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_prepare_answer_batches.py

@@ -0,0 +1,97 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+from scripts.prepare_answer_batches import (
+    build_batches,
+    load_tuning_questions,
+    prepare,
+    validate_answer_output,
+)
+
+
+def test_batch_builder_excludes_reserved_and_caps_batch_size(tmp_path: Path) -> None:
+    path = tmp_path / "questions.json"
+    path.write_text(
+        json.dumps(
+            [
+                {"id": "Q01", "partition": "tuning", "question": "First?"},
+                {"id": "Q99", "partition": "reserved", "question": "Secret?"},
+                {"id": "Q02", "partition": "tuning", "question": "Second?"},
+            ]
+        ),
+        encoding="utf-8",
+    )
+    questions = load_tuning_questions(path)
+    batches = build_batches(questions, 2)
+    assert [[row["id"] for row in batch] for batch in batches] == [["Q01", "Q02"]]
+    with pytest.raises(ValueError, match="between 1 and 4"):
+        build_batches(questions, 5)
+
+
+def test_prepared_batches_include_limits_but_no_reserved_content(tmp_path: Path) -> None:
+    source = tmp_path / "questions.json"
+    source.write_text(
+        json.dumps(
+            [
+                {"id": f"Q{i:02d}", "partition": "tuning", "question": f"Question {i}?"}
+                for i in range(1, 7)
+            ]
+            + [{"id": "Q99", "partition": "reserved", "question": "Reserved answer clue"}]
+        ),
+        encoding="utf-8",
+    )
+    output = tmp_path / "out"
+    result = prepare(
+        type("Args", (), {"questions": source, "output": output, "batch_size": 4})()
+    )
+    assert result["question_count"] == 6
+    assert result["batch_count"] == 2
+    prompts = "\n".join(path.read_text(encoding="utf-8") for path in output.glob("*-prompt.md"))
+    assert "Never repeat an identical search" in prompts
+    assert "40000" in prompts
+    assert "Reserved answer clue" not in prompts
+    assert all((path.stat().st_mode & 0o077) == 0 for path in output.iterdir())
+
+
+def test_answer_validator_requires_exact_ids_and_complete_citations(tmp_path: Path) -> None:
+    expected = tmp_path / "expected.json"
+    answers = tmp_path / "answers.json"
+    expected.write_text(json.dumps([{"id": "Q01"}, {"id": "Q02"}]), encoding="utf-8")
+    answers.write_text(
+        json.dumps(
+            {
+                "answers": [
+                    {"id": "Q01", "answer": "A", "citations": [], "evidence_gap": False},
+                    {
+                        "id": "Q02",
+                        "answer": "B",
+                        "citations": [
+                            {
+                                "node_id": "n",
+                                "claim_id": "c",
+                                "source_revision_id": "r",
+                                "extraction_id": "e",
+                                "locator": {"kind": "text_span"},
+                                "quote": "exact quote",
+                            }
+                        ],
+                        "evidence_gap": False,
+                    },
+                ]
+            }
+        ),
+        encoding="utf-8",
+    )
+    report = validate_answer_output(expected, answers)
+    assert report["status"] == "valid"
+
+    malformed = json.loads(answers.read_text(encoding="utf-8"))
+    malformed["answers"].pop()
+    answers.write_text(json.dumps(malformed), encoding="utf-8")
+    report = validate_answer_output(expected, answers)
+    assert report["status"] == "invalid"
+    assert any("exactly match" in issue for issue in report["issues"])
+