoring_service.py r7tree/scripts/rectification/scoring_service.py
+--- r5tree/scripts/rectification/scoring_service.py	2026-11-09 07:17:58.857906592 +0800
+++ r7tree/scripts/rectification/scoring_service.py	2026-11-09 07:17:58.861906532 +0800
+@@ -254,6 +254,17 @@
+     return {**request, "events": [event for event in request["events"] if is_scoreable_event(event)]}
+ 
+ 
+import atexit as _atexit, sys as _sys
+_REPLAY_EMPTY = [0]
+_REPLAY_TOTAL = [0]
+_atexit.register(lambda: print(f"REPLAY_ADAPTER empty_evidence_rows={_REPLAY_EMPTY[0]} of {_REPLAY_TOTAL[0]}", file=_sys.stderr))
+
+
+def _replay_empty(event):
+    _REPLAY_EMPTY[0] += 1
+    return {"event_id": event["id"], "points": 0.0, "rule_ids": []}
+
+
+ def build_event_contribution_matrix(
+     request: RectificationRequest,
+     row_provider: Callable[[dict[str, Any]], Sequence[CandidateScoreRow]] | None = None,
+@@ -277,10 +288,14 @@
+         sample_rows = []
+         for sampled in samples:
+             rows = list(provider(_legacy_request(scoring_request, event, sampled)))
+            _REPLAY_TOTAL[0] += len(rows)
+             sample_rows.append([
+                 {**row, "score": adjusted["points"], "evidence": [adjusted]}
+                 for row in rows
+-                for adjusted in [_kind_adjusted_evidence(event, row["evidence"][0])]
+                # REPLAY ADAPTER (round9 integration, 2026-11-09): a candidate whose mandatory layer is missing for this
+                # event (e.g. Narayana blocked) has no evidence row; the product adds nothing for it, so it counts 0 here.
+                for adjusted in [_kind_adjusted_evidence(event, row["evidence"][0]) if row["evidence"]
+                                 else _replay_empty(event)]
+             ])
+         grids = [[row["time"] for row in rows] for rows in sample_rows]
+         if any(grid != grids[0] for grid in grids[1:]) or (candidate_grid is not None and grids[0] != candidate_grid):
+diff -ruN r5tree/scripts/research/reported_offset_sweep.py r7tree/scripts/research/reported_offset_sweep.py
+--- r5tree/scripts/research/reported_offset_sweep.py	2026-11-09 07:17:58.857906592 +0800
+++ r7tree/scripts/research/reported_offset_sweep.py	2026-11-09 07:17:58.861906532 +0800
+@@ -66,8 +66,11 @@
+ # 原值: docs/research/reported_offset_float_split_2026_11_08.{freeze.json,json}
+ # 新值: docs/research/reported_offset_round5_pvr_narayana_2026_11_08.{freeze.json,json}
+ # 原因: 产品 2026-11-08 批准；占星师第五轮裁定 1、3；与报告/卡口径一致（BUG-1759：校正功能吉凶改用默认 PVR 表 30，Narayana 改用来源版 narayana_rath、出生时刻起算；回放结果变了，另存新标签）
+-FREEZE = ROOT / "docs/research/reported_offset_round5_pvr_narayana_2026_11_08.freeze.json"
+-REPORT = ROOT / "docs/research/reported_offset_round5_pvr_narayana_2026_11_08.json"
+# 原值: docs/research/reported_offset_round5_pvr_narayana_2026_11_08.{freeze.json,json}
+# 新值: docs/research/reported_offset_round7_integration_2026_11_09.{freeze.json,json}
+# 原因: 产品负责人 2026-11-09 批准；第七轮整合（R1 默认 p50、R2 局部传播 seed_only、可比领先）；BUG-1809（staging b453505d 全部 scripts/references 叠到回放树，回放结果变了，另存新标签）
