Executive summary
Spec Mirror is a practical review aid for engineering prose. A JSON taxonomy describes wording patterns that make components, actions, conditions, and requirements easier to identify. Each tag carries detection signals, a rewrite rule, examples, and guidance on what should remain untouched. A Python helper generates a smaller view suitable for a project instruction file. Reviewers use the source requirements as a baseline, preserve established terminology, and report missing or conflicting facts for the author to resolve.
What was built
- Seven review layers containing 51 named tags.
- Preferred patterns and correction patterns with contextual examples.
- Per-passage assessment guidance and a structured finding format.
- Two recorded project corrections illustrating the review approach.
- A standard-library Python compact-view generator with optional rewrite rules
and JSON output, plus marked-block updates for instruction files.
Evidence and validation
The scope counts come from the committed JSON structure. The README documents 85,781 bytes for the full v0.3.1 taxonomy and 15,125 bytes for its default compact Markdown view. The insertion helper checks marker structure before writing and preserves surrounding content. These are inspectable artifact and behavior claims; the repository does not report a measured reviewer-accuracy benchmark.
Boundaries
This is a wording review tool, not a standards checker or design verification system. Clear prose does not prove correct behavior. A reviewer must not add actors, conditions, or requirements without support from the supplied material. Taxonomy scores remain editorial guidance, not calibrated engineering measures.
Repository map and first run
Generate a compact view without editing any instruction file:
python3 compact.pyUse spec_taxonomy.json with the draft, intended engineering field, and source requirements. The README supplies both review-only and rewrite prompts. Review each proposed change against the source before accepting it.