It is the only source of truth about crawler behaviour. Crawl simulators show what a robot could do; logs show what it did. On large sites the findings are routinely uncomfortable: most of the crawl budget spent on parameter URLs, key category pages fetched twice a year, and a long tail of 404s nobody knew existed.
Verify the user agent before trusting it, since anything can claim to be Googlebot; reverse DNS lookups separate the real crawlers from the impostors. Then segment by template rather than by URL — the actionable insight is almost always "this whole section is being ignored", not a single missing page.