The §29 exit bar includes revocation p95 < 60 s across three regions for 14 days. This is how you run it for real. The instrument only ever measures — every number is the real wall-clock time from a revocation to when a root-dark verifier in a region first sees INVALID(revoked). Nothing is hardcoded; a missed SLO is recorded honestly and fails closed.

• A live registrar (or its public mirror) reachable from each region — the one whose revocations you're measuring.
• ≥3 vantage points: any three hosts you control in three regions (VMs, containers, a Pi in a closet — independence of location is what matters, not the provider). Each needs Node 18+ and outbound HTTPS to the registrar.
• The published directory.json + roots.json (so each vantage runs a real root-dark verifier), and a run CHALLENGE nonce you pick and keep (it binds this run's log + report; a collector pins it out of band).

On each host, from kits/soak/ (with @ainra/sdk installed):

node soak.mjs \
  --registrar https://registrar.example \
  --directory ./directory.json --roots ./roots.json --now $(date +%s) \
  --vantages <this-region>            # e.g. eu-west  (one label per process)
  --duration-sec $((14*24*3600)) --poll-ms 1000 --slo-p95-sec 60 \
  --challenge <RUN_NONCE> --out ./soak-out-<this-region>

Each process continuously issues + revokes real lineages and appends every measurement to an append-only, hash-chained log (soak-out-*/soak-log.jsonl), renders a live status page (soak-status.html) from that log, and writes a signed soak-report.json. Run one process per region so each report reflects what that region actually observed. (A single coordinator polling three regional endpoints also works; what matters is that each latency is measured from the region.)

soak-status.html is rendered from the measured log — open it, or serve the soak-out-*/ dir. The genesis board (make genesis-status) ingests the finished soak-report.jsons and shows p95 vs the 60 s target with elapsed-time honesty ("day 3 of 14").

Any third party checks a finished run:

# structural integrity (signature + tamper-evident chain + numbers recompute from the log):
make soak-verify OUT=./soak-out-eu-west
# to also GATE the SLO, pin the target + challenge YOURSELF (never read them from the report — D-024):
make soak-verify OUT=./soak-out-eu-west SLO=60 CHALLENGE=<the-nonce-you-pinned>

A tampered log or a re-signed PASS over a breaching log is rejected. A BREACH is recorded honestly and blocks the DoD declaration — that's the point.

Before committing to 14 days, confirm the whole mechanism end-to-end on one host:

make soak-smoke        # real registrar, ~12 issue/revoke cycles, 3 local vantage points, signed report, verify + tamper-reject

The smoke's vantages all poll one localhost registrar, so its numbers are real single-host latency (ms) — it proves the mechanism, not WAN propagation. The real run is the identical instrument with --duration-sec and regional --registrar URLs.