Field Notes from a Timeout That Was Set by a Demo, Not a Design

A thirty-second timeout chosen to make a demo feel responsive survived into production, where it silently abandoned legitimate work every busy afternoon.

Nobody chose thirty seconds for production; someone chose it for a demo two years earlier, and production inherited it the way systems inherit everything: silently.

What the postmortem surfaced

• The timeout fired on eleven percent of legitimate peak-hour runs.
• Abandoned work retried into the same congestion, worsening it.
• No document recorded why the value was what it was.

Use the architecture guide, benchmarks, and examples to trace every timeout to a measurement instead of a memory.