Skip to content

feat(t27b): bless accounts for a cap rise; doctor alarms from per-spec records - #6242

Merged
gHashTag merged 1 commit into
masterfrom
claude/t27b-cap-accounting
Oct 4, 2026
Merged

gHashTag merged 1 commit into
masterfrom
claude/t27b-cap-accounting

Conversation

@gHashTag

@gHashTag gHashTag commented Oct 4, 2026

Copy link
Copy Markdown
Owner

Closes #6237
Refs #6063 #6198

Slice 5 of moving the t27b steward into .t27. Two decisions move into specs/tri/t27b/steward.t27:

  • cap_rise_is_new(not_pass, old_cap, new_not_pass): a cap rise is covered only if the non-pass entries the old ledger already named still fit the old cap.
  • is_alarm(t27b, reference): a mismatch or a crash is always an alarm; a fail or a timeout is an alarm only where the reference passes.

Tool

  • tri t27b ratchet --bless prints the accounting when it refuses a rise, and says whether --accept-new would cover it.
  • --accept-new raises the cap by exactly the number of new specs. A rise from a spec the ledger already named is still refused.
  • The doctor's LAB-ERROR counts fail and timeout from results through is_alarm. When a run has no per-spec results, it falls back to the summary.

Evidence

  • t27b test --check: 62/62. t27c lab test-report: 62/62.
  • gen/c/tri/t27b/steward.c regenerated on the t27c lab. tri t27b gen-check answers SAME b46d92e084be.
  • Live run against lab run 88862a9 and the master ledger (cap 582):
  • Live doctor: the LAB-ERROR t27b_fail 1 false alarm for build_verify.t27 is gone. The reference fails that spec too.
  • CI tests: steward PASS, ratchet PASS (new cases for a refused rise and an accepted rise), tick 47 ok (a shared-fail fixture with no anomaly, plus an own-fail control that raises LAB-ERROR), gen-check PASS.

🤖 Generated with Claude Code

…c records

steward.t27 gains cap_rise_is_new and is_alarm (62 tests, both compilers).
`tri t27b ratchet --bless` prints the accounting on a refused rise and
`--accept-new` raises the cap by exactly the specs new to the ledger; a
rise from a named spec is still refused. The doctor's LAB-ERROR counts
fail/timeout through is_alarm, so a fail the reference shares
(build_verify.t27) is no longer an alarm.

Closes #6237
Refs #6063 #6198

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

github-actions Bot commented Oct 4, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-04 19:27:17 UTC

Summary

Status Count
Total Open PRs 46
PRs with Failing Checks 28
PRs with All Checks Green 18
READY 12
FAILING 28
PENDING 0
NO CHECKS YET 0

These columns do not partition: 12 + 28 + 0 + 0 = 40, and there are 46 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=8597b6ded596 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@gHashTag
gHashTag merged commit c5406e6 into master Oct 4, 2026
24 of 26 checks passed
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.

t27b: bless accounts for a cap rise, doctor stops alarming on a fail the reference shares

1 participant