Local Verification

Use Zig 0.17.0 and run from a Git checkout:

tools/verify.sh fmt    # formatting and whitespace only
tools/verify.sh quick  # build, debug tests, tidy, downstream consumer
tools/verify.sh        # full native optimization and external validation matrix

Set ZIG=/path/to/zig to select a toolchain. The full command runs debug, safe, and fast tests plus all six pinned external validations and the campaign-to-regression validation, then compiles the disabled Win64 fiber path and an SVE AArch64 Linux build (zig build check-sve) and checks release symbols. The ordinary test step includes tidy, the package consumer, supported host fiber checks, and the bounded queue, network KV, and storage compatibility validations. External validations may fetch their pinned lazy dependencies on the first run. Ochi validation is unavailable until its upstream build and dependencies support Zig 0.17; zig build validate-ochi reports this explicitly. Its harness is retained in validation/ochi_store.zig and syntax-ported to Zig 0.17, but it is not compiled until Ochi builds again (previous pin: f8b2e9c804571e4d2203658589123e974feac4c2). Negative-run tests intentionally print failure summaries; check the build summary and exit status to distinguish those from failing tests.

Formatting checks the .zig and .zon files Git tracks or would track, so ignored dependencies and build caches are skipped. It works in a dirty checkout, and CI runs the same command.

The native matrix only validates the current host. Linux and macOS CI jobs are still release gates; a green local run is not evidence that another OS passed. The release checklist also requires Pages and clean installation.

tests/compatibility_072.zig compares lifecycle and reduction traces, tapes, capsules, and manifests byte for byte with fixtures captured from v0.7.1. Keep pinned-build identity rejection tests enabled even when these bytes match.

zig build validate-campaigns runs, kills, resumes, replays, and reduces campaigns on pinned xitdb and redis.zig in separate processes; see Campaigns. Its evidence lands in zig-out/campaigns unless -Dcampaign-dir says otherwise.

For repeatable measurements before performance work, see Performance Baseline.