Executive summary
Rhetorical Mirror helps an author inspect how a draft is packaged before accepting an AI-assisted rewrite. Its taxonomy describes common persuasion frameworks, narrative devices, biases, fallacies, and pressure tactics. A review prompt asks for substance extraction, a substance check, and a plainer rewrite that preserves the author's voice. The repository includes self-evaluations of its current and initial README, keeping the earlier text available for comparison. It is a small personal editing framework rather than an automated classifier for judging other writers.
What was built
- A JSON taxonomy with seven layers and 41 named tags.
- A reusable prompt for analysis and low-rhetoric rewriting.
- Two worked README self-evaluations and the original draft.
- A workflow in which the author reviews detected patterns and accepts or
rejects the proposed changes.
Evidence and validation
The tag and layer counts are derived from taxonomy/taxonomy.json. The two example evaluations illustrate the format and its use on the repository's own prose. They are demonstrations, not a held-out benchmark or evidence of consistent scoring across models and reviewers.
Boundaries
A high rhetorical-density score does not imply falsehood, and a low score does not imply truth. The project is not an authorship detector, scientific benchmark, or honesty test. The author remains responsible for checking substance and deciding whether a proposed rewrite still sounds like them.
Repository map and first run
Read prompts/mirror-prompt.md, provide the full taxonomy/taxonomy.json and a draft to the reviewer, then inspect the findings and rewrite. Compare the examples/ evaluations with README.md and archive/INITIAL_README.md for a small worked example. There is no runtime application to install.