This material is a specification and reference set with a reference rule engine. It is not an editing application, a rendering pipeline or a completed production, and the worked example is entirely fictional — it exists to demonstrate the method.

Endpoints

The machine-readable description lives at /v1/openapi.json (OpenAPI 3.1). Agent onboarding: /llms.txt.

Run the rule engine

Submit a context object; rule conditions read dotted paths from it. The engine evaluates all 196 rules in priority order — hard before soft — applies the actions of every match, and returns the accumulated state plus a trace.

curl -s https://video.bowtiekreative.com/v1/evaluate \
  -H 'content-type: application/json' \
  -d '{
    "context": {
      "transcript": {"unresolved_pronoun": true},
      "story": {"has_opening_promise": true}
    }
  }'

The response includes state (flags, proposals, requirements, human reviews, block status), hard_matched and soft_matched rule ids, and a trace with each match's priority, rationale and actions. Add "domains": ["transcript"] to evaluate one domain, or "matched_only": false for the full audit trace over every rule.

Validate a document

Name one of the 12 packaged schemas and submit a document:

curl -s https://video.bowtiekreative.com/v1/validate \
  -H 'content-type: application/json' \
  -d '{"schema": "clip", "document": {"clip_id": "clip_001"}}'

Validation is structural: required fields, types, enumerations, patterns, ranges and cross-file references. It does not run the rule engine or judge editorial quality.

Caching & terms

• GET responses are cacheable for five minutes with a day of stale-while-revalidate.
• POST endpoints are never cached.
• The corpus is versioned; counts and content only change with a new release.