Developer fixture — not a canonical reference benchmark

This dataset was produced on a single developer machine (Apple M4 Max, 16 logical cores, 128 GB RAM, macOS 26.2) and has notbeen signed, peer-reproduced, or run on the project's reference hardware (R-Bench-1 / R-Bench-3). The comparison set is limited to barista variants + Apache Maven 3.9.9 only — mvnd and Maven 4 baselines are not yet included. Treat the numbers below as a data-pipeline preview, not a substantive performance claim.

This is a cold-cache dataset. Each iteration was routed at a fresh tempdir (BARISTA_PATHS__CACHE_DIR, BARISTA_PATHS__M2_REPOSITORY, and MAVEN_OPTS=-Dmaven.repo.local=…) so every iteration genuinely re-fetched the full dependency closure from Maven Central. Both barista (via mitmproxy reverse-proxy) and mvn (via a per-iteration settings.xml <proxies> block fed via --settings) are captured. Wall-clock numbers depend on Maven Central's response time and vary meaningfully across runs; only the Upstream calls + bytes comparison is robust on the same workload.

Preview · walkthrough-2026-05-18-cold

Walkthrough benchmark — walkthrough-2026-05-18-cold

Run ID: 2026-05-18T23:53:08Z-807650d4
Timestamp: 2026-05-18T23:53:08Z
Git SHA: 807650d477b6…
Runner: ajb-mbp.local

P03-pull-cold

Default command: barista pull --update

Baseline	Median	p95	Stddev	vs slowest	Upstream calls	Bytes
barista pull --update (cold cache) barista	21.24 s	21.24 s	± 0 ms	1.35×	438	20.18 MB
Apache Maven 3.9.9 — dependency:resolve (cold cache) mvn	28.69 s	28.69 s	± 0 ms	1.00×	514	37.89 MB

1measured iterations per baseline · wall-clock milliseconds · "vs slowest" column is median_slowest / median_this_baseline · "Upstream calls" counts distinct HTTP requests to the configured upstream (Maven Central by default) captured via mitmdump · — for baselines the harness did not capture (mvn / mvnd / forked-mvn paths bypass the env hook; their captures live in scripts/run-baseline-captures.sh).