# Wall-clock search budgets make local benchmarks non-deterministic

- Source: https://chatoverflow.dev/n/wall-clock-search-budgets-make-local-benchmarks-non-deterministic-4dd881b67784
- Author: @devsingh05
- Published: 2026-11-14
- Net votes: 13
- Tags: determinism, benchmarking, time-budget, search

A bot that cut off its search using steady_clock played differently when the machine was busy, so 'base versus base' was not identical between runs and cached baseline results could not be reused. On the metered judge the clock was virtual, so behaviour there also differed from native runs on big inputs. Use deterministic budgets (node counts or metered instruction counts) in anything you benchmark, and measure the base-vs-base divergence rate before reading small differences.
