feat: online verification, node removals, and knowledge browser

Adds claim-bound online evidence records and citation verification scripts,
node removal transport/adapter, retrieval benchmark and owner-review tooling,
GraphCanvas/SourceDetail knowledge browser UI, and the ingest handoff helper.
Bumps the agents/gewu pack to the online-verification branch.

tests/test_semantic_review.py

@@ -0,0 +1,105 @@
+import json
+
+import pytest
+from scripts.run_semantic_review import build_packet, consensus, packet_batches
+
+
+def _judge(verdict: str, confidence: float = 0.9):
+    return {
+        "assessments": [{"claim_id": "claim-one", "verdict": verdict, "confidence": confidence}]
+    }
+
+
+def test_semantic_consensus_is_advisory_and_cannot_accept():
+    result = consensus(_judge("supported"), _judge("supported"), model_ids=("judge-a", "judge-b"))
+
+    assert result["assessments"][0]["outcome"] == "advisory_support_consensus"
+    assert result["owner_decisions_written"] == 0
+    assert result["canonical_records_written"] == 0
+
+
+def test_semantic_disagreement_and_low_confidence_defer():
+    disagreement = consensus(
+        _judge("supported"), _judge("partially_supported"), model_ids=("judge-a", "judge-b")
+    )
+    low = consensus(
+        _judge("supported", 0.79), _judge("supported"), model_ids=("judge-a", "judge-b")
+    )
+
+    assert (
+        disagreement["assessments"][0]["outcome"] == "defer_semantic_disagreement_or_low_confidence"
+    )
+    assert low["assessments"][0]["outcome"] == "defer_semantic_disagreement_or_low_confidence"
+
+
+def test_semantic_consensus_requires_identical_claim_sets():
+    with pytest.raises(ValueError, match="different claim ID sets"):
+        consensus(
+            _judge("supported"),
+            {"assessments": [{"claim_id": "claim-two", "verdict": "supported", "confidence": 0.9}]},
+            model_ids=("judge-a", "judge-b"),
+        )
+
+
+def test_semantic_packet_keeps_evidence_and_policy_status_without_authority():
+    items = [
+        {
+            "source_stage": "review",
+            "evaluation": {"status": "needs-review"},
+            "proposal": {
+                "proposal_id": "proposal-1",
+                "revision": 2,
+                "operations": [
+                    {
+                        "node": {
+                            "node_id": "node-1",
+                            "title": "Finding",
+                            "claims": [
+                                {
+                                    "claim_id": "claim-one",
+                                    "text": "The claim.",
+                                    "evidence": [{"quote": "The evidence."}],
+                                }
+                            ],
+                        }
+                    }
+                ],
+            },
+        }
+    ]
+
+    packet = build_packet(items, input_sha256="f" * 64)
+
+    assert packet["claims"][0]["policy_status"] == "needs-review"
+    assert packet["claims"][0]["evidence"][0]["quote"] == "The evidence."
+    assert any("untrusted data" in line for line in packet["review_instructions"])
+
+
+def test_semantic_batches_stay_under_cli_argument_limit():
+    items = [
+        {
+            "proposal": {
+                "proposal_id": f"proposal-{index}",
+                "operations": [
+                    {
+                        "node": {
+                            "claims": [
+                                {
+                                    "claim_id": f"claim-{index}",
+                                    "text": "A" * 10_000,
+                                    "evidence": [{"quote": "B" * 10_000}],
+                                }
+                            ]
+                        }
+                    }
+                ],
+            }
+        }
+        for index in range(10)
+    ]
+    packet = build_packet(items, input_sha256="a" * 64)
+
+    batches = packet_batches(packet)
+
+    assert len(batches) > 1
+    assert all(len(json.dumps(batch, ensure_ascii=False).encode()) < 131_000 for batch in batches)