test(kernel-deploy): red-path contract guards for the sf37/38 fixes #2069

Merged
pipeline-bot merged 1 commit from test/kernel-deploy-regression-guards into trunk 2026-08-18 23:28:28 +00:00
Owner

Closes the regression-test gap from the /btw review: the marker-file gate, the already-on-target no-op leg, staged-or-booted tolerances, the durable MANIFEST fallback, and the prestage verdicts each get a purpose-built guard, plus a throwaway-fixture contract proving 7 mutations go red with evidence. 132/132 guards green.

Closes the regression-test gap from the /btw review: the marker-file gate, the already-on-target no-op leg, staged-or-booted tolerances, the durable MANIFEST fallback, and the prestage verdicts each get a purpose-built guard, plus a throwaway-fixture contract proving 7 mutations go red with evidence. 132/132 guards green.
test(kernel-deploy): red-path contract guards for the sf37/38 fixes
All checks were successful
pipeline/ci CI green @ 4ddec739c4fc
4ddec739c4
The 2026-08-16..18 timed rolls hardened the kernel-deploy and
build-images templates five times, and every fix landed as inline shell
with no guard that would go red if it regressed. That is the class of
silent loss this repo has been burned by before. Three new guards close
the gap:

- check-build-images-marker-gate.sh — FUNCTIONAL: extracts the real
  CHANGED derivation from the prepare script and executes it under
  set -euo pipefail against fixture diffs. A marker-only diff must
  select nothing AND survive pipefail (kzwh4 + 98t5x together); a
  nested file that shares a marker name stays content. Non-vacuity
  floor on the extraction.
- check-kernel-roll-leg-tolerances.sh — pins the five load-bearing
  behaviors: the already-on-target no-op leg BEFORE the cordon, the
  kstage staged-or-booted verify, the kreboot staged-or-booted
  refusal, the durable in-cluster MANIFEST fallback with the frozen
  digest deciding AFTER it, and the prestage-verdict line. A missing
  anchor is a FAILURE, never a skip.
- check-kernel-deploy-regressions-contract.sh — proves both guards can
  fail: 7 mutations (each a historical or plausible regression) applied
  to throwaway fixture copies of the real templates, each asserted red
  WITH its evidence string; 2 green baselines.

132/132 guards pass with the three wired into tests/check.sh.
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
sean/pipelines!2069
No description provided.