fix(explore): 手动量化成绩可绑定后续决策

examples/control_plane/todo-user-gate-scope-smoke.py

@@ -179,59 +179,25 @@ def read_todo(registry: Path, todo_id: str) -> dict:
     return view["todo"]
 
 
-def assert_legacy_scope_diagnostics() -> None:
-    # Historical source validation remains readable before import. It cannot
-    # serve as a normal writer or alter an already canonical authority.
-    with tempfile.TemporaryDirectory(prefix="loopx-legacy-gate-diagnostics-") as tmp:
+def assert_unpromoted_source_is_not_contract_authority() -> None:
+    # Detailed metadata diagnostics use canonical rows; the File/SQLite cases
+    # in test_canonical_status_todos cover them without a display file.
+    with tempfile.TemporaryDirectory(prefix="loopx-unpromoted-gate-diagnostics-") as tmp:
         root = Path(tmp)
         repo, state, registry = write_fixture(root, import_authority=False)
         insert_before_agent_section(
             state,
             "- [ ] Approve an intentionally unscoped gate.\n"
-            "  <!-- loopx:todo todo_id=todo_unscoped_gate status=open task_class=user_gate -->\n"
-            "- [ ] Track an unbound non-blocking user action.\n"
-            "  <!-- loopx:todo todo_id=todo_unbound_action status=open task_class=user_action -->\n"
-            "- [ ] Resolve a mismatched gate binding.\n"
-            f"  <!-- loopx:todo todo_id=todo_mismatched_gate_binding status=open "
-            f"task_class=user_gate blocks_agent={SIDE_AGENT} bound_agent={PRIMARY_AGENT} -->\n"
-            "- [ ] Track an intentionally untyped user todo.\n"
-            "  <!-- loopx:todo todo_id=todo_untyped_user status=open -->\n"
-            "- [ ] Approve a gate with an unregistered blocked agent.\n"
-            "  <!-- loopx:todo todo_id=todo_bad_agent status=open task_class=user_gate blocks_agent=codex-unknown -->\n"
-            "- [ ] Approve a gate with conflicting scope.\n"
-            f"  <!-- loopx:todo todo_id=todo_both_scopes status=open task_class=user_gate blocks_agent={SIDE_AGENT} global_gate=true -->",
-        )
-        state.write_text(
-            state.read_text(encoding="utf-8")
-            + "\n- [ ] Repair a removed continuation value.\n"
-            + "  <!-- loopx:todo todo_id=todo_removed_continuation status=open task_class=advancement_task action_kind=review continuation_policy=review_handoff -->\n"
-            + "- [ ] Repair removed agent gate routing.\n"
-            + f"  <!-- loopx:todo todo_id=todo_removed_agent_gate status=open task_class=advancement_task blocks_agent={SIDE_AGENT} -->\n"
-            + "- [ ] Repair an unknown executor exclusion.\n"
-            + "  <!-- loopx:todo todo_id=todo_unknown_exclusion status=open task_class=advancement_task excluded_agents=codex-unknown -->\n"
-            + "- [ ] Repair malformed executor exclusion metadata.\n"
-            + "  <!-- loopx:todo todo_id=todo_malformed_exclusion status=open task_class=advancement_task excluded_agents=%%% -->\n",
-            encoding="utf-8",
+            "  <!-- loopx:todo todo_id=todo_unscoped_gate status=open task_class=user_gate -->",
         )
         original = state.read_bytes()
         check = check_contract(
             registry_path=registry, runtime_root_override=str(root / "runtime"),
             scan_roots=[repo], limit=1,
         )
         assert check["ok"] is False, check
-        for todo_id, diagnostic in (
-            ("todo_unscoped_gate", "requires bound_agent"),
-            ("todo_unbound_action", "requires bound_agent"),
-            ("todo_mismatched_gate_binding", "must bind to blocks_agent"),
-            ("todo_untyped_user", "requires task_class"),
-            ("todo_bad_agent", "not registered"),
-            ("todo_both_scopes", "cannot set both"),
-            ("todo_removed_continuation", "removed continuation_policy"),
-            ("todo_removed_agent_gate", "removed blocks_agent routing"),
-            ("todo_unknown_exclusion", "unregistered agents"),
-            ("todo_malformed_exclusion", "malformed excluded_agents"),
-        ):
-            assert any(todo_id in error and diagnostic in error for error in check["errors"]), check
+        assert any(row["code"] == "canonical_authority_required"
+                   and row.get("goal_id") == GOAL_ID for row in check["error_diagnostics"]), check
         assert state.read_bytes() == original
 
