Add JSON Schema and CI validation for data.json

[ksallee]

Treats data.json as a spec — adds a JSON Schema, a validator, and a GitHub Actions workflow that regenerates data from doc_export.html and validates on every PR, so the parser and the committed artifact cannot silently drift.

While wiring this up the schema surfaced 43 empty-titled subsections in data.json caused by convert_doc.py appending a subsection for every <h2> element including styling-only ones. This PR includes a one-line skip for empty <h2>s and regenerates data.json; CI is green after that fix. The source Google Doc likely still contains those empty <h2>s — I don't have edit access there.

The schema is strict by default (additionalProperties: false at the top level and on data entries) so future field additions surface as CI failures and prompt a schema bump alongside the data change. Easy to relax to true if you'd rather have the schema document the shape without constraining it.

Files:

• data/schema.json — draft-2020-12 schema
• data/validate.py — local + CI validator (run via python data/validate.py)
• .github/workflows/validate.yml — round-trip + validate on push to main and on every PR
• data/convert_doc.py — skip empty <h2> elements
• data/data.json — regenerated (43 phantom subsections removed)
• requirements.txt — added jsonschema

[richardssam]

This is a great addition. Does the change you made to convert_doc fix all the cases?
What I'm wondering is whether we should be doing the validation as we convert the doc to data.json (using the convert_doc.py script).

[ksallee]

This is a great addition. Does the change you made to convert_doc fix all the cases? What I'm wondering is whether we should be doing the validation as we convert the doc to data.json (using the convert_doc.py script).

Thanks!

The empty <h2> skip cleared all 43 violations the schema caught against the current doc_export.html. It doesn't guarantee every future export is clean, but that's exactly what the schema is for: next time the doc changes shape, CI fails and we either tighten the parser or bump the schema deliberately.

Happy to move the validation call into convert_doc.py so the check runs as part of the conversion anyone regenerating locally gets the same guardrail without remembering a second command. I'd keep validate.py as a standalone entry point too, so it can be pointed at any data.json independent of regeneration. Want me to add that in this PR?