fix(queue): report actual executed model in completion evidence

project/TICKETS.md

@@ -439,4 +439,6 @@ project analysis or human-owned ticket inputs.
 | **ticket-439** | [`README.md`](./ticket-439/README.md) | - | - | - | - | - |
 | **ticket-440** | [`README.md`](./ticket-440/README.md) | - | - | - | - | - |
 | **ticket-441** | [`README.md`](./ticket-441/README.md) | - | - | - | - | - |
+| **ticket-442** | [`README.md`](./ticket-442/README.md) | - | - | - | - | - |
+| **ticket-443** | [`README.md`](./ticket-443/README.md) | - | - | - | - | - |
 <!-- AUTO:TICKET_INDEX:END -->

project/ticket-445/README.md

@@ -0,0 +1,23 @@
+# Ticket 445: Report actual SubLLM model in queue execution evidence
+
+- **ID**: ticket-445
+- **Owner**: agent:codex
+- **Status**: IN_PROGRESS
+- **Workflow state**: PUBLICATION
+- **Created**: 2026-10-08
+
+## Goal and scope
+
+SESSION_EXECUTION_AUTHORIZATION: the user requested autonomous service and bug fixing, verification of Willman/Koru/SubLLM/OpenCode and protected delivery. Intake: paxlet-com/willman PLF-113, bounded follow-up PLF-115. A real local GPT-OSS response was labeled cursor/grok-4.6 by the queue summary. Correct the label from the runner result without changing execution or admission.
+
+## Acceptance criteria
+
+- [x] AC-01: The executed result model wins over the requested/default model; legacy runners retain fallback labels and exit codes.
+- [x] AC-02: Targeted regression and managed governance checks pass.
+- [ ] AC-03: Exact-head OneDev and independent Validator approve protected publication.
+
+## Evidence
+
+External receipt set: receipt:minis-autonomy-20261008/koru-provenance. Source scope: src/koru/queue/runner.py and tests/test_queue_model_provenance.py. No fleet writer is enabled by this change.
+
+Targeted regression: 45 passed. Managed governance: PASS, 0 warnings. Source/test lint: PASS. Protected exact-head CI and independent merge approval remain required.

project/ticket-445/intent.json

@@ -0,0 +1,91 @@
+{
+  "schema": "new-project.intent/v3",
+  "ticket": "ticket-445",
+  "summary": "Report actual SubLLM model in queue execution evidence",
+  "workstream": "application",
+  "classification": {
+    "kind": "BUG",
+    "priority": "P2",
+    "origin": "requested"
+  },
+  "allowedPaths": [
+    "project/ticket-445/**",
+    "TODO.md",
+    "project/TICKETS.md",
+    "src/koru/queue/runner.py",
+    "tests/test_queue_model_provenance.py"
+  ],
+  "forbiddenPaths": [
+    "project/ticket-*/user-*.md"
+  ],
+  "stacks": [
+    "python",
+    "docker"
+  ],
+  "dependsOn": [],
+  "conflictsWith": [],
+  "integrationTicket": null,
+  "delivery": {
+    "acceptedBaseSha": "552ed81026c9213d05029c9ce395d2e01fa0ca7a",
+    "targetBranch": "main",
+    "outcome": "Queue completion and run-log labels report the actual SubLLM result model, including a local model chosen instead of the requested/default cloud model.",
+    "nonGoals": [
+      "No provider route or permission changes",
+      "No automatic fleet execution or issue closure"
+    ],
+    "complexity": "S",
+    "estimatedMinutes": 20,
+    "budgets": {
+      "maxImplementationFiles": 2,
+      "maxAffectedComponents": 1,
+      "maxPublicInterfaceChanges": 0,
+      "maxRuntimeDependencies": 0
+    },
+    "architecture": {
+      "status": "accepted",
+      "decision": "Prefer optional result.model from the executed runner when building the queue action label; retain requested/default label for legacy CommandResult implementations without model metadata.",
+      "components": [
+        {
+          "name": "queue-execution-evidence",
+          "paths": [
+            "src/koru/queue/runner.py",
+            "tests/test_queue_model_provenance.py"
+          ]
+        }
+      ],
+      "responsibilityChanges": false,
+      "interfaceChanges": [],
+      "dataChanges": [],
+      "ui": {
+        "impact": "none",
+        "states": [],
+        "evidence": []
+      },
+      "rollback": "Revert the source/test delta; no stored state migration."
+    },
+    "runtimeDependencies": [],
+    "validation": [
+      {
+        "criterion": "AC-01",
+        "commands": [
+          "python -m pytest -q tests/test_queue_model_provenance.py tests/test_queue_runners.py tests/test_queue_evidence.py"
+        ],
+        "evidence": "Executed model wins; requested/default fallback and failure propagation retain behavior."
+      },
+      {
+        "criterion": "AC-02",
+        "commands": [
+          "./project/governance-check.sh"
+        ],
+        "evidence": "Bounded source ownership and delivery contract validated."
+      },
+      {
+        "criterion": "AC-03",
+        "commands": [
+          "protected OneDev exact-head checks and independent Validator"
+        ],
+        "evidence": "Reviewed exact-head PR merged by protected controller."
+      }
+    ]
+  }
+}