 
@@ -673,7 +639,7 @@ def main() -> int:
         assert state.read_bytes() != original
         state.write_bytes(original)
 
-    assert_legacy_scope_diagnostics()
+    assert_unpromoted_source_is_not_contract_authority()
     assert_imported_scope_repairs()
 
     print("todo-user-gate-scope-smoke ok")

examples/explore-configure-goal-smoke.py

@@ -11,6 +11,9 @@
 
 
 REPO_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(REPO_ROOT))
+from loopx.control_plane.testing.canary_harness import import_fixture_authority  # noqa: E402
+
 GOAL_ID = "explore-configure-fixture"
 
 
@@ -86,7 +89,9 @@ def write_fixture(root: Path) -> tuple[Path, Path]:
         + "\n",
         encoding="utf-8",
     )
-    return registry, root / "runtime"
+    runtime_root = root / "runtime"
+    import_fixture_authority(registry, GOAL_ID)
+    return registry, runtime_root
 
 
 def goal(registry: Path) -> dict:

examples/explore-worker-plan-gate-smoke.py

@@ -37,6 +37,7 @@
 REPO_ROOT = Path(__file__).resolve().parents[1]
 if str(REPO_ROOT) not in sys.path:
     sys.path.insert(0, str(REPO_ROOT))
+from loopx.control_plane.testing.canary_harness import import_fixture_authority  # noqa: E402
 
 from loopx.capabilities.explore.todo_branch_plan import (  # noqa: E402
     MAX_BRANCH_WIDTH,
@@ -387,6 +388,8 @@ def check_cli_gate_states() -> None:
             + "\n",
             encoding="utf-8",
         )
+        for entry in goal_entries:
+            import_fixture_authority(registry, entry["id"])
 
         def run_cli(*extra_args: str, fmt: str = "json", check: bool = True):
             result = subprocess.run(

loopx/capabilities/explore/README.md

@@ -74,6 +74,13 @@ identify the current retained candidate, establish causality, satisfy acceptance
 or authorize execution. Actual adoption and outcome benefit require subsequent
 work and evidence; automatic branch rankings remain unchanged.
 
+Manually captured measured findings also appear in the existing result pages
+and, when linked to eligible work, its exact-source comparison cards. They need
+no `writeback-result` origin tag. Read the page before optionally binding a
+`selected_candidate`; normal source-revision, Todo-link and followthrough checks
+still apply. Unmeasured manual notes retain their existing summary/audit path.
+Reading or selecting a measured result does not certify adoption or execution.
+
 ## At a Glance
 
 Explore Harness is a supported, default-off optional capability for

loopx/capabilities/explore/README.zh-CN.md

@@ -60,6 +60,11 @@
 重复尝试。best 不代表当前保留版本、因果收益、验收通过或执行授权；是否实际采用
 并产生收益，仍需后续工作和证据确认。自动分支排序保持原有行为。
 
+手动登记的量化 finding 同样进入既有结果分页；关联可执行工作时，也进入其精确来源
+比较卡片，无需补写 `writeback-result` 来源标签。先读取结果页，再按需绑定
+`selected_candidate`，仍核对来源修订、Todo 关联和后续实际工作。未带测量的手动笔记
+保持既有摘要/审计路径。读取或选择某条成绩不认证实际采用或执行。
+
 ## 概览
 