+FREEZE = ROOT / "docs/research/reported_offset_round7_integration_2026_11_09.freeze.json"
+REPORT = ROOT / "docs/research/reported_offset_round7_integration_2026_11_09.json"
+ LEGACY_REPORT = ROOT / "docs/research/reported_offset_2026_09_20.json"
+ 
+ 
+diff -ruN r5tree/scripts/research/sealed_holdout_rerun.py r7tree/scripts/research/sealed_holdout_rerun.py
+--- r5tree/scripts/research/sealed_holdout_rerun.py	2026-11-09 07:17:58.857906592 +0800
+++ r7tree/scripts/research/sealed_holdout_rerun.py	2026-11-09 07:17:58.861906532 +0800
+@@ -63,8 +63,11 @@
+ # 原值: docs/research/sealed_holdout_rerun_float_split_2026_11_08.{freeze.json,json}
+ # 新值: docs/research/sealed_holdout_rerun_round5_pvr_narayana_2026_11_08.{freeze.json,json}
+ # 原因: 产品 2026-11-08 批准；占星师第五轮裁定 1、3；与报告/卡口径一致（BUG-1759：校正功能吉凶改用默认 PVR 表 30，Narayana 改用来源版 narayana_rath、出生时刻起算；回放结果变了，另存新标签）
+-FREEZE = ROOT / "docs/research/sealed_holdout_rerun_round5_pvr_narayana_2026_11_08.freeze.json"
+-REPORT = ROOT / "docs/research/sealed_holdout_rerun_round5_pvr_narayana_2026_11_08.json"
+# 原值: docs/research/sealed_holdout_rerun_round5_pvr_narayana_2026_11_08.{freeze.json,json}
+# 新值: docs/research/sealed_holdout_rerun_round7_integration_2026_11_09.{freeze.json,json}
+# 原因: 产品负责人 2026-11-09 批准；第七轮整合（R1 默认 p50、R2 局部传播 seed_only、可比领先）；BUG-1809（staging b453505d 全部 scripts/references 叠到回放树，回放结果变了，另存新标签）
+FREEZE = ROOT / "docs/research/sealed_holdout_rerun_round7_integration_2026_11_09.freeze.json"
+REPORT = ROOT / "docs/research/sealed_holdout_rerun_round7_integration_2026_11_09.json"
+ LEGACY_REPORT = ROOT / "docs/research/sealed_holdout_rerun_2026_09_20.json"
+ ARCHIVE = ROOT / "docs/research/history/rectification_pre_cross_midnight_2026_09_20"
+ PRODUCTION_FILES = [
+diff -ruN stg/scripts/active_rectification_event_engine.py r7g/scripts/active_rectification_event_engine.py
+--- staging/scripts/active_rectification_event_engine.py	2026-11-09 07:17:58.869906413 +0800
+++ r7guard/scripts/active_rectification_event_engine.py	2026-11-09 07:17:58.869906413 +0800
+@@ -107,7 +107,7 @@
+     and started every period from midnight of that day.
+     """
+     if not isinstance(birth_at, datetime):
+-        raise TypeError("birth_at must be the candidate birth datetime")
+        raise RuntimeError("REPLAY_GUARD birth_at must be the candidate birth datetime")
+     timeline, _, _, _ = birth_moment.mahadashas(birth_at, float(moon_longitude))
+     return timeline
+ 
+@@ -125,7 +125,7 @@
+     with ``birth_moment.split_period`` at both levels.
+     """
+     if not isinstance(birth_at, datetime):
+-        raise TypeError("birth_at must be the candidate birth datetime")
+        raise RuntimeError("REPLAY_GUARD birth_at must be the candidate birth datetime")
+     if timeline is None:
+         timeline = _vimshottari_timeline(birth_at, moon_longitude)
+     _, major = dasha_analyzer.find_current(timeline, event_at)
+@@ -200,7 +200,7 @@
+     step; there is no fallback to the old algorithm.
