[prathamtharwani]

Could you point us to any specific past instances? I'd be interested to read about them.

[jeffreyq]

https://blog.railway.com/p/incident-report-february-11-2026

[x0x0]

"we did not have the monitoring or controls to prevent our anti-fraud from hard killing 3% of workloads, including many instances of pg"

Oof.

[rzmmm]

Needs an anti-anti-fraud service which terminates malfunctioning anti-fraud services.

[x0x0]

When I've written similar services, there was a (low) hard cap on how many fraud decisions they could action before they quit and paged. If we were getting hit with a wave of something, a human had to temporarily bump that limit.