src/koru/queue/runner.py

@@ -299,7 +299,8 @@ def _execute_action(
     elif executor_kind == "llm":
         action = _enrich_llm_request_with_context(action, project)
         result = llm_runner(action, project)
-        action_label = f"llm {action.get('model') or _DEFAULT_LLM_MODEL}"
+        runtime_model = getattr(result, "model", None)
+        action_label = f"llm {runtime_model or action.get('model') or _DEFAULT_LLM_MODEL}"
     elif executor_kind in ("taskand", "process"):
         result = taskand_runner(action, project)
         action_label = f"taskand {action.get('uri') or 'orchestrator'}"

tests/test_queue_model_provenance.py

@@ -0,0 +1,47 @@
+from types import SimpleNamespace
+
+import pytest
+
+from koru.queue.runner import _execute_action
+from koru.queue.runners import _DEFAULT_LLM_MODEL
+
+
+@pytest.mark.parametrize(
+    ("runtime_model", "requested_model", "expected_model"),
+    [
+        ("local-development/gpt-oss:20b", "cursor/grok-4.6", "local-development/gpt-oss:20b"),
+        ("local-development/gemma4:12b", None, "local-development/gemma4:12b"),
+        ("", "requested/model", "requested/model"),
+        (None, "requested/model", "requested/model"),
+        (None, None, _DEFAULT_LLM_MODEL),
+    ],
+)
+@pytest.mark.parametrize("exit_code", [0, 7])
+def test_queue_label_uses_executed_model(
+    monkeypatch, tmp_path, runtime_model, requested_model, expected_model, exit_code
+):
+    monkeypatch.setattr("koru.queue.runner._enrich_llm_request_with_context", lambda action, _: action)
+    result = SimpleNamespace(returncode=exit_code, stdout="answer", stderr="diagnostic")
+    if runtime_model is not None:
+        result.model = runtime_model
+    request = {"prompt": "fixture"}
+    if requested_model is not None:
+        request["model"] = requested_model
+
+    def run_llm(action, project):
+        assert action == request
+        assert project == tmp_path
+        return result
+
+    def unexpected_runner(*args):
+        pytest.fail("another executor was selected")
+
+    actual, label = _execute_action(
+        "llm", request, tmp_path, "PLF-001", unexpected_runner, run_llm, unexpected_runner
+    )
+
+    assert actual is result
+    assert actual.returncode == exit_code
+    assert actual.stdout == "answer"
+    assert actual.stderr == "diagnostic"
+    assert label == f"llm {expected_model}"