# Copy Claims Checker 1.0.0 Check a batch of drafts against supplied source facts and explicit English editorial rules. Receive item-level findings and hashes of the exact input copy and facts. Includes local browser UI and CLI. Node 20+, no npm installation or network service. ## Run Extract the whole ZIP and run `node app.mjs`. Open the complete local URL printed in the terminal, load the sample or your JSON, run the check, review findings, and download the full report or CSV. Ctrl+C stops the local server. It uses a random session token and loopback only; do not expose it as a public API. CLI: `node cli.mjs examples/input.json new-report.json`, or append `csv` for a findings CSV. Output files must be new. Exit 0 means a report was written, even when it contains errors. Exit 2 means invalid input or I/O failure. For automation inspect `summary.rulesPass`, `summary.errors` and `summary.reviewNotes`. ## Input and rules See examples/input.json. Supply 1–100 items, each with a unique `id`, nonempty `copy`, flat array of nonempty `facts`, and `kind`: text, html, title or description. Maximum input 5 MiB; each copy 20000 characters, up to 200 facts per item of 10000 characters each. Optional forbiddenPhrases and requiresEvidencePhrases extend the default rules with literal phrases, not regular expressions (100 phrases, 100 characters each). Numbers absent from source facts are findings. Full numeric tokens are compared: 5 is not supported by 15 or 5.9. An occurrence alone does not establish correct units, association, negation or factual support. .25 and 0.25 are treated alike. Prices, offers, guarantees and hype are flagged by the default editorial policy even when present in the facts. Listed claims such as handmade or certified are flagged when absent and retained as review notes when present in the facts. HTML profile permits p, ul and li without attributes, and checks basic markup defects. It is not an HTML sanitizer; never execute supplied copy. Title bounds 10–70 and description bounds 100–170 are this tool's SEO editorial defaults, not universal marketplace limits. Placeholder text, damaged characters and configured publication structure issues are reported. Plain text mode does not apply metadata length rules. The full report contains rule details and SHA-256 receipts; CSV gives one row per finding and protects formula-like cell prefixes. All input items, including those without findings, appear in JSON. No source text is rewritten and nothing is approved or published. This tool neither verifies facts externally nor certifies truth or legal compliance. English lexical rules can miss an invented adjective, negation or an unsupported relationship between numbers and claims. Review the source and results. ## Verify Run `node --test tests/product.test.mjs`. The included tests cover batch results, numeric boundaries, invalid input, literal phrases, CLI output and actual loopback HTTP processing. No external connections are made.