[CodeWithEugene]

Why

Production hit the Hobby fair-use limit for Fluid Active CPU (hack-village was ~53% of the team's ~3h50m over 30 days) with almost no traffic. instrumentation.ts starts pg-boss on every serverless boot. It registers about 10 workers at the default 2s poll, so each warm instance runs roughly 5 queries per second and burns CPU while idle.

What

• Job workers poll every 30s (was 2s).
• Cron workers poll every 60s.
• pg-boss cronMonitorIntervalSeconds and cronWorkerIntervalSeconds are 45s (the maximum) and maintenanceIntervalSeconds is 600s.

Impact

• About 15x fewer polling queries per warm instance.
• Jobs may start up to 30s late and cron sweeps up to 60s late. That is well inside the 1-hour payout KPI.
• Timing only: no change to payout, escrow or rollback logic.

Notes

• This touches the payout.* and escrow.* worker registrations (timing only), so the escrow/payout review rules apply.
• tsc --noEmit passes. Integration tests not run locally (need Postgres); CI will run them.
• If CPU is still high, the next step is moving the worker off Vercel.