+     """
+     if not isinstance(birth_at, datetime):
+-        raise TypeError("birth_at must be the candidate birth datetime")
+        raise RuntimeError("REPLAY_GUARD birth_at must be the candidate birth datetime")
+     if periods is None:
+         periods = _narayana_periods(ascendant_index, planet_longitudes)
+     if not periods:
+diff -ruN stg/scripts/rectification/event_probes.py r7g/scripts/rectification/event_probes.py
+--- staging/scripts/rectification/event_probes.py	2026-11-09 07:17:58.869906413 +0800
+++ r7guard/scripts/rectification/event_probes.py	2026-11-09 07:17:58.873906354 +0800
+@@ -509,7 +509,7 @@
+     if birth_at is None:
+         return None
+     if not isinstance(birth_at, datetime):
+-        raise TypeError("birth_at must be the candidate birth datetime")  # a date string would silently drop the track
+        raise RuntimeError("REPLAY_GUARD birth_at must be the candidate birth datetime")  # a date string would silently drop the track
+     periods = _narayana_periods(ascendant_index, planet_longitudes)
+     if not periods:
+         return None
@@ -0,0 +1,116 @@
+{
+  "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/functional_benefics.py",
+      "scripts/jaimini.py",
+      "scripts/minute_rectification_fact_ranker_v4.py",
+      "scripts/minute_rectification_feature_facts_v4.py",
+      "scripts/narayana_dasha.py",
+      "scripts/rectification/api_service.py",
+      "scripts/rectification/candidate_contrast.py",
+      "scripts/rectification/candidate_window.py",
+      "scripts/rectification/case_holdout.py",
+      "scripts/rectification/contracts.py",
+      "scripts/rectification/dasha_transition_proximity.py",
+      "scripts/rectification/decision_policy.py",
+      "scripts/rectification/event_probes.py",
+      "scripts/rectification/refinement_packet.py",
+      "scripts/rectification/scoring_service.py",
+      "scripts/shadbala.py",
+      "scripts/varga.py"
+    ],
+    "production_scoring_sha256": "9575adabfc825fe515d571ea2a8e8550b85ccbf089f9e4b185d330821d37ace8",
+    "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": "b7130fd8a9dd0a734d1c10aa0a4900a19702ecf75044ad1740e596f20cada0b5",
+    "file_sha256": {
+      "scripts/active_rectification_event_engine.py": "3bda6fcf8855bec90dd62e64fae167ce3e6afe5956549bbd26587ea59eba3d94",
+      "scripts/active_rectification_events.py": "3d8a625d24aac6f7f05a3265a23b488153daa03b5b81c6b993160011cc07eb1c",
+      "scripts/ashtakavarga.py": "c524da3f54b2ed9c080b78b33854854a0ca3d5d79b7e536bb1c3bc2be3e02852",
+      "scripts/dasha_analyzer.py": "3949b96813217fe624899856afec9f0b787c4cb0fea3a9a0d961990dbd11943a",
+      "scripts/divisional_charts_extended.py": "b13ad68c45685568d6eb17f87017624a9288e31fad74662735cd6d6564f7755d",
+      "scripts/domain_calculation_service.py": "66923a2bc69daf65c965d2973c4b9234edbec5147909eaf3609b6b377ce3c7ca",
+      "scripts/functional_benefics.py": "d34dd7745ba9db0a61e3f7a2aa1e6acd1f819961442cb3412b5dbe22c02ef032",
+      "scripts/jaimini.py": "e474b3a1d5c50de87fafdbdcc3df9bd724de004aa2495463763f5b2688b8f5e2",
+      "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": "4661d725d4f04935169d9d8a86ce8d0ef4061915a3a985c4912b93a6e035689e",
+      "scripts/minute_rectification_holdout_validator.py": "b60b1c752e58672beaee8b36eb3b2348169726cda070ec969db0f13e7ed715af",
+      "scripts/narayana_dasha.py": "5a924f06f76426ab587163fd8c0b5e32a6a8c6917e8670029bd84a2eb6a6172e",
