Docs inventory + scoring pipeline: npm run audit:inventory

Part of the docs & demos tracking issue: #140. Phase 0 of the docs-quality program: make the audit repeatable instead of one-shot.

Problem

The docs corpus is 176 pages / ~308k words (2026-07-09 measurement) with large quality variance that currently requires manual archaeology to see:

• Critical pages that are far too thin: oss/getting-started/pro-quick-start.md is 207 words — the entry point for the paid tier
• Pages too long to navigate: oss/migrating/rsc-troubleshooting.md (8.7k words), oss/upgrading/changelog.md (35.8k words inside the docs tree)
• A stale cluster untouched since 2026-03-10 (angular-js-integration-migration, troubleshooting-when-using-webpacker, asset-pipeline at 92 words, rails-engine-integration with zero site references, project-structure)

Docs churn is high (dozens of pages changed in the first week of July alone), so any manual audit rots within weeks. We need npm run audit:inventory to regenerate the picture on demand.

Work

• scripts/inventory-docs.mjs following the audit-docs.mjs conventions (ESM, --docs-root/--output/--fail-on args, report to workspace root)
• Resolve the monorepo the same way sync-docs.mjs does (REACT_ON_RAILS_REPO env var, then sibling ../react_on_rails) and analyze <repo>/docs directly so git history is available for last-modified dates
• Per-page metrics: word count, heading/code-block counts, title, frontmatter description presence, last commit date (one batched git log --name-only pass), inbound links from other docs pages (resolved relative markdown links, not substring matches), sidebar presence (prototypes/docusaurus/sidebars.ts)
• Per-page flags: tooShort (<300 words, excluding index/release-note pages), tooLong (>4000 words), stale (>90 days untouched), orphan (no inbound links and not in sidebar), missingDescription
• Output: JSON dataset + markdown summary report (extremes tables, flag counts), mirroring the VALIDATION_REPORT pattern
• npm scripts: audit:inventory and test:inventory-docs (unit tests for the pure helpers via node --test)

Follow-ups (not this issue)

• Join with an importance signal (Cloudflare Web Analytics page views, or GSC if a property exists) to complete the Importance × Quality triage
• Feed tooShort/stale/orphan flags into audit:docs CI severities once thresholds are tuned

Acceptance criteria

• npm run audit:inventory runs against a sibling react_on_rails checkout and produces the JSON + markdown report; unit tests pass via node --test.

🤖 Generated with Claude Code

Start by reading scripts/audit-docs.mjs and scripts/sync-docs.mjs for the reporting and repository-resolution conventions, then inspect prototypes/docusaurus/sidebars.ts and the docs tree. Implement the inventory and pure-helper tests, and run npm run audit:inventory plus node --test to verify the JSON dataset, markdown summary, flags, and acceptance criteria.