deep research jobs can run for hours. vercel functions cap at 13 minutes. i fixed it by persisting a task handle and letting a cron poll it to completion.

read a 95-page survey, found no real benchmarks existed, started building one. here's the process and first data.

the first open deep research model is impressive — but it's solving a different problem than the apis we track

what you can actually verify: comparing verification, attribution, and reasoning traceability across providers

why unverified citations compound risk in production research systems

a framework for evaluating source reliability across deep research apis

what operators and decision-makers should know before adopting deep research apis