compass-kb-validation 2.6.0__tar.gz → 2.7.0__tar.gz

{compass_kb_validation-2.6.0 → compass_kb_validation-2.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compass-kb-validation
-Version: 2.6.0
+Version: 2.7.0
 Summary: Knowledge-base readiness validation for RAG & AI agents — config-driven, deterministic pipeline integrity checks.
 Author: COMPASS
 License: MIT
@@ -229,8 +229,14 @@ compass ask --project my-kb --batch questions.jsonl --format eval --output eval_
 compass quality   --project my-kb
 compass freshness --project my-kb --stale-after 30
 
-# Adjudicate an eval's grounding finding against the KB: did the bot's counts match reality?
-# Deterministic - attributes a failure to a layer (retrieval vs the model), never scores answer quality.
+# Validate an EXTERNAL eval's output against the KB COMPASS already has access to. Not an eval,
+# not a tracer: the eval (an LLM judge, or the bot's captured trace) claims "grounded",
+# "cited MIS p.1", "28%" - COMPASS deterministically checks those claims against ground truth
+# in the KB and attributes any failure to a layer (retrieval vs the model), never scoring answer
+# quality. Verifies each CITED claim at its cited location: a fabricated number, a wrong page, a
+# fact present in the KB but not retrieved, or a source not in the KB. Uses the eval's retrieved
+# chunks when present (Level 1) and the live KB (Level 2); the answer text + a reachable KB alone
+# still catches fabricated figures and phantom citations.
 compass grounding-audit --project my-kb --eval-result case.json --fail-on-fabrication
 ```
 

{compass_kb_validation-2.6.0 → compass_kb_validation-2.7.0}/README.md

@@ -162,8 +162,14 @@ compass ask --project my-kb --batch questions.jsonl --format eval --output eval_
 compass quality   --project my-kb
 compass freshness --project my-kb --stale-after 30
 
-# Adjudicate an eval's grounding finding against the KB: did the bot's counts match reality?
-# Deterministic - attributes a failure to a layer (retrieval vs the model), never scores answer quality.
+# Validate an EXTERNAL eval's output against the KB COMPASS already has access to. Not an eval,
+# not a tracer: the eval (an LLM judge, or the bot's captured trace) claims "grounded",
+# "cited MIS p.1", "28%" - COMPASS deterministically checks those claims against ground truth
+# in the KB and attributes any failure to a layer (retrieval vs the model), never scoring answer
+# quality. Verifies each CITED claim at its cited location: a fabricated number, a wrong page, a
+# fact present in the KB but not retrieved, or a source not in the KB. Uses the eval's retrieved
+# chunks when present (Level 1) and the live KB (Level 2); the answer text + a reachable KB alone
+# still catches fabricated figures and phantom citations.
 compass grounding-audit --project my-kb --eval-result case.json --fail-on-fabrication
 ```
 

{compass_kb_validation-2.6.0 → compass_kb_validation-2.7.0}/compass_kb_validation.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: compass-kb-validation
-Version: 2.6.0
+Version: 2.7.0
 Summary: Knowledge-base readiness validation for RAG & AI agents — config-driven, deterministic pipeline integrity checks.
 Author: COMPASS
 License: MIT
@@ -229,8 +229,14 @@ compass ask --project my-kb --batch questions.jsonl --format eval --output eval_
 compass quality   --project my-kb
 compass freshness --project my-kb --stale-after 30
 
