The content model for product copy.

Most UI copy is fine. The stakes are concentrated in a handful of moments: an error message that tells the user what went wrong and what to do next; a destructive confirmation that names the thing being destroyed; a permissions button that asks for access instead of declaring submission; an empty state that points somewhere useful.

Engineers and PMs without content-design training can't hold all that context in their heads while they're shipping features. ContentRX holds it for them. It knows that the same sentence reads differently in destructive_action than in browsing_discovery, and applies the standards that match the moment it detected.

A senior content designer looks at an error message and sees whether it owns the failure or blames the user. That's not a rule you can look up in a style guide — it's pattern recognition built from years of practice. ContentRX encodes that pattern recognition. The 47 standards are the visible surface of it; the moments carry the situational weights; the evaluation chain publishes its own accuracy with confidence intervals and no composite score.

Those tools check that your sentence is grammatical and inclusive. Excellent at what they do. ContentRX checks a different thing: that your error message shouldn't be an error message at all; that your destructive confirmation names what will be destroyed; that your permissions button says Request access and not Submit.

You can run both. The layers don't compete — they stack.

Stripe Radar is a model. Stripe sells the model — the rules engineers write on top are secondary; the learned patterns from every transaction Stripe has ever seen are the moat.

ContentRX is a model. We sell the model. The taxonomy is browsable and claimable; every source that shaped it is named with its license and opt-out path; calibration is reported honestly with 95% CIs and no composite headline. The rules you can disable per team; the commitments you can't.

Content-standards enforcement is moving upstream. Engineers write product copy in the IDE now; PRs carry more strings than design files; LLMs draft the first pass before a content designer sees it. ContentRX meets that reality by leading with the surfaces where copy is actually written today, with the Figma plugin alongside for design-time checks.

• MCP server — Claude Code, Cursor, and any MCP client call evaluate_copy, classify_moment, and the standards catalog directly. Inline content review during generation, not after. Install.
• CLI — contentrx "Click here" or --batch strings.txt. --explain prints the full rationale chain. Stdlib-only install. Install.
• GitHub Action — evaluates strings touched in a pull request. fail-on: review gates the merge on review-recommended verdicts. Drops into any repo with a YAML snippet. Install.
• Figma plugin — design-time check. Scan a frame, per-string verdicts with moment banners and rationale chains. Three-button stance per finding (Agree / Disagree / Ship anyway). Best for the copy that comes in through design, not code. Install.

ContentRX is the content model that Robo — a senior content designer — would run on their own work. The moments, the weights, the 47 standards: all carry one designer's judgment calls, attributed and published. Read the about-the-model page for the longer story.