METHOD / PROOFGATE BETA
How the review works.
Transparent, source-based screening—not an independent certificate of truth.
Four stages
1. Separate atomic claims. Cloudflare-hosted Qwen or an explicitly configured OpenRouter model proposes individual testable statements and editorial notes from your pasted draft. Coverage is imperfect. The system separately checks that original numbers remain visible in extracted claims.
2. Compare to your sources. Qwen examines each claim against user-supplied evidence and returns one of: source-aligned, contradicted, insufficient.
3. Enforce quotation validity. A deterministic check requires cited passages to occur verbatim in the supplied text. Unsupported numbers, missing quotations and missing evidence prevent source alignment.
4. Require human interpretation. Even a source-aligned statement might be false if the source itself is unreliable, outdated or out of context. Important work needs human review.
Current limits
This beta does not independently search the web, authenticate source documents, parse scanned PDFs or Word files, validate every mathematical relationship, or guarantee complete claim extraction. It accepts short pasted drafts and evidence or local text-file imports.
Data handling
Submitted text is sent to a Cloudflare Worker and processed by Cloudflare Workers AI or, if the operator configured OpenRouter, an eligible OpenRouter provider selected for this request. The default provider selection prioritizes low price and requires no provider data collection and zero data retention. This beta application does not deliberately store submitted content in a database, but infrastructure-level processing may still occur. Please do not submit confidential, medical, account, password or other sensitive information. No app account is required; usage limits apply.
For concerns or feedback, email hello@kin-etiq.ai.