feat: sync skill truth audit safeguards

tests/test_cross_repo_strategy_diff_audit.py

@@ -0,0 +1,30 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[1]
+AUDIT = ROOT / "references" / "cross_project_contract" / "cross_repo_strategy_diff_audit_2026_07_19.json"
+
+
+def test_strategy_diff_audit_blocks_blind_overwrite() -> None:
+    data = json.loads(AUDIT.read_text(encoding="utf-8"))
+
+    assert data["scope"] == "cross_repo_strategy_diff_audit"
+    assert data["status"] == "do_not_blind_overwrite"
+    assert data["summary"]["diff_count"] == 7
+
+
+def test_strategy_diff_audit_tracks_vedastro_and_holdout_policy_diffs() -> None:
+    data = json.loads(AUDIT.read_text(encoding="utf-8"))
+    items = {item["path"]: item for item in data["items"]}
+
+    assert "scripts/day_level_holdout_validator.py" in items
+    assert "commercial_retains_observational_validation_mode_for_product_intake" in items[
+        "scripts/day_level_holdout_validator.py"
+    ]["notes"]
+    assert "scripts/vedastro_identity_archive.py" in items
+    assert "research_has_stricter_self_host_truth_upgrade_gate" in items[
+        "scripts/vedastro_identity_archive.py"
+    ]["notes"]

tests/test_day_level_holdout_pilot_source_queue.py

@@ -0,0 +1,33 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+QUEUE = ROOT / "references" / "real_case_calibration" / "day_level_holdout_v3_pilot_source_queue_2026_07_19.json"
+
+
+def test_pilot_source_queue_has_three_public_subjects_but_no_truth_upgrade() -> None:
+    data = json.loads(QUEUE.read_text(encoding="utf-8"))
+
+    assert data["status"] == "candidate_requires_adjudication"
+    assert data["production_tuning_allowed"] is False
+    assert data["candidate_count"] == 3
+    assert "not holdout annotations" in data["truth_boundary"]
+    assert "independent adjudicator" in data["required_next_step"]
+
+    subjects = {subject["subject_id"]: subject for subject in data["subjects"]}
+    assert set(subjects) == {"steve_jobs", "barack_obama", "albert_einstein"}
+    for subject in subjects.values():
+        assert subject["candidate_positive_event"]["status"] == "candidate_requires_adjudication"
+        assert len(subject["candidate_negative_windows"]) == 2
+        assert all(window["status"] == "candidate_requires_adjudication" for window in subject["candidate_negative_windows"])
+        assert all(url.startswith("https://") for url in subject["candidate_positive_event"]["source_urls"])
+
+
+def test_pilot_source_queue_is_not_misrepresented_as_day_level_holdout_manifest() -> None:
+    data = json.loads(QUEUE.read_text(encoding="utf-8"))
+
+    assert "annotations" not in data
+    assert data["scope"] == "day_level_holdout_v3_pilot_source_queue"
+    assert data["status"] != "ready_for_blind_replay"

tests/test_day_level_holdout_pilot_source_queue_report.py

@@ -0,0 +1,40 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+from scripts.day_level_holdout_pilot_source_queue_report import build_report
+
+ROOT = Path(__file__).resolve().parents[1]
+QUEUE = ROOT / "references/real_case_calibration/day_level_holdout_v3_pilot_source_queue_2026_07_19.json"
+
+
+def test_pilot_source_queue_expands_to_nine_unscored_windows() -> None:
+    report = build_report(QUEUE)
+
+    assert report["scope"] == "day_level_holdout_pilot_source_queue_report"
+    assert report["status"] == "awaiting_independent_human_labels"
+    assert report["production_tuning_allowed"] is False
+    assert report["blind_scoring_allowed"] is False
+    assert report["summary"] == {
+        "subject_count": 3,
+        "window_count": 9,
+        "positive_candidate_count": 3,
+        "negative_candidate_count": 6,
+        "ready_annotation_count": 0,
+        "blocked_annotation_count": 9,
+    }
+
+
+def test_pilot_source_queue_windows_keep_public_source_and_boundary() -> None:
+    report = build_report(QUEUE)
+
+    for window in report["windows"]:
+        assert window["claim_status"] == "candidate_not_label"
+        assert window["required_next_step"] == "independent_human_adjudication"
+        assert window["scoring_status"] == "blocked_not_frozen"
+        if window["label_candidate"] == "target_event":
+            assert window["source_urls"]
+            assert all(url.startswith("https://") for url in window["source_urls"])
+        else:
+            assert window["event_absent_assertion"]
+

tests/test_skill_truth_boundary_text.py

@@ -0,0 +1,19 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+SKILL = ROOT / "SKILL.md"
+
+
+def test_skill_text_requires_effective_capability_view_before_claims() -> None:
+    text = SKILL.read_text(encoding="utf-8")
+    assert "references/oracle/effective_skill_capability_view_2026_07_19.json" in text
+    assert "references/oracle/skill_truth_overlay_2026_07_19.json" in text
+    assert "不得直接把 `references/technique_registry.json` 的旧 `covered` 当作完整闭环" in text
+
+
+def test_skill_text_does_not_overclaim_kp_muhurta_sahams() -> None:
+    text = SKILL.read_text(encoding="utf-8")
+    assert "KP/Muhurta/Gochara/Sahams/Sphuta/Tajika等高阶分支必须按 skill truth overlay 降级使用" in text
+    assert "KP系统 | reference-only / partial" in text

tests/test_technique_promotion_audit_vimsopaka_avastha.py

@@ -0,0 +1,42 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+from scripts.technique_promotion_audit_vimsopaka_avastha import build_audit
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def test_vimsopaka_avastha_audit_corrects_call_status() -> None:
+    audit = build_audit(ROOT)
+    assert audit["scope"] == "technique_promotion_audit_vimsopaka_avastha"
+    assert audit["truth_policy"] == "runtime_presence_not_oracle_closure"
+    assert audit["production_tuning_allowed"] is False
+    statuses = {item["technique_id"]: item["current_call_status"] for item in audit["items"]}
+    assert statuses["vimsopaka_bala"] == "formally_called_in_full_reading"
+    assert statuses["avastha_states"] == "formally_called_via_deep_varga_avastha_endpoint"
+
+
+def test_vimsopaka_avastha_audit_records_entrypoints_and_boundaries() -> None:
+    audit = build_audit(ROOT)
+    for item in audit["items"]:
+        assert item["main_artifacts"]
+        assert item["entrypoints"]
+        assert item["next_action"] in {
+            "add_formula_source_and_oracle_packet",
+            "add_display_contract_and_source_oracle_packet",
+        }
+        assert item["claim_boundary"]
+        assert item["reuse_decision"] == "do_not_duplicate_runtime"
+
+
+def test_vimsopaka_avastha_audit_artifact_exists() -> None:
+    artifact = ROOT / "references/oracle/technique_promotion_audit_vimsopaka_avastha_2026_07_19.json"
+    data = json.loads(artifact.read_text(encoding="utf-8"))
+    assert data["summary"] == {
+        "items_checked": 2,
+        "formally_called_count": 2,
+        "duplicate_runtime_needed": 0,
+    }
+