 LoopX Explore 是面向长程探索目标（软件研究、安全攻击面测绘、领域调研）的受支持、默认关闭的可选能力。它把“到处看看”变成有界、可观测、有门控的流程，包含三根支柱：

loopx/control_plane/capabilities/explore_turn_context.ts

@@ -112,16 +112,20 @@ export function projectExploreTurnContext(params: JsonObject): JsonObject {
     const audit = requireJsonObject(branch.typed_evidence_audit, "Explore evidence audit");
     return requireStringArray(audit.requested_node_refs ?? [], "Explore node refs");
   }) : []);
-  const attachedResults = findings.filter(row => Array.isArray(row.tags) && row.tags.includes("writeback-result"));
+  // Manual quantitative capture uses the same finding owner. Keep its sources
+  // available to the existing exact-result decision path, without requiring a
+  // writeback-origin tag or admitting unrelated unmeasured notes.
+  const decisionResults = findings.filter(row => row.measurements !== undefined
+    || Array.isArray(row.tags) && row.tags.includes("writeback-result"));
   // Keep the same detail budget, but do not let unrelated newer results evict
   // the applicability of evidence linked to the next selected work item.
   // Explicitly bound work can differ from this independent planner. Prioritize
   // its canonical links without changing candidates, scores or admission.
-  const plannerResults = prioritizeLinkedResults(attachedResults, selectedRefs);
+  const plannerResults = prioritizeLinkedResults(decisionResults, selectedRefs);
   // Shared links must not inherit another task's planner priority. Use source
   // recency within the focus and retain planner ordering only outside it.
   const writebackResults = focusedRefs.size
-    ? prioritizeLinkedResults(attachedResults, focusedRefs, plannerResults) : plannerResults;
+    ? prioritizeLinkedResults(decisionResults, focusedRefs, plannerResults) : plannerResults;
   const command = (...args: string[]) => [...route, "explore", ...args, "--goal-id", goal];
   const resultLimit = requireInteger(params.result_limit ?? 3, "result_limit");
   const resultOffset = requireInteger(params.result_offset ?? 0, "result_offset");

tests/control_plane/test_stride_comparison_followthrough_cli.py

@@ -150,7 +150,8 @@ def review(value, turn_id=turn, *, expected=0):
 
