Skip to content

ci(live-provers): drop continue-on-error on Tier-2/3 live-test steps#118

Merged
hyperpolymath merged 1 commit into
mainfrom
ci/drop-soft-fail-on-live-prover-tests
May 28, 2026
Merged

ci(live-provers): drop continue-on-error on Tier-2/3 live-test steps#118
hyperpolymath merged 1 commit into
mainfrom
ci/drop-soft-fail-on-live-prover-tests

Conversation

@hyperpolymath

Copy link
Copy Markdown
Owner

Summary

The two "Run live test for ${{ matrix.backend }} (allow-fail while Wave-X wires up)" steps in .github/workflows/live-provers.yml carried continue-on-error: true. The comments explicitly disclosed the soft-fail as transitional, but the gate silently masks real prover regressions: a successful Tier-2 / Tier-3 matrix run reports PASS even when the cargo test step crashed with a real prover error.

This is the same anti-pattern class as echo-types' N5Falsifier xfail gate (resolved via echo-types#136): a self-decaying CI assertion that masks proof-validity once the underlying state matures.

Scope

  • Removed continue-on-error: true from the Tier-2 live-test step (line 280).
  • Removed continue-on-error: true from the Tier-3 live-test step (line 403).
  • Renamed both steps to drop the "(allow-fail while Wave-X wires up)" qualifier.

What is NOT changed

Intentional environment resilience preserved:

  • Job-level continue-on-error on tier1-guix / tier3 / tier4 / tier4-gpu (best-effort prover installations).
  • Step-level continue-on-error on "Provision ${{ matrix.backend }}" steps (apt installation of 30+ heterogeneous provers is genuinely best-effort).
  • Step-level continue-on-error on the Tier-4-GPU "Run live version check (SKIP if binary absent)" step (version check is best-effort by design).

Verification

YAML lints clean. No Agda/Coq/Lean files touched — this is a CI-policy change only.

Refs

Estate prover-CI hygiene audit, recorded at:
`standards/adoption-readiness-grades/AUDIT-FINDINGS-2026-05-28.adoc`.

Test plan

  • CI runs on this PR (will fail-fast on real prover errors instead of silently masking).
  • After merge, observe Tier-2/Tier-3 scheduled runs to confirm honest pass/fail signal.

🤖 Generated with Claude Code

The two "Run live test for ${{ matrix.backend }} (allow-fail while
Wave-X wires up)" steps carried `continue-on-error: true`. Comments
explicitly disclosed the soft-fail was transitional, but the gate
silently masks real prover regressions: a successful Tier-2 / Tier-3
matrix run says "live test PASSED" even when the cargo test step
crashed with a real prover error.

This is the same anti-pattern class as echo-types' N5Falsifier xfail
gate (resolved via PR #136): a self-decaying CI assertion that masks
proof-validity once the underlying state matures. Removing both
soft-fails so Tier-2/3 results are honest.

Provision steps and job-level continue-on-error on best-effort
environment setup are preserved — those are intentional resilience
for apt-installing 30+ heterogeneous provers, not proof-validity
masking.

Refs hygiene audit 2026-05-28 (`standards/adoption-readiness-grades/
AUDIT-FINDINGS-2026-05-28.adoc`).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@hyperpolymath
hyperpolymath merged commit 299344a into main May 28, 2026
34 of 38 checks passed
@hyperpolymath
hyperpolymath deleted the ci/drop-soft-fail-on-live-prover-tests branch May 28, 2026 08:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant