evalvault 1.65.0__py3-none-any.whl → 1.66.0__py3-none-any.whl

evalvault/domain/services/prompt_registry.py

@@ -109,30 +109,59 @@ def build_prompt_summary(bundle: PromptSetBundle) -> dict[str, Any]:
 
 def build_prompt_inputs_from_snapshots(
     snapshots: dict[str, dict[str, Any]] | None,
+    *,
+    kind: PromptKind = "ragas",
+    source: str | None = None,
 ) -> list[PromptInput]:
     if not snapshots:
         return []
     prompt_inputs: list[PromptInput] = []
     for metric_name, entry in snapshots.items():
-        prompt_text = entry.get("prompt") if isinstance(entry, dict) else None
+        if not isinstance(entry, dict):
+            continue
+        entry_source = entry.get("source")
+        resolved_source = source if source else entry_source
+        metadata = {key: value for key, value in entry.items() if key != "prompt"}
+
+        prompts_map = entry.get("prompts")
+        if isinstance(prompts_map, dict) and prompts_map:
+            for prompt_key, prompt_text in prompts_map.items():
+                if not isinstance(prompt_text, str):
+                    continue
+                normalized = prompt_text.strip()
+                if not normalized:
+                    continue
+                prompt_inputs.append(
+                    PromptInput(
+                        content=normalized,
+                        name=f"{kind}.{metric_name}.{prompt_key}",
+                        kind=kind,
+                        role=f"{metric_name}.{prompt_key}",
+                        source=(
+                            resolved_source
+                            if isinstance(resolved_source, str) and resolved_source
+                            else kind
+                        ),
+                        metadata=metadata or None,
+                    )
+                )
+            continue
+
+        prompt_text = entry.get("prompt")
         if not isinstance(prompt_text, str):
             continue
         prompt_text = prompt_text.strip()
         if not prompt_text:
             continue
-        source = entry.get("source") if isinstance(entry, dict) else None
-        metadata = {
-            key: value
-            for key, value in entry.items()
-            if key != "prompt" and isinstance(entry, dict)
-        }
         prompt_inputs.append(
             PromptInput(
                 content=prompt_text,
-                name=f"ragas.{metric_name}",
-                kind="ragas",
+                name=f"{kind}.{metric_name}",
+                kind=kind,
                 role=str(metric_name),
-                source=source if isinstance(source, str) and source else "ragas",
+                source=resolved_source
+                if isinstance(resolved_source, str) and resolved_source
+                else kind,
                 metadata=metadata or None,
             )
         )

evalvault/domain/services/threshold_profiles.py

@@ -8,6 +8,10 @@ SUMMARY_RECOMMENDED_THRESHOLDS = {
     "summary_faithfulness": 0.90,
     "summary_score": 0.85,
     "entity_preservation": 0.90,
+    "summary_accuracy": 0.90,
+    "summary_risk_coverage": 0.90,
+    "summary_non_definitive": 0.80,
+    "summary_needs_followup": 0.80,
 }
 QA_RECOMMENDED_THRESHOLDS = {
     "faithfulness": 0.70,

evalvault/domain/services/visual_space_service.py

@@ -233,13 +233,13 @@ def _aggregate_stage_metrics(metrics: Iterable[StageMetric]) -> dict[str, dict[s
 
