narrator-as-builtlisted

# narrator-as-built You are writing an as-built document for the subsystem changed by a deadreckon run. The document must help a user or maintainer understand the delivered application shape quickly. Return exactly one JSON object: ```json { "subject": "one line", "system_overview": "paragraph", "components": "markdown table", "topology": "fenced ASCII block or empty string", "file_layout": "fenced text tree/list", "external_interactions": ["bullet"], "cross_references": ["bullet"] } ``` ## Inputs - Goal: `{{ goal }}` - Run summary: `{{ run_summary }}` - Source layout seed: ```markdown {{ source_layout }} ``` - Diff samples: ```markdown {{ diff_samples }} ``` - Tool stdout/stderr: ```markdown {{ tool_stdout }} ``` ## Requirements - Components table columns must be exactly `Layer | Responsibilities | Key entrypoints`. - Every component row must include at least one `file:line` entrypoint. - The layer name `Project files` is forbidden. If the layer cannot be inferred, omit it. - `system_overview` must include two short sub-paragraphs beginning `What's load-bearing:` and `Where the seams are:`. - Include process/data topology ASCII when the seed has one; omit rather than emit a generic placeholder. - Include wire protocols, local files, subprocesses, or external services touched by the run when present in the evidence. - Link to `RUN-NARRATIVE.md`, `RUN-DECISIONS.md`, source AS-BUILT when available, traces, provenance, snapshots, and acceptance. - Exclu