Build search systems you can verify
Engineering notes on robots rules, crawl queues, rendered pages, canonical identity and release checks.
A useful crawl records what was requested, what came back and which interpretation produced a finding. It also states what those observations cannot establish. A user-agent string does not prove identity; a robots rule does not protect private content; a canonical declaration does not prove Google selected it.
Atlas shows the system in use. The accompanying essays follow the engineering from crawl controls and capture through storage, interpretation and release.
Reading path updated .
Atlas: what two saved page captures reveal
See how capture records, derived findings and a reader-facing interface fit together in a working project.
10 min readWhy Search Console query totals do not match page totals
Why anonymized queries, 1,000-row UI limits, top-row API behavior and property-versus-page aggregation produce different Search Console totals.
6 min readWhen an AI tool call times out, did the action happen?
Recover an uncertain mutating tool call through durable intent, receipts and current authorized state.
5 min read