     aggregated: dict[str, dict[str, float]] = {}
     for name, entries in buckets.items():
-        scores = [m.score for m in entries]
+        scores = [m.score for m in entries if m.score is not None]
         threshold = next(
             (m.threshold for m in entries if m.threshold is not None),
             DEFAULT_STAGE_THRESHOLDS.get(name),
         )
         aggregated[name] = {
-            "avg": mean(scores) if scores else None,
+            "avg": mean(scores) if scores else 0.0,
             "threshold": threshold if threshold is not None else DEFAULT_METRIC_THRESHOLD,
         }
     return aggregated
@@ -770,6 +770,77 @@ def _build_case_coords(result: TestCaseResult) -> dict[str, float | None]:
             ),
         ]
     )
+
+    if x_value is None:
+        x_value = _weighted_average(
+            [
+                (
+                    _centered_norm(
+                        scores.get("summary_accuracy"), thresholds.get("summary_accuracy")
+                    ),
+                    0.4,
+                ),
+                (
+                    _centered_norm(
+                        scores.get("summary_risk_coverage"),
+                        thresholds.get("summary_risk_coverage"),
+                    ),
+                    0.3,
+                ),
+                (
+                    _centered_norm(
+                        scores.get("summary_faithfulness"),
+                        thresholds.get("summary_faithfulness"),
+                    ),
+                    0.2,
+                ),
+                (
+                    _centered_norm(scores.get("summary_score"), thresholds.get("summary_score")),
+                    0.1,
+                ),
+                (
+                    _centered_norm(
+                        scores.get("entity_preservation"),
+                        thresholds.get("entity_preservation"),
+                    ),
+                    0.2,
+                ),
+            ]
+        )
+
+    if y_value is None:
+        y_value = _weighted_average(
+            [
+                (
+                    _centered_norm(
+                        scores.get("summary_accuracy"), thresholds.get("summary_accuracy")
+                    ),
+                    0.35,
+                ),
+                (
+                    _centered_norm(
+                        scores.get("summary_non_definitive"),
+                        thresholds.get("summary_non_definitive"),
+                    ),
+                    0.35,
+                ),
+                (
+                    _centered_norm(
+                        scores.get("summary_needs_followup"),
+                        thresholds.get("summary_needs_followup"),
+                    ),
+                    0.3,
+                ),
+                (
+                    _centered_norm(
+                        scores.get("entity_preservation"),
+                        thresholds.get("entity_preservation"),
+                    ),
+                    0.2,
+                ),
+            ]
+        )
+
     return {"x": x_value, "y": y_value}
 
 
@@ -799,8 +870,12 @@ def _build_cluster_points(
 
     points = []
     for cluster_id, coords_list in clusters.items():
-        x_values = [c.get("x") for c in coords_list if c.get("x") is not None]
-        y_values = [c.get("y") for c in coords_list if c.get("y") is not None]
+        x_values = [
+            value for value in (c.get("x") for c in coords_list) if isinstance(value, (int, float))
+        ]
+        y_values = [
+            value for value in (c.get("y") for c in coords_list) if isinstance(value, (int, float))
+        ]
         x_avg = mean(x_values) if x_values else None
         y_avg = mean(y_values) if y_values else None
         quadrant = _quadrant_label(x_avg, y_avg)

{evalvault-1.65.0.dist-info → evalvault-1.66.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: evalvault
-Version: 1.65.0
+Version: 1.66.0
 Summary: RAG evaluation system using Ragas with Phoenix/Langfuse tracing
 Project-URL: Homepage, https://github.com/ntts9990/EvalVault
 Project-URL: Documentation, https://github.com/ntts9990/EvalVault#readme
@@ -25,6 +25,7 @@ Classifier: Topic :: Software Development :: Quality Assurance
 Classifier: Topic :: Software Development :: Testing
 Classifier: Typing :: Typed
 Requires-Python: >=3.12
+Requires-Dist: chainlit>=2.9.5
 Requires-Dist: chardet
 Requires-Dist: fastapi>=0.128.0
 Requires-Dist: instructor
@@ -137,12 +138,17 @@ English version? See `README.en.md`.
 ## Quick Links
 
 - 문서 허브: `docs/INDEX.md`
+- CLI 실행 시나리오 가이드: `docs/guides/RAG_CLI_WORKFLOW_TEMPLATES.md`
 - 사용자 가이드: `docs/guides/USER_GUIDE.md`
 - 개발 가이드: `docs/guides/DEV_GUIDE.md`
 - 상태/로드맵: `docs/STATUS.md`, `docs/ROADMAP.md`
 - 개발 백서(설계/운영/품질 기준): `docs/new_whitepaper/INDEX.md`
 - Open RAG Trace: `docs/architecture/open-rag-trace-spec.md`
 
+### 다음 개선 작업 메모
+- 보험 요약 메트릭 확장 계획: `docs/guides/INSURANCE_SUMMARY_METRICS_PLAN.md`
+- Prompt 반복 적용 계획: `docs/guides/repeat_query.md`
+
 ---
 
 ## EvalVault가 해결하는 문제
@@ -470,6 +476,24 @@ npm run dev
 - Ragas 계열: `faithfulness`, `answer_relevancy`, `context_precision`, `context_recall`, `factual_correctness`, `semantic_similarity`
 - 커스텀 예시(도메인): `insurance_term_accuracy`
 
