PatrickNDevelopment

Work / Project · Published · 2026-10-19

ScriptAudit

Four deterministic audits for video scripts, with suggestions accepted one at a time.

ScriptAudit spoken-readability findings with a side-by-side preview of a suggested change
ScriptAudit spoken-readability findings with a side-by-side preview of a suggested change. Screenshot of the implemented application with its synthetic sample data.

The problem

A script draft needs checking for unsupported claims, loose narrative threads, hard-to-say sentences and packaging issues, with every change reviewable and reversible.

What it does

  • Grounding, narrative, spoken-readability and packaging audits with the reason for each finding.
  • Suggestions accepted or rejected one at a time with a side-by-side diff; every change is a revision.
  • Optional local backend for an opt-in AI explanation broker with a usage ledger.

Engineering choices

  • All audits run in the browser as transparent rules; the optional FastAPI backend only brokers AI explanations.
  • AI rewrites are filtered so they cannot add numbers, superlatives, attributions or links.

Tested results

  • 64 Vitest and 31 pytest tests; 13 Playwright tests.
  • Independent acceptance review on 2026-10-18: 9 defects fixed with regression tests.
  • Published 2026-10-19: CI green on the first run on Linux, GitHub Pages deployed, live check passed.

Last recorded verification 2026-10-19; latest recorded commit bb81fcd. The full commands and outcomes are in the repository's docs/test-report.md.

Stack

  • TypeScript
  • React
  • Vite
  • zod
  • Python
  • FastAPI
  • SQLite
  • Vitest
  • pytest
  • Playwright

Candid limitations

  • Rule-based and English-only; it does not verify facts or predict performance.
  • The AI broker was tested only with a fake provider.

How it was made

Directed and reviewed by the owner; specification, code, tests and documentation written with Claude Code (AI). A separate Claude Code session that did not write the code then re-ran every check and fixed what it found; that is an AI review, not a human code audit. See the repository's AI_ASSISTANCE.md.