-# Adjudicate an eval's grounding finding against the KB: did the bot's counts match reality?
-# Deterministic - attributes a failure to a layer (retrieval vs the model), never scores answer quality.
+# Validate an EXTERNAL eval's output against the KB COMPASS already has access to. Not an eval,
+# not a tracer: the eval (an LLM judge, or the bot's captured trace) claims "grounded",
+# "cited MIS p.1", "28%" - COMPASS deterministically checks those claims against ground truth
+# in the KB and attributes any failure to a layer (retrieval vs the model), never scoring answer
+# quality. Verifies each CITED claim at its cited location: a fabricated number, a wrong page, a
+# fact present in the KB but not retrieved, or a source not in the KB. Uses the eval's retrieved
+# chunks when present (Level 1) and the live KB (Level 2); the answer text + a reachable KB alone
+# still catches fabricated figures and phantom citations.
 compass grounding-audit --project my-kb --eval-result case.json --fail-on-fabrication
 ```
 

{compass_kb_validation-2.6.0 → compass_kb_validation-2.7.0}/compass_kb_validation.egg-info/SOURCES.txt

@@ -15,6 +15,7 @@ ingestion_validation/__main__.py
 ingestion_validation/ask.py
 ingestion_validation/baseline.py
 ingestion_validation/chunk_coherence.py
+ingestion_validation/citation.py
 ingestion_validation/cli.py
 ingestion_validation/config_lint.py
 ingestion_validation/content_inspect.py
@@ -67,6 +68,7 @@ tests/test_ask.py
 tests/test_baseline.py
 tests/test_benchmark.py
 tests/test_check_catalog.py
+tests/test_citation.py
 tests/test_cli.py
 tests/test_config.py
 tests/test_config_lint.py
@@ -101,6 +103,7 @@ tests/test_remediation.py
 tests/test_reporting.py
 tests/test_retrieval_assertions.py
 tests/test_retrieval_quality.py
+tests/test_review_hardening.py
 tests/test_run_cli.py
 tests/test_sources.py
 tests/test_storage_validator.py

{compass_kb_validation-2.6.0 → compass_kb_validation-2.7.0}/discover_and_validate.py

@@ -138,14 +138,14 @@ def _resolve_skip_steps(settings, *, include_graphdb: bool, include_retrieval: b
     (BUG-22). ``--keep s3_storage`` overrides the auto-skip.
     """
     skip = ["API Upload"]
-    # Graph validation runs by default when a graph block is actually configured (a
-    # sparql_query or an endpoint) - a configured neptune:/graphdb: block should not be
-    # silently ignored just because --include-graphdb was not passed. Opt out with
-    # --skip neptune_metadata (or graphdb_metadata). --include-graphdb still forces it on.
+    # Graph validation runs by default when THIS PROJECT declares a per-document graph
+    # query (sparql_query) - the signal that graph validation is actually intended. It must
+    # NOT key off `neptune.endpoint`, which resolves from a shared NEPTUNE_ENDPOINT env var:
+    # doing so turned graph on for every project whenever that env var was set, so a project
+    # with no graph block inherited another project's Neptune and failed all 19 graph checks
+    # (regression). Opt out with --skip neptune_metadata; --include-graphdb still forces on.
     nep, gdb = settings.neptune, settings.graphdb
-    graph_configured = bool(
-        getattr(nep, "sparql_query", "") or getattr(nep, "endpoint", "")
-        or getattr(gdb, "sparql_query", "") or getattr(gdb, "base_url", ""))
+    graph_configured = bool(getattr(nep, "sparql_query", "") or getattr(gdb, "sparql_query", ""))
     if not (include_graphdb or graph_configured):
         skip += ["GraphDB Metadata", "Neptune Metadata"]
     if not include_retrieval:
@@ -479,7 +479,11 @@ def main(argv=None) -> int:
     print(f"  Run ID: {run_id}")
     print("=" * 60)
 
-    documents = discover_documents(settings, folder_filter=args.folder, limit=args.limit)
+    try:
+        documents = discover_documents(settings, folder_filter=args.folder, limit=args.limit)
+    except Exception as exc:  # noqa: BLE001 - an unreachable/misconfigured source must not traceback
+        print(f"\n  Could not discover documents (is the storage/index reachable and configured?): {exc}")
+        return EXIT_CONFIG
     print(f"\n  Found {len(documents)} document(s)\n")
     # Corpus-level minimum-document gate (s3.min_documents). A no-arg --limit run that
     # finds fewer than expected is a real problem worth failing on.

{compass_kb_validation-2.6.0 → compass_kb_validation-2.7.0}/ingestion_validation/__init__.py

@@ -5,4 +5,4 @@ end-to-end: API upload -> object storage -> knowledge graph -> vector store ->
 retrieval quality, plus cross-store consistency.
 """
 
-__version__ = "2.6.0"
+__version__ = "2.7.0"