+### 요약 메트릭 설계 근거 (summary_score, summary_faithfulness, entity_preservation)
+
+### 커스텀 메트릭 스냅샷 (평가 방식/과정/결과 기록)
+- 평가 방식/입출력/규칙/구현 파일 해시를 `run.tracker_metadata.custom_metric_snapshot`에 기록합니다.
+- Excel `CustomMetrics` 시트와 Langfuse/Phoenix/MLflow artifact에도 함께 저장됩니다.
+
+- `summary_faithfulness`: 요약의 모든 주장이 컨텍스트에 근거하는지 평가합니다. 환각/왜곡 리스크를 직접적으로 측정합니다.
+- `summary_score`: 컨텍스트 대비 요약의 핵심 정보 보존/간결성 균형을 평가합니다. 정답 요약 단일 기준의 편향을 줄입니다.
+- `entity_preservation`: 금액·기간·조건·면책 등 보험 약관에서 중요한 엔티티가 요약에 유지되는지 측정합니다.
+
+**보험 도메인 특화 근거**
+- 보험 약관에서 치명적인 요소(면책, 자기부담, 한도, 조건 등)를 키워드로 직접 반영하고, 금액/기간/비율 같은 핵심 엔티티를 보존하도록 설계했습니다.
+- 범용 규칙(숫자/기간/금액)과 보험 특화 키워드를 함께 사용하므로, 현재 상태는 “보험 리스크 중심의 약한 도메인 특화”로 보는 것이 정확합니다.
+
+**해석 주의사항**
+- 세 메트릭 모두 `contexts` 품질에 크게 의존합니다. 컨텍스트가 부정확/과도하면 점수가 낮아질 수 있습니다.
+- `summary_score`는 키프레이즈 기반이므로, 표현이 달라지면 점수가 낮게 나올 수 있습니다.
+
 정확한 옵션/운영 레시피는 `docs/guides/USER_GUIDE.md`를 기준으로 최신화합니다.
 
