A stratified sample of 30 public URLs (10 reference/docs pages, 10 e-commerce/listing/review sites, 10 sites known to run anti-bot protection or heavy client-side rendering) scraped once each via Crawlora's own POST /web/scrape endpoint — formats=["markdown","metadata"], max_age=0 (cache bypassed, forces a fresh fetch), render="auto" (HTTP first, escalates to a headless browser when blocked or JS-rendered), 45-second wall-clock timeout per request. A row counts as a success only if the response was HTTP 200 and returned more than 200 characters of markdown.

Scope — what this is not

This is Crawlora's own measured reliability on a single disclosed run — not a head-to-head benchmark against Firecrawl, Scrapfly, or any other named competitor. We don't hold API keys for competitor products and won't publish a number for a tool we haven't actually run.

Tier labels (easy / medium / hard) were assigned before the run, by reputation, and left unchanged afterward — several sites labeled “medium” (Reddit, Indeed, Yelp) turned out harder in practice than several labeled “hard” (Nike, Walmart, Instagram). That mismatch is disclosed, not smoothed over.