Datum
A benchmarking harness that refuses to lie. Statistical rigour by default: confidence intervals, outlier detection, and a hard stop when the signal is noise.
Placeholder entry — replace with a real project.
Datum runs micro-benchmarks and reports them the way a lab would: with error bars, not vibes. It grew out of frustration with tools that print a single number to three decimal places and call it a result.
The survey
- Every measurement carries a 95% confidence interval.
- The runner detects a warm-up plateau instead of guessing an iteration count.
- Results export to a format a spreadsheet and a CI check can both read.
Field notes
Write the story here.
Return to the register.