 ---

{evalvault-1.65.0.dist-info → evalvault-1.66.0.dist-info}/RECORD

@@ -5,10 +5,11 @@ evalvault/mkdocs_helpers.py,sha256=1AKVQ1W2_VO4qclhfyefyU9Dz1Hzkh1DWDwsFMe24jc,3
 evalvault/adapters/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 evalvault/adapters/inbound/__init__.py,sha256=SG1svel1PwqetnqVpKFLSv612_WwGwLTbFpYgwk6FMw,166
 evalvault/adapters/inbound/api/__init__.py,sha256=LeVVttCA3tLKoHA2PO4z3y8VkfVcf3Bq8CZSzo91lf4,34
-evalvault/adapters/inbound/api/adapter.py,sha256=tYkJciUUFOK80QcSwzrqkXP1G4qUFItFV7uBYbjBGqU,68473
-evalvault/adapters/inbound/api/main.py,sha256=lRuyg3aBs5jIk7tq4p4d7jrRkFpV_brZypoOq8s56Rk,6896
+evalvault/adapters/inbound/api/adapter.py,sha256=HgWSYyUxvJPlaSG158WVzpPckpPCYV9Ec3CWN8rLFdI,69118
+evalvault/adapters/inbound/api/main.py,sha256=skYtmDngdOBryyLXQpNGlSd2Te6RF6GtfIwcMACPHFU,7068
 evalvault/adapters/inbound/api/routers/__init__.py,sha256=q07_YF9TnBl68bqcRCvhPU4-zRTyvmPoHVehwO6W7QM,19
 evalvault/adapters/inbound/api/routers/benchmark.py,sha256=yevntbZcNtMvbVODsITUBgR1Ka4pdFQrXBJJ4K4Jyr4,4477
+evalvault/adapters/inbound/api/routers/chat.py,sha256=3S6-ljiY1COlDuVDH5yzMJs9SO0EkuosRcJIYScHWvI,18143
 evalvault/adapters/inbound/api/routers/config.py,sha256=LygN0fVMr8NFtj5zuQXnVFhoafx56Txa98vpwtPa4Jc,4104
 evalvault/adapters/inbound/api/routers/domain.py,sha256=RsR7GIFMjccDN7vpG1uDyk9n1DnCTH18JDGAX7o4Qqc,3648
 evalvault/adapters/inbound/api/routers/knowledge.py,sha256=yb_e7OEPtwldOAzHTGiWe7jShHw2JdpOFnzGPMceRsg,7109
@@ -42,8 +43,8 @@ evalvault/adapters/inbound/cli/commands/pipeline.py,sha256=NeqWLzO9kRDuZd0pHAIHg
 evalvault/adapters/inbound/cli/commands/profile_difficulty.py,sha256=nOJH3iqgLAlXq4keLBj5oqpiRCg0jjGgT-7Q57HxEh8,6665
 evalvault/adapters/inbound/cli/commands/prompts.py,sha256=lddde5VbjYaqN_9gHPLNu6DWpg5fE-KqZzjN-XYwvJw,27153
 evalvault/adapters/inbound/cli/commands/regress.py,sha256=Dy8hUOdjapxOW9Hoov0DHHblkMaExiqWfYS14CaC9Kk,8806
-evalvault/adapters/inbound/cli/commands/run.py,sha256=X19rgXhajhvZNA4c0JMmzmPatTxhZgfapuW07bZL9xA,119265
-evalvault/adapters/inbound/cli/commands/run_helpers.py,sha256=hu2TioocitUZzGR7HUwZ6gOeEJSvt5tGNjwXOlo4Eic,40336
+evalvault/adapters/inbound/cli/commands/run.py,sha256=aKoZcQbOJ1KB_4zPk4L-AWw3u9vGWg3SaooR7A3Xd_Y,119910
+evalvault/adapters/inbound/cli/commands/run_helpers.py,sha256=93jFUg8QLrD38QU2JhOhFMoHDWUphSEKRdJ5KcUvrkQ,40806
 evalvault/adapters/inbound/cli/commands/stage.py,sha256=oRC9c5CysLX90Iy5Ba1pc_00DaOBS78lcBvzkbdrGRM,17123
 evalvault/adapters/inbound/cli/utils/__init__.py,sha256=QPNKneZS-Z-tTnYYxtgJXgcJWY6puUlRQcKrn7Mlv1M,685
 evalvault/adapters/inbound/cli/utils/analysis_io.py,sha256=RHkKEq4e-PtbtRDlXAJWU80RYHNPw-O5V9_GujdaGfc,13393
@@ -177,10 +178,10 @@ evalvault/adapters/outbound/nlp/korean/toolkit_factory.py,sha256=x3v-AAkVInOabC4
 evalvault/adapters/outbound/phoenix/sync_service.py,sha256=i6gHpNiZXKQ5yzV9B2TPb-P1N45k_Ck5ruzh3oqp4d8,9122
