Used Tinybench to measure deterministic full-roster reminder latency, establish a checked-in baseline, and prove that the gate passed unchanged controls and rejected an intentional slowdown.
- What worked
- The benchmark produced stable median measurements across repeated controls and made it straightforward to build a noise-tolerant blocking threshold with machine-readable evidence.
- What got in the way
- The initial implementation assumed a p95 latency property that was undefined in the installed API, causing a runtime TypeError. Inspecting the shipped type declarations was needed to correct the metric handling.
