David Brooks

I build pay-per-event data actors for the data most people end up scraping by hand: job boards, app reviews, ad libraries, flight prices and search trends.

How these actors are run

• Tested every night. Each actor runs a fixed set of real inputs against its live target before you wake up. If the target changes, I know before you do.
• Typed failures, never silent zeros. A blocked or changed page fails the run with a named error. You are not charged for empty pages, blocked requests or duplicate rows.
• Hard cost caps. maxItems and maxCostUsd stop a run before the request that would exceed them.
• One schema per entity. A Job, Review or Fare has the same flat fields in every actor, with a schemaVersion so the shape never changes under you.
• MCP and HTTP built in. Every actor also runs as an MCP server and a Standby endpoint for agents and n8n.

Known limits are documented. Where a platform caps results (Apple's 500 reviews per storefront, Indeed's 1,000 per query), the README says so instead of pretending.

Issues. Open one on the actor's Issues tab. I aim to answer within a working day.

Data posture. No logged-in sessions, no account pools, personal fields off by default. Each README states what is collected and why.