 evalvault/adapters/outbound/report/__init__.py,sha256=8OUduTHnWkBLHYrc7mBg45DnAwz0RgvSJmz1HqxVjLY,477
 evalvault/adapters/outbound/report/dashboard_generator.py,sha256=Dcu18NTK4lS8XNKnnnquagpZkd-4TSf5Mb2isFNW5Pk,7800
-evalvault/adapters/outbound/report/llm_report_generator.py,sha256=HUDA_IPBbl54cyEjTTJzdKTQ6H4IoZi-1VBdVmZf0uI,26593
+evalvault/adapters/outbound/report/llm_report_generator.py,sha256=i_iXfY8qutIb8TsvLKyMLnijsA0yiNJ3rBEFg4zVqcE,26858
 evalvault/adapters/outbound/report/markdown_adapter.py,sha256=5PS72h_qe4ZtYs-umhX5TqQL2k5SuDaCUc6rRw9AKRw,16761
 evalvault/adapters/outbound/storage/__init__.py,sha256=n5R6thAPTx1leSwv6od6nBWcLWFa-UYD6cOLzN89T8I,614
-evalvault/adapters/outbound/storage/base_sql.py,sha256=7jWtmNDBHncLDABf5ewwQJnfhFjySTfpfDJmEbPBD1w,40823
+evalvault/adapters/outbound/storage/base_sql.py,sha256=bNjJr941wqeLgv4E772JlOer1Q8OpJWxyotsNNn_R98,42536
 evalvault/adapters/outbound/storage/benchmark_storage_adapter.py,sha256=Qgf9xSSIkYQRpG4uLzcUdoYO9LTQDQ4tFRkkMYer-WA,9803
 evalvault/adapters/outbound/storage/postgres_adapter.py,sha256=HLaoQ3YJDFwOxeY0S92oPIqb-7EgWSasgt89RM86vr0,47148
 evalvault/adapters/outbound/storage/postgres_schema.sql,sha256=A9MfO0pjf4kjxoRj2KPI0Gg1cbX13I2YE3oieT-PGiI,8906
@@ -193,10 +194,10 @@ evalvault/adapters/outbound/tracer/open_rag_trace_decorators.py,sha256=LFnk-3FSL
 evalvault/adapters/outbound/tracer/open_rag_trace_helpers.py,sha256=D48Mbj-ioDKztjhV9513Q5DiUNiVdO60B_2sWMFEmnI,3520
 evalvault/adapters/outbound/tracer/phoenix_tracer_adapter.py,sha256=inmTAolAVsm0IrszE9VTJoI7HSvGGAnGNZVu_vZRAGg,741
 evalvault/adapters/outbound/tracker/__init__.py,sha256=Suu5BznOK5uTuD5_jS8JMZd8RPfQNlddLxHCBvMTm_4,358
-evalvault/adapters/outbound/tracker/langfuse_adapter.py,sha256=oi6bnz2qjDSXEea_bACFRUEKk-tKw53d1xMfL4FE5dI,18749
+evalvault/adapters/outbound/tracker/langfuse_adapter.py,sha256=uI-t5v9AC5VUMYsIc1FHYImourZeErGMXB0_prOMErc,18839
 evalvault/adapters/outbound/tracker/log_sanitizer.py,sha256=ilKTTSzsHslQYc-elnWu0Z3HKNNw1D1iI0_cCvYbo1M,2653
-evalvault/adapters/outbound/tracker/mlflow_adapter.py,sha256=m4xj3XBULFYg27U3twKrldLhbLyLNefezmb2pCpHJrw,7180
-evalvault/adapters/outbound/tracker/phoenix_adapter.py,sha256=VmGkVuBX5Nae6zTaaSbpcA94spFJGmwjW0gOh2vIxnk,25332
+evalvault/adapters/outbound/tracker/mlflow_adapter.py,sha256=6pSxbxSDZE7jN7uSMU6VFg0JlO7cBiMLYcd53NYpfcY,7350
+evalvault/adapters/outbound/tracker/phoenix_adapter.py,sha256=8p2qJeKn6OvIxNbD16h_QrhmCzKIBIf8_ej535MNn_A,26443
 evalvault/config/__init__.py,sha256=UCgeDx62M2gOuFvdN29wWwny2fdH4bPY_uUC3-42eDw,1297
 evalvault/config/agent_types.py,sha256=EP2Pv3ZtOzDXIvIa-Hnd1to9JIbMUtGitrlwzZtx0Ys,13418
 evalvault/config/domain_config.py,sha256=rOgNA2T8NWlDzcEFC0shdUCCww0lI1E5fUm5QrKQSZI,9264
@@ -205,7 +206,7 @@ evalvault/config/langfuse_support.py,sha256=DEzVMfMGGf1V45W_2oUG-NCDfsYI4UUdnYJI