+      "scripts/rectification/api_service.py": "089730097d3872a4f7b5d7dd5574bb7e9ab4d2e246d4a380800bc824d7979b95",
+      "scripts/rectification/candidate_contrast.py": "eee7da71c0574a08876404b78ca111153fa95fb83ace82b9fdca141d1992e273",
+      "scripts/rectification/candidate_window.py": "2fbc5058b12049b15fca375cdc125b8283e19f9f3e4bedcc291e344501891bef",
+      "scripts/rectification/case_holdout.py": "fe0e699b617d9e216acdfeff87af4221c1d6f4c101bf33e1dbf14e2bcd75a8b5",
+      "scripts/rectification/contracts.py": "c6dc6bd230e38f9f2126c224aae9ddbd6415304c15340f8c7cd3f06f6ef3160f",
+      "scripts/rectification/dasha_transition_proximity.py": "66d29c0b182309d17264dd2645b5cb426e28b4066d6c7004fa66f3334286753f",
+      "scripts/rectification/decision_policy.py": "80e8de4680d19a052213bb821d2a9c94143a2962235aeb27f6ad81374d2e60d5",
+      "scripts/rectification/event_probes.py": "283ac525efa5ba5b12e804d9b87e6725741ede3365659002224fffe61722bc3f",
+      "scripts/rectification/refinement_packet.py": "dbeea5aa780889785ce850f3881e8b7f67bb44594e9bc382c65b2e730884a2f4",
+      "scripts/rectification/scoring_service.py": "d1530ea3569f86747c4e6d7bccf18639e50c870526c06702bc1d9b3fec07ed9b",
+      "scripts/research/cluster_width_lib.py": "1d8f2ffa7795d62069deebb2de8e16749317037d4ed3f24a2c0092676e5cfcc6",
+      "scripts/research/probe_supply_after_six.py": "098a5b4398d6f8d996917d809c2fc922ad64031d3b70660f8ef7c876aba92438",
+      "scripts/research/reported_offset_sweep.py": "1ef4645a954da5d8374efc807aeeb2daf6e3431db771ed1e346b46ebbe83de62",
+      "scripts/research/sealed_holdout_rerun.py": "388b8af89fb89da9c3f2ce0e2bb6351ed20fd54fa19b8be34438540eed128617",
+      "scripts/shadbala.py": "cf1127cb60ce9aedc01050e7e391257d1c354b9ca735c7a2293baefb7d57e343",
+      "scripts/varga.py": "c1c8521320176e31b1f897807e50322f147654251c323731666482fc98f6df0a"
+    },
+    "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": "2026-11-08T23:15:54.758056+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": "b32140f91dd8c65a4380ad31a29d77902b2ec2cd06495b9ce387eb7e60b68825",
+  "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": "388b8af89fb89da9c3f2ce0e2bb6351ed20fd54fa19b8be34438540eed128617",
+  "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 +1,80 @@
+"""BUG-1809: rectification test sets resealed on the round-7 integration (staging b453505d).
+
+The tag pins sensitivity measures, not accuracy (event scoring is near chance, BUG-1644).
+Older tags must stay byte-for-byte what they were; this file only adds a new tag.
+"""
+import json
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+RESEARCH = ROOT / "docs/research"
+TAG = "round7_integration_2026_11_09"
+OLD = "round5_pvr_narayana_2026_11_08"
+REASON = "产品负责人 2026-11-09 批准；第七轮整合（R1 默认 p50、R2 局部传播 seed_only、可比领先）"
+
+
+def _load(name: str) -> dict:
+    return json.loads((RESEARCH / name).read_text(encoding="utf-8"))
+
+
+def test_sealed_20_pins_the_new_tag() -> None:
+    # 原值：sealed_holdout_rerun_round5_pvr_narayana_2026_11_08 top-1 0.35、top-3 0.45、平均分钟误差 6.80
+    # 新值：sealed_holdout_rerun_round7_integration_2026_11_09 top-1 0.30、top-3 0.45、平均分钟误差 6.75
+    # 原因：产品负责人 2026-11-09 批准；第七轮整合（R1 默认 p50、R2 局部传播 seed_only、可比领先）
+    report = _load(f"sealed_holdout_rerun_{TAG}.json")
+    assert report["trial_count"] == 20 and report["implementation_hash_matches_at_replay"] is True
+    assert report["metrics"]["top_1_rate"] == 0.3
+    assert report["metrics"]["top_3_rate"] == 0.45
+    assert report["metrics"]["mean_absolute_minute_error"] == 6.75
