Skip to content

ci: two stale seals (petal_menu, bench) make spec-guards and coverage red on every PR; no reseal automation exists #11640

Open

Description

@gHashTag

spec-guards and coverage are red on every open pull request (#11583, #11634, #11635) from two stale seals on master itself, not from any diff, and nothing in the repository repairs them automatically: a grep -r reseal specs/crons/ specs/jobs/ returns nothing, so the repair has only ever been someone remembering to type it. Measured 2026-10-22 on a clean origin/master worktree (aeb772f46b):

The repair is the gate's own remedy line, for these two specs only: t27c seal <spec> --save && tri seals sync-twins. Baselining (check_seal_coverage.py --update-baseline) is the WRONG repair here, by that tool's own wording: it records drift as accepted debt instead of recording what the compiler now produces.

Whether a steward job should reseal on a schedule is out of scope for this issue -- this issue makes the two gates green and records the measurement; the missing automation is a separate decision for the owner (it needs a cron card in specs/crons/ first, per the repo's cron registry rule).

Boundary

  • .trinity/seals/t27f_T27fPetalMenu.json
  • .trinity/seals/t27os_T27osBench.json
  • tools/seal_baseline.txt

User Scenarios

Given a fresh clone of master, when both seal checks run with a built t27c, then check_seal_currency.py --stale-specs prints an empty list and check_seal_coverage.py exits 0.

Given the resealed tree, when t27c corpus --json is compared to the pre-reseal baseline taken on the same binary, then the acceptance columns (generates / compiles per backend) are identical -- a reseal changes only what seals claim, never what the compiler emits.

Given a pull request opened after this fix, when spec-guards and coverage run on it, then they no longer fail on petal_menu and bench, and any red they do report is about the PR's own change.

Requirements

  • FR-001: The reseal MUST be produced by t27c seal <spec> --save for exactly the two stale specs, never by hand-editing a hash in .trinity/seals/.
  • FR-002: tri seals sync-twins MUST run after the reseal, so a twinned seal (if either spec carries one) inherits the same claims; it must refuse rather than propagate gen_hash=none.
  • FR-003: The two specs' own tests MUST pass before the reseal is committed (measured: 12/12 and 28/28); a failing test makes the reseal the wrong move and this issue must be reopened with that fact.
  • FR-004: The pre/post t27c corpus --json acceptance columns MUST be identical; any moved column stops the repair and is reported here instead of being resealed past.
  • FR-005: No baseline line may be added to tools/seal_baseline.txt for these two specs (baselining is the explicitly wrong repair for gen-drift).

Success Criteria

  • python3 tools/check_seal_currency.py --stale-specs exits 0 with an empty list on the merged tree.
  • python3 tools/check_seal_coverage.py exits 0 on the merged tree (0 newly-unheld seals).
  • git diff origin/master -- tools/seal_baseline.txt is empty.
  • The post-reseal t27c corpus --json acceptance columns equal the pre-reseal baseline, and both JSON lines are attached to this issue.
  • On a PR opened after the merge, spec-guards and coverage pass.

Activity

  1. gHashTag commented on Oct 25, 2026

    @gHashTag
    CollaboratorAuthor

    One more instance, measured on PR #12060 at head 8dd35fbc4b (master merged in at 15acb60995, 2026-10-25 10:49Z), spec-guards job:

    STALE generated-code hash     : 2
      StorageKv.json:         gen_hash_rust sealed=34fcb9c62268 current=d93c8652c75b  (specs/storage/kv.t27)
      storage_StorageKv.json: gen_hash_zig  sealed=3bfef2b80546 current=f31afd9e6274  (specs/storage/kv.t27)
      (c and verilog moved too: babf48544cbe -> 90d92edcc258, 9b5ae5e084a6 -> d6a6e8b0350e)
    

    specs/storage/kv.t27 last moved in #4660 (0bdd185d9f, "Wave 684"), which did not re-save the seal. No open PR re-seals it (gh pr list --search "StorageKv OR kv.t27 seal" is empty). The four seals of this morning (#12348) are green again; StorageKv is the one that keeps spec-guards red on every PR that merges today's master.

    Fix is the usual one: t27c seal specs/storage/kv.t27 --save from a master-tip build, then tri seals sync-twins, in one PR. Not done from the WASM lane: its local t27c predates #12335 and would write the wrong hashes.

  2. gHashTag commented on Oct 25, 2026

    @gHashTag
    CollaboratorAuthor

    Follow-up to the StorageKv note above, 2026-10-25.

    This issue stays open for the missing reseal automation.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions