[engine-failure] PROJ-448-quantifying-composition-dependent-grain- @ tasks: RuntimeError

Automated report (spec 023 / FR-016) — an unexpected engine failure occurred while processing **PROJ-448-quantifying-composition-dependent-grain-** at stage `tasks`.

```
RuntimeError: LLMReviewer[ordering]: review frontmatter has neither `verdict:` nor `concerns:`/`action_items:` — a contentless/truncated review MUST NOT count as acceptance (false-convergence guard). First 200 chars: '```yaml\n---\nreviewer_name: ordering\nreviewer_kind: llm\nstage: tasks\nartifact_path: projects/PROJ-448-quantifying-composition-dependent-grain-/specs/001-quantifying-composition-dependent-grain/tasks.md'
```

The project remains schedulable; it will retry on its normal cadence and recover once the underlying defect is fixed. Closing this issue does not require any project-state change.

Start with the engine's handling of the ordering review at stage `tasks`, using the reported artifact path `projects/PROJ-448-quantifying-composition-dependent-grain-/specs/001-quantifying-composition-dependent-grain/tasks.md`. Inspect how the review frontmatter is produced and validated, then retry the normal schedule. Done means a non-truncated review contains `verdict:` or `concerns:`/`action_items:` and no false acceptance occurs.

Written by the indexing model from the issue text.