+    assert report["metrics"]["false_confirmation_rate"] == 0.0
+    assert report["verified_minute_claim_allowed"] is False and report["results_previously_seen"] is True
+    assert report["status"] == "blocked_independent_blind_evidence"
+
+
+def test_reported_offset_900_pins_the_new_tag() -> None:
+    # 原值：reported_offset_round5_pvr_narayana_2026_11_08 top-1 0.0156、交付覆盖率 0.900
+    # 新值：reported_offset_round7_integration_2026_11_09 top-1 0.0100、交付覆盖率 0.883（164/900 次、7 例变）
+    # 原因：产品负责人 2026-11-09 批准；第七轮整合（R1 默认 p50、R2 局部传播 seed_only、可比领先）
+    report = _load(f"reported_offset_{TAG}.json")
+    trials = report["trials"]
+    assert len(trials) == 900 and report["implementation_hash_matches_at_replay"] is True
+    assert report["freeze_record_path"] == f"docs/research/reported_offset_{TAG}.freeze.json"
+    assert round(sum(t["top_1_hit"] for t in trials) / 900, 4) == 0.01
+    assert round(sum(t["delivery_covers_truth"] for t in trials) / 900, 4) == 0.8833
+    old = {(t["case_ordinal"], t["offset_minutes"], t["radius_minutes"]): t for t in _load(f"reported_offset_{OLD}.json")["trials"]}
+    changed = [t for t in trials if old[(t["case_ordinal"], t["offset_minutes"], t["radius_minutes"])] != t]
+    assert len(changed) == 164 and len({t["case_ordinal"] for t in changed}) == 7
+
+
+def test_live_v5_and_comparability_are_recorded_in_the_tag() -> None:
+    # 原值：live_astrologer_v5_round5_pvr_narayana_2026_11_08 最佳 = 真值 8、真值在前三 19
+    # 新值：live_astrologer_v5_round7_integration_2026_11_09 最佳 = 真值 8、真值在前三 18；
+    #       可比性策略 common_evidence_coverage_v1，暂列 / 可比领先计数写进标签
+    # 原因：产品负责人 2026-11-09 批准；第七轮整合（R1 默认 p50、R2 局部传播 seed_only、可比领先）
+    live = _load(f"live_astrologer_v5_{TAG}.json")
+    assert live["tag"] == TAG and live["baseline_tag"] == OLD
+    assert len(live["trials"]) == 231 and live["summary"]["new"]["best_is_truth"] == 8
+    assert live["summary"]["changed"]["any"] == 124
+    assert live["summary"]["pre_integration_staging"]["new_vs_it"]["any"] == 115
+    assert live["narayana_strength_scope"].startswith("seed_only_v0")
+    assert live["comparability_policy"].startswith("common_evidence_coverage_v1")
+    assert live["leader_coverage_status"] == {"comparable": 226, "provisional_unequal_coverage": 5}
+    comparability = _load(f"{TAG}_comparability.json")
+    assert comparability["policy"] == "common_evidence_coverage_v1"
+    assert comparability["sealed_20"]["leader_status"] == {"comparable": 40}
+    assert comparability["reported_offset_900"]["leader_status"] == {
+        "comparable": 863, "provisional_unequal_coverage": 35, "no_single_leader": 2}
+    assert comparability["live_v5_231"]["leader_status"] == live["leader_coverage_status"]
+
+
+def test_older_tags_stay_intact() -> None:
+    sealed = _load(f"sealed_holdout_rerun_{OLD}.json")["metrics"]
+    assert (sealed["top_1_rate"], sealed["top_3_rate"], sealed["mean_absolute_minute_error"]) == (0.35, 0.45, 6.8)
+    offset = _load(f"reported_offset_{OLD}.json")["trials"]
+    assert round(sum(t["top_1_hit"] for t in offset) / 900, 4) == 0.0156
+    assert _load(f"live_astrologer_v5_{OLD}.json")["summary"]["new"]["best_is_truth"] == 8
+    for old_tag in ("float_split_2026_11_08", "astrologer_166d_2026_11_05", OLD):
+        for kind in ("sealed_holdout_rerun", "reported_offset"):
+            assert (RESEARCH / f"{kind}_{old_tag}.json").exists()
+            assert (RESEARCH / f"{kind}_{old_tag}.freeze.json").exists()
+    assert REASON in Path(__file__).read_text(encoding="utf-8")