 evalvault/config/model_config.py,sha256=KlzDbGyDLeOGE7ElekFFk5YjjT5u8i6KO2B4EyZkLnI,3542
 evalvault/config/phoenix_support.py,sha256=e6RPWd6Qb7KU6Q8pLaYTpJGWULtvEEU6B0xHWyVyOH0,13604
 evalvault/config/secret_manager.py,sha256=YjPMuNqeBrAR2BzCJvsBNUExaU4TBSFyZ8kVYZZifqA,4172
-evalvault/config/settings.py,sha256=JKJf8t20sOHYnHoCfTxqupQixNgfmWYJhChiGMNz-W0,17617
+evalvault/config/settings.py,sha256=xvoNma4CHAd8R_nF0DL4MUWXBWCR5M0C68NPSPLT5JQ,18285
 evalvault/config/playbooks/improvement_playbook.yaml,sha256=9F9WVVCydFfz6zUuGYzZ4PKdW1LLtcBKVF36T7xT764,26965
 evalvault/domain/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 evalvault/domain/entities/__init__.py,sha256=wszRJ1Imdc5NJ1bQPC2udk-mAgFdlw4uZV5IPNjLpHQ,3669
@@ -222,20 +223,24 @@ evalvault/domain/entities/judge_calibration.py,sha256=fhQEI7g2nZuG1OliikhxgefcFA
 evalvault/domain/entities/kg.py,sha256=8awN1M4vxAGQZk_ZG8i2CXKTizQ8FA1VCLhUWHZq0U8,3001
 evalvault/domain/entities/memory.py,sha256=bfS75q8K8_jNrB7IYh4mjP8Lkyj-I0TVsmHCP0FuICw,8423
 evalvault/domain/entities/method.py,sha256=a3jZi7SjcpK3HeVyVwQkUMwpnmg2RbxCnH4NqYPLCOI,1157
-evalvault/domain/entities/prompt.py,sha256=VzuUzqkqXv0FwTSNGyV5sSCft5sxTbG_Noq6Ymnke5o,2910
+evalvault/domain/entities/prompt.py,sha256=lQlRnHEKY69GWTC-cUIu0DMuPfJ9UWm6Sm4KTNjVwfY,2920
 evalvault/domain/entities/prompt_suggestion.py,sha256=Ep_XSjdYUj7pFSCMyeeZKs8yTnp74AVx05Zqr7829PE,1243
 evalvault/domain/entities/rag_trace.py,sha256=sZgnkG4fK6KOe3Np6TYAZ_tPnsRbOmucDSQns35U1n4,11868
 evalvault/domain/entities/result.py,sha256=OaGHMDLWMW2O4fNVuVTUvWFVBQ1iu93OD_oI3NumrCQ,10697
 evalvault/domain/entities/stage.py,sha256=UqS59sjoMs_bhMupNtvagbIx8QgHgFjWoRPhJ3uJP2s,7426
-evalvault/domain/metrics/__init__.py,sha256=fxjC5Z_8OuBIeMn80bYgnZZxpNoay2wH-qtG3NqCUvk,797
+evalvault/domain/metrics/__init__.py,sha256=Ros3CWg5in1xlEdMa0WUSG602SBVkxw2Zbro-XUlmxU,1214
 evalvault/domain/metrics/analysis_registry.py,sha256=JZpBrBs7-JExHKYuEML6Vg_uYLm-WniBE3BfiU5OtJg,7641
 evalvault/domain/metrics/confidence.py,sha256=AX4oeN28OvmMkwD0pT-jskkOlXh87C1pe2W9P1sF69g,17224
 evalvault/domain/metrics/contextual_relevancy.py,sha256=xAPYUv_0TM4j4WOutOSGftNln_l-2Ev6qpANeu4REv8,11057
 evalvault/domain/metrics/entity_preservation.py,sha256=uSCbaETceE5PbGn-230Rm8pryOA8jDkkeOwAkWxA65g,6500
 evalvault/domain/metrics/insurance.py,sha256=5NPeAi_86rpuZRgV4KhzomGrq3Uw2jjglN6FfA_AO8o,4040
 evalvault/domain/metrics/no_answer.py,sha256=x6vRyOa1jw-qsnw9kOYT8YMPdLElaDRu7zjNCpyJfqM,8237
-evalvault/domain/metrics/registry.py,sha256=QKjo4RNHxCqObGg36xJP3KAHqFpHM50Jy7GeSksdz0Y,5665
+evalvault/domain/metrics/registry.py,sha256=1CKPxSjdXK739zhzwodukGDL-dIhlJJH30cuP-czQWc,6926
 evalvault/domain/metrics/retrieval_rank.py,sha256=F55ByadJBowyKHKBmKAZ0T0qN_R1_7UNu-MiLnT4Ypg,14675
+evalvault/domain/metrics/summary_accuracy.py,sha256=Hr4QS1e4Rxt1MgcTj5rElKuPw9rWS-zGkI0d8wB5dwA,5988
