Build-Time Reduction Program

Instruments

vox ci build-bench — runs contracts/ci/build-bench-scenarios.v1.json, writes a snapshot, and --compares against contracts/ci/build-bench-baseline.v1.json to emit a per-scenario delta. Cumulative report: graphify-out/build-bench/REPORT.md.
vox ci dep-cycles — Tarjan SCC over cargo metadata. HARD-fails on normal-dependency cycles; inventories dev-dep back-edges to graphify-out/DEP_CYCLES.md. (Fills the gap that vox-arch-check does only pairwise layer-ordering with no cycle detection.)
scripts/crate-build-audit.vox — the dependency/blast-radius map (fan-in, LoC, self-time) → graphify-out/crate_audit.json + CRATE_BUILD_AUDIT.md.

Refreshing the baseline

After an intentional, accepted change to build cost, regenerate:

cargo check --workspace   # warm the cache first
vox ci build-bench --label baseline --write contracts/ci/build-bench-baseline.v1.json --repeat 3
git add contracts/ci/build-bench-baseline.v1.json
git commit -m "chore(ci): refresh build-bench baseline"

Adding or removing a scenario means editing build-bench-scenarios.v1.json and regenerating the baseline in the same PR.

Selective CI + soundness backstop

PR-time selective CI (PR #348) builds only affected crates on PRs; the merge-queue gate and nightly run the full --workspace. No build-time optimization here weakens that backstop.

Historical log

See docs/src/architecture/build-time-log.md for the measured phase-by-phase deltas.