可疑实体找「最接近的名字」时重建的目录（_scope_catalog）不收分析结论：候选只看表和字段，用不到结论的内容，以前每点开一个可疑的名字都把这次运行的分析结论读盘、复验一遍哈希（build_catalog 给 A 条目装整份结论）。查询、表和字段的编号不受影响。测试：台账里有分析结论时重建目录不读任何工件，按报告节点和退回不分范围两条路都是

backend/app/api/evidence.py

@@ -4048,18 +4048,22 @@ def _scope_catalog(sealed: _Sealed, report_node: str) -> dict[str, Any]:
 
     doc.catalog 只留了正文用到的实体，给「是不是想写…」找候选不够；运行状态（checkpoint）不在封存
     范围里，也不能用。图读不出来、找不到报告节点时退回全部封存台账。
+
+    台账里的分析结论不进：候选只看表和字段，用不到结论的内容，进了目录就要把每份结论读盘、复验一遍哈希（build_catalog
+    给 A 条目装整份结论）。查询、表和字段的编号不受影响（分析结论另编 A）。
     """
     from app.engine.nodes.report import report_catalog
     from app.engine.schema import GraphSpec
 
+    entries = [e for e in sealed.entries if e.get("kind") != LEDGER_KIND]
     try:
         spec = GraphSpec.model_validate(sealed.graph)
         node = spec.node_map().get(report_node)
     except Exception:  # noqa: BLE001 - 老运行的图可能已经不合现在的格式，退回不分范围
         node = None
     if node is None:
-        return build_catalog(nodes={}, ledger=sealed.entries)
-    return report_catalog({"evidence": sealed.entries, "nodes": {}, "input": {}}, spec, node)  # type: ignore[arg-type]
+        return build_catalog(nodes={}, ledger=entries)
+    return report_catalog({"evidence": entries, "nodes": {}, "input": {}}, spec, node)  # type: ignore[arg-type]
 
 
 def _suspicious_step(seg: dict[str, Any], cite: dict[str, Any], sealed: _Sealed, report_node: str) -> dict[str, Any]:

backend/tests/test_analysis_evidence_api.py

@@ -287,6 +287,25 @@ async def test_a_tampered_result_is_said_and_not_shown(client):
     assert seal["hash_ok"] is False and seal["sealed"] is False and body["seal"]["covered"] is False
 
 
+def test_suspicious_name_hints_do_not_read_the_analysis_results(monkeypatch):
+    """可疑实体找「最接近的名字」时按封存的台账重建目录（_scope_catalog）：分析结论不进，不读盘、不复验哈希（候选只看表和
+    字段，用不到结论的内容）。以前每点开一个可疑的名字，就把这次运行的分析结论全读一遍。"""
+    from app.api import evidence as api
+    from app.engine import evidence as engine_evidence
+
+    read: list[str] = []
+    monkeypatch.setattr(engine_evidence, "load_artifact", lambda artifact: read.append(artifact))
+    q, a = "1" * 64, "2" * 64
+    entries = [_query_entry(SNAPSHOT, q), {"kind": "analysis", "node_id": "an", "exec": 1, "artifact": a,
+                                           "op": "decompose", "mode": "additive", "item": "drop", "inputs": [q]}]
+    graph = {**GRAPH, "edges": [{"source": "fetch", "target": "an"}, {"source": "an", "target": "write"}]}
+    sealed = api._Sealed(run_id="r", graph=graph, seal={"sealed": True, "ok": True}, events=[], entries=entries)
+    for report in ("write", "nope"):
+        catalog = api._scope_catalog(sealed, report)
+        assert "Q1" in catalog and not [e for e in catalog.values() if e.get("kind") == "analysis"]
+    assert read == []
+
+
 async def test_agent_analyses_are_marked_as_such(client):
     made = await sealed_run(TEXT, item=None, call_id="call-9")
     body = (await client.get(f"/api/runs/{made['run']}/evidence/segments/{seg_id(made['doc'], 'a:A1')}")).json()