Run npm run dev and open http://localhost:5177/script-audit/. For the AI part, also run the backend
with the deterministic fake provider: in backend, set SCRIPT_AUDIT_AI_PROVIDER=fake and run
..\.venv\Scripts\python -m script_audit.
- Editor: choose
examples/draft.mdunder "Markdown or text file". The title becomes "The Shocking Story of the Paper Bridge"; the outline shows three sections with words and durations. - Audits: open each tab and press its Run button: grounding, narrative, spoken, packaging. Each finding names its rule; "Why this rule?" explains it.
- Packaging: reject the "Sensational phrasing" title suggestion. It moves to "Rejected" and can be restored.
- Spoken readability: on "Double space", press Preview change (side-by-side diff), then Apply. Revision 2 is created; the rejection is still there.
- History → Compare revision 1 with the current one: the double space on the left, one space on the right.
- Editor → Download Markdown:
the-shocking-story-of-the-paper-bridge-r2.md, title unchanged.
- Editor → Load the sample script (invented). Audits → Grounding: run it. Coverage is 1 of 10.
- In the Editor, add
[c:worst-storm]before the full stop of "The worst storm took 3 metres of cliff in a single night." and save. Back in Grounding, that warning is gone, coverage is 2 of 10, and "She was the first keeper ..." in the same paragraph is still flagged. - Tag "She was the first keeper" with
[c:keeper-log]: the warning becomes "Not in the linked claim: first", because that claim is about 212 storms, not about being first. - On Claims, add a source with a citation and quote to
storm-record; its sentence is covered.
- Approve the script in the Editor: it becomes read-only and every Apply button is disabled.
- History → Undo last change: the new revision has the exact earlier text.
- With the fake backend: Audits → AI explanations → opt in, tick findings, Ask. Preview and apply an AI rewrite, or see one dropped when it would add a number.