+evalvault/domain/metrics/summary_needs_followup.py,sha256=5kExtZxxankP7csAAIZe_1uRFeBD7NQK-N15b5d0awM,1357
+evalvault/domain/metrics/summary_non_definitive.py,sha256=1EE-z0Ib66gpjc0MGZHmZJHJfpoACSIldgOwFkUNxg0,1029
+evalvault/domain/metrics/summary_risk_coverage.py,sha256=Fo-dMg_jU4MCr0YqOZzBZymwEbG9y2H6eLX-jmuS8IU,1777
 evalvault/domain/metrics/terms_dictionary.json,sha256=-ZQmpx6yMOYoAOpcLj-xK2LkAeCbAw0EUb6-syIOKS0,3801
 evalvault/domain/metrics/text_match.py,sha256=P-YTZs9ekDqEmxLNBP8eXnMRymPdC8V4dJPtwG2ajVM,10219
 evalvault/domain/services/__init__.py,sha256=X5Af1kf_vSt3S3mFwOV6OQdro-lFxwbVdNd7nJznkC8,1024
@@ -248,6 +253,7 @@ evalvault/domain/services/benchmark_runner.py,sha256=4tvQEDrfvp2fC2luUPuPBcRjEPL
 evalvault/domain/services/benchmark_service.py,sha256=TrmnvBMAPmcs0PewGZcn2rxHbviZ8KxmDvJCeyqm28I,6286
 evalvault/domain/services/cache_metrics.py,sha256=FKNZoxym30lc1SxTGmTn3Pr-PDNoAqgC9_d_IdF_jOQ,3463
 evalvault/domain/services/cluster_map_builder.py,sha256=qPKMPj-eSqECJSCOKvv3ZETgIwxwiKWbU3d6_feCoDg,6885
+evalvault/domain/services/custom_metric_snapshot.py,sha256=_MLOzBlHTRyTQ2NuunZ_lrLVF0__kvEcCUxXVVCeoRA,9684
 evalvault/domain/services/dataset_preprocessor.py,sha256=PnhLiPk0E9DIzjUr8N75296CCfl1AUXGv-lpaXBi0Ok,14797
 evalvault/domain/services/debug_report_service.py,sha256=SGdFh8tctAIq7RotFbg47eetxdYSS4Yju7-LOzpCMCM,4386
 evalvault/domain/services/difficulty_profile_reporter.py,sha256=uIj9-eiO2dDvQ6tP-DJBddfBq8VT63st0wtNC8Co4NQ,680
@@ -257,7 +263,7 @@ evalvault/domain/services/document_versioning.py,sha256=M1qZaMpQ2exVT1wkVAmvEPPu
 evalvault/domain/services/domain_learning_hook.py,sha256=rhKBmdnrJyfGzFNsNxzyv8jZO26-WOosHSmBV_9qdJg,7176
 evalvault/domain/services/embedding_overlay.py,sha256=ZTNxUPXpHGbQ3Uri5DD3feTUFn7qrhuNshhyCQEvRuM,3559
 evalvault/domain/services/entity_extractor.py,sha256=f3Rf5saK8QsgetLNK1Hbxzt8PtttJZCicSR63S8DJ5k,14141
-evalvault/domain/services/evaluator.py,sha256=cKcPRoqkUFc00lgx_blbIFl0qJyPIDjPeCciWAPm7m8,67669
+evalvault/domain/services/evaluator.py,sha256=Fvth2VdckDJvGuwxbXPnvPfQU59WZSJHV63H4qji4lM,78815
 evalvault/domain/services/experiment_comparator.py,sha256=IBrxIwux-8GucwlLx6e5lUqB9miSPvBLGJK9ctoW7Y0,3299
 evalvault/domain/services/experiment_manager.py,sha256=2k-qGiAUyZuqqmcp4P-M3Z9HTXwwcqW5HQYKNkcIHuI,4863
 evalvault/domain/services/experiment_reporter.py,sha256=QYlVmCFSx8hKTPMezc7QjJE07b3MSQ82Q4QVucSHLVY,1420
@@ -276,7 +282,7 @@ evalvault/domain/services/pipeline_orchestrator.py,sha256=yriVlEVZYDtt0Vwt4Ae6xy
 evalvault/domain/services/pipeline_template_registry.py,sha256=aWqXLQ24grpSZo9M4tZLRo1ysD10c6hUpW3JupZH9e0,28083
 evalvault/domain/services/prompt_candidate_service.py,sha256=Ibyb5EaWK28Ju2HnTqHHGOoiA9Q-VwY3hjxVODALwGY,3997
 evalvault/domain/services/prompt_manifest.py,sha256=5s5Kd6-_Dn-xrjjlU99CVo6njsPhvE50H5m_85U-H6U,5612
-evalvault/domain/services/prompt_registry.py,sha256=THcNs4jqp4FTLv9uO-VVvN6XGorkGrcIUwQH1dL74o8,4166
