Perf calibration
Question: does hammering a fixture for 10–60s change ranking vs a short autocannon run?
Measured on existing nuxt-i18n-next fixture builds (plain-nuxt / i18n / i18n-micro), same machine, sequential loads:
| Fixture | autocannon 5s RPS | autocannon 10s RPS | Δ | Artillery 2s+8s RPS |
|---|---|---|---|---|
| plain-nuxt | 64.0 | 65.2 | 1.9% | 68 |
| i18n | 98.8 | 97.1 | 1.7% | 100 |
| i18n-micro | 173.8 | 187.4 | 7.3% | 112 |
Rank micro > i18n > plain holds for 5s and 10s.
micro/i18n ratio: 1.76 (5s) vs 1.93 (10s) — relative gap change ≈ 9% (under the 15% budget). Absolute RPS for a single fixture still jitters; use the same duration every time and compare ratios/ranks, not one-off absolutes against an old report.
Published long Artillery (6s+60s) historically showed the same rank with a larger micro/i18n gap (~1.69 Artillery RPS). Short Artillery was noisier on absolute micro RPS in this pass — so the single default is autocannon 10 connections × 5 seconds, not a separate “quick” profile.
Reproduce:
pnpm exec tsx scripts/perf-calibrate.tsArtifact: .untestutils/perf-calibrate/calibration.json.