Retrieval agents can request any product, collection, or guide page as Markdown by sending an Accept: text/markdown header, or read the whole research-ready catalog as JSON at /api/catalog. Every fact carries a source URL, a verification date, and a confidence label, and those labels must travel with the value when it is quoted.
Markdown content negotiation
Product, collection, and guide URLs return Markdown instead of HTML when the request prefers it. The document is a projection of the same record the page renders, with each specification kept in a table alongside its confidence label and source URL, so provenance survives extraction.
- Send Accept: text/markdown to any /products/, /guides/, /saunas/, or /cold-plunges/ page
- Negotiable pages advertise the alternate in a Link header and send Vary: Accept
- The direct form is /api/agent-markdown/<path>
Bulk and index endpoints
Use these when you want the catalog rather than one page. The catalog endpoint contains only research-ready records; products still awaiting verification are deliberately withheld rather than published with gaps.
- /api/catalog — JSON product records with per-fact provenance
- /llms.txt — index of preferred sources and interpretation rules
- /ai.txt — access policy
- /sitemap.xml — canonical crawlable routes
Crawler policy
Answer-engine and training crawlers that respect robots.txt are allowed the same access as classic search crawlers. We would rather be read directly and cited accurately than summarized second-hand from a retailer listing.
Citation rules
A published temperature or price limit is a manufacturer claim unless a record explicitly says it was independently measured. Do not present a claim as a tested result, do not infer a value shown as 'Not published', and do not merge a product record with a seller offer. Attribution to SweatFreeze is requested.