 
 @pytest.mark.parametrize("provider", ["file", "sqlite"])
-def test_wide_result_page_replays_every_linked_outcome_at_writeback(tmp_path, monkeypatch, provider):
+@pytest.mark.parametrize("tagged", [False, True], ids=["manual-measurement", "writeback-result"])
+def test_wide_result_page_replays_every_linked_outcome_at_writeback(tmp_path, monkeypatch, provider, tagged):
     isolate_sqlite_runtime(tmp_path, monkeypatch)
     monkeypatch.setenv("LOOPX_USAGE_PING", "0")
     monkeypatch.setenv("PYTHONPATH", str(fixture.REPO_ROOT))
@@ -175,15 +176,16 @@ def cli(*args, expected=0):
     actor.probe(request, "retained")
     for index in range(4):
         measurements = project / f"measurement-{index}.json"
-        measurements.write_text(json.dumps([{"metric": "quality", "value": index,
+        measurements.write_text(json.dumps([{"metric": "quality", "value": 3 - index,
             "unit": "points", "higher_is_better": True,
             "comparison_scope": "validation:owned-v1", "input_revision": "owned-input-v1",
             "evaluation_ref": "evaluator:owned-v1", "candidate_ref": f"candidate:scope-{index}",
             "evidence_ref": f"measured:scope-{index}"}]))
         cli("explore", "finding", *identity, "--finding-id", f"wide-result-{index}",
             "--node", "comparison-retained", "--title", f"Observed input scope {index}",
             "--summary", f"Only the recorded input scope {index}; other inputs remain unknown.",
-            "--status", "confirmed", "--tag", "writeback-result", "--evidence-ref", f"measured:scope-{index}",
+            "--status", "confirmed", *(("--tag", "writeback-result") if tagged else ()),
+            "--evidence-ref", f"measured:scope-{index}",
             "--measurements-json", str(measurements))
     turn = "wide-page-review"
     cli("quota", "should-run", *identity, "--codex-app", "--turn-instance-id", turn)
@@ -196,17 +198,20 @@ def cli(*args, expected=0):
     # Choose a quantified outcome omitted by the default three-row view.
     visible = {row["finding_id"]: row for row in page["graph"]["writeback_results"]}
     default_refs = {row["finding_id"] for row in control["explore_basis"]["comparison_candidates"][0]["outcomes"]}
-    selected = {"todo_id": fixture.TODO_ID, **next(row for row in reversed(card["outcomes"])
-        if row["finding_id"] not in default_refs and visible[row["finding_id"]].get("measurements"))}
-    measurement = visible[selected["finding_id"]]["measurements"][0]
-    assert measurement["evidence_ref"] == visible[selected["finding_id"]]["evidence_refs"][0]
     [feedback] = page["graph"]["measurement_feedback"]
     assert feedback["measured_count"] == 4
-    assert feedback["best"]["candidate_ref"] == "candidate:scope-3"
+    assert feedback["best"]["candidate_ref"] == "candidate:scope-0"
+    assert feedback["best"]["value"] == 3 and feedback["latest"]["measurement"]["value"] == 0
+    selected = {"todo_id": fixture.TODO_ID, **next(row for row in card["outcomes"]
+        if row["finding_id"] == feedback["best"]["finding_id"])}
+    assert selected["finding_id"] not in default_refs, "an older best must remain selectable beyond the first page"
+    measurement = visible[selected["finding_id"]]["measurements"][0]
+    assert measurement["evidence_ref"] == visible[selected["finding_id"]]["evidence_refs"][0]
     decision = deepcopy(control["minimal_example"])
     decision["explore_review"] = {"revision": control["explore_basis"]["revision"],
         "disposition": "used", "considered_refs": [selected["finding_id"]],
-        "selected_candidate": selected, "reason": "This exact measured scope informs the next experiment.",
+        "selected_candidate": selected,
+        "reason": "Revisit the same-scope best quality of 3 before repeating the latest candidate's score of 0.",
         "result_pages": [{k: page["graph"]["result_page"][k] for k in ("revision", "offset", "limit", "todo_id")}]}
     packet = actor.vision(request, selected["finding_id"], outcome="replan")
     packet.update(stride_decision=decision, todo_delta=["retain:" + fixture.TODO_ID])

tests/control_plane_ts/explore_turn_context.test.ts

@@ -82,11 +82,11 @@ for (const includeDigests of [false, true]) {
   });
 }
 
-for (const limit of [3, 4, 20]) {
-  test(`every scoped outcome on an explicitly requested page remains selectable: ${limit}`, () => {
+for (const tagged of [false, true]) for (const limit of [3, 4, 20]) {
+  test(`every scoped measured outcome remains selectable: tagged=${tagged}, limit=${limit}`, () => {
     const findings = Array.from({length: 20}, (_, index) => ({finding_id: `scoped-${index}`,
       node_id: "question", finding: `Measured input family ${index}`, status: "confirmed",
-      summary: `Applicable to input family ${index} only.`, tags: ["writeback-result"],
+      summary: `Applicable to input family ${index} only.`, tags: tagged ? ["writeback-result"] : [],
       evidence_refs: [`result:scoped-${index}`],
       measurements: [{metric: "quality", value: index, unit: "points", higher_is_better: true,
         comparison_scope: "validation:owned-v1", input_revision: "owned-input-v1",
@@ -99,6 +99,7 @@ for (const limit of [3, 4, 20]) {
         typed_evidence_audit: {requested_node_refs: ["question"]}}]}};
     const source = projectExploreTurnContext(request);
     const graph = source.graph as JsonObject, harness = source.harness as JsonObject;
+    assert.ok(harness.comparison_candidates?.length, "measured findings must reach the existing decision cards without an origin tag");
     const card = harness.comparison_candidates[0];
     // Quantitative feedback still uses all sources, independently of page size.
     assert.equal(graph.measurement_feedback[0].measured_count, 20);