+evalvault/domain/services/prompt_registry.py,sha256=QyL4yIcKT93uv6L0-Q_iaNXno8QnsC19YcGekuSRMtE,5247
 evalvault/domain/services/prompt_scoring_service.py,sha256=SlvfuIbhj92RJu4RQAJ1BGKhKkOAUOt3cZNH21HtsX4,9833
 evalvault/domain/services/prompt_status.py,sha256=r1dFLGz4SfRxXaxsULQsr0-HpJkG9YfZ_yLIxF1MMBo,6731
 evalvault/domain/services/prompt_suggestion_reporter.py,sha256=Fc6sCPebUMk8SZVpjoJ6bCEun0ma-YmayEQnulBVv8s,10577
@@ -292,9 +298,9 @@ evalvault/domain/services/stage_metric_service.py,sha256=_u6ThZ8rGw8H9h3TNpu0j8X
 evalvault/domain/services/stage_summary_service.py,sha256=VVtuAr4vwzvmNFn8rqURJrhKFqAMG4CaBmyGiUk_xG0,1590
 evalvault/domain/services/synthetic_qa_generator.py,sha256=aiOTPoHZbKRTEeodABQ2I5lq8-Vs_kQtuzcGWd4MTGE,16526
 evalvault/domain/services/testset_generator.py,sha256=6IpiZ0pqhKEymo-AlUdfJjDkF2P1n8Md_QKV4nOheyg,4470
-evalvault/domain/services/threshold_profiles.py,sha256=YuOrD5CkXugAdSQYbMsFzS5VS1R201JOJtpKTs4dpXU,1296
+evalvault/domain/services/threshold_profiles.py,sha256=yYJ7o8SIRufI7kUN8edh8am-dVOq_TEhvDqlHe0WQUQ,1433
 evalvault/domain/services/unified_report_service.py,sha256=lG3VpMLC1MTYUlcGl-MUEE4PUopkyrhcgj4_ye9c_vM,11829
-evalvault/domain/services/visual_space_service.py,sha256=xG2jxKuRuqmQgbWsXOqmytKr6pQ7igQujNgdpb5gyB0,32569
+evalvault/domain/services/visual_space_service.py,sha256=3_qyBsThr5lzP1le6qkXf9ByX3JjoYGX15iMIHe8gQs,34958
 evalvault/ports/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 evalvault/ports/inbound/__init__.py,sha256=2Wsc0vNzH8_ZaErk4OHxP93hRonLUkMbn3W28DtTDO0,562
 evalvault/ports/inbound/analysis_pipeline_port.py,sha256=RJfKtp22AYEqnmRk6RDawAK52rEmyAhuk0FUPJQUwQU,1758
@@ -332,8 +338,8 @@ evalvault/reports/__init__.py,sha256=Bb1X4871msAN8I6PM6nKGED3psPwZt88hXZBAOdH06Y
 evalvault/reports/release_notes.py,sha256=pZj0PBFT-4F_Ty-Kv5P69BuoOnmTCn4kznDcORFJd0w,4011
 evalvault/scripts/__init__.py,sha256=NwEeIFQbkX4ml2R_PhtIoNtArDSX_suuoymgG_7Kwso,89
 evalvault/scripts/regression_runner.py,sha256=SxZori5BZ8jVQ057Mf5V5FPgIVDccrV5oRONmnhuk8w,8438
-evalvault-1.65.0.dist-info/METADATA,sha256=1f23dU3LtSQwrlalym1j8mH-vZaM7HWMzDh-ByjVOjo,24276
-evalvault-1.65.0.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-evalvault-1.65.0.dist-info/entry_points.txt,sha256=Oj9Xc5gYcyUYYNmQfWI8NYGw7nN-3M-h2ipHIMlVn6o,65
-evalvault-1.65.0.dist-info/licenses/LICENSE.md,sha256=3RNWY4jjtrQ_yYa-D-7I3XO12Ti7YzxsLV_dpykujvo,11358
-evalvault-1.65.0.dist-info/RECORD,,
+evalvault-1.66.0.dist-info/METADATA,sha256=f6jzeYkN1iuFwYJTcI8r5L52hVNZwACOlQuWYvVz_JY,26159
+evalvault-1.66.0.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+evalvault-1.66.0.dist-info/entry_points.txt,sha256=Oj9Xc5gYcyUYYNmQfWI8NYGw7nN-3M-h2ipHIMlVn6o,65
+evalvault-1.66.0.dist-info/licenses/LICENSE.md,sha256=3RNWY4jjtrQ_yYa-D-7I3XO12Ti7YzxsLV_dpykujvo,11358
+evalvault-1.66.0.dist-info/RECORD,,