Skip to content

fix(seals): reseal the seals the corpus and the compiler moved - #5578

Merged
dmitrii-f-t27 merged 2 commits into
masterfrom
fix/reseal-stale-seals
Oct 2, 2026
Merged

fix(seals): reseal the seals the corpus and the compiler moved#5578
dmitrii-f-t27 merged 2 commits into
masterfrom
fix/reseal-stale-seals

Conversation

@dmitrii-f-t27

@dmitrii-f-t27 dmitrii-f-t27 commented Oct 2, 2026

Copy link
Copy Markdown
Collaborator

What

The seals no longer describe what the tree generates:

Nothing in the merge path reseals. The first gate sat behind Spec Guards' first red step, and the second was empty until #5454.

Why they moved (measured, not assumed)

90 specs did not change; their output did. I built t27c at seven checkpoints between 2026-09-08 and 2026-10-01, then three more inside 2026-09-17, and hashed each spec's output per backend. Every attributable drift lands on the 2026-09-17 compiler changes:

commit specs moved (by backend)
#3962 empty array literal ("and the seal moves for it") zig 66, verilog 47, c 47
#4114 sized array element type zig 70
#3973 single string element zig 34, c 34, rust 30, verilog 29

The rest were sealed before 2026-09-09, inside the 2026-09-08 gen-c/gen-rust wave (#3401..#3502). Twelve seals from 2026-10-01 are the S07-S10 seals made with a pinned binary.

245 specs changed after sealing: 154 last edited by bee commits that implement bodies and never reseal, 91 by repairs on master.

Change

  • 336 specs resealed with the t27c built from master. The compiler sources are unchanged since 756bcff, which is what the binary was built from.
    • The 307 that check_seal_currency.py --stale-specs lists.
    • 29 more that the coverage gate flags, 34 of them hollow seals (gen_hash=none) whose specs now generate and get real hashes.
  • specs/ar/ternary_logic.t27 still generates in no backend (it is in specs_generate_baseline.txt). It is resealed with --force, on the record as hollow.
  • tri seals sync-twins: all twins consistent. Four seal files are new, each a twin of an existing seal whose spec's module name changed (fifo_tb, cordic, adapter, lsp/language).
  • The seven specs/ml/ specs are left to fix(specs): seven ml specs stop discarding tokens #5575, which reseals them with their repair.

Result

before after
currency: stale gen hashes 544 14 (the seven ml specs, #5575)
coverage: seals that do not hold 604 14 (same)
hollow seals 65 31

With #5575 both reach 0. A reseal records what the compiler produces, not that it is correct; this follows #3440 (57 specs) and #2909 (926 seals).

Refs #5497

🤖 Generated with Claude Code

@dmitrii-f-t27 @claude
fix(seals): reseal the seals the corpus and the compiler moved
c359cda 
check_seal_currency.py reported 544 stale generated-code hashes on master
db870cd, and the seal-coverage gate restored by #5454 604 seals that no
longer hold. Nothing in the merge path reseals; the first gate sat behind
Spec Guards' first red step and the second was empty until #5454.

Why they moved, measured. 90 specs never changed while their output did:
bisecting t27c across the history puts it on the 2026-09-17 compiler changes
(#3962 "and the seal moves for it", #3973, #4114) and the 2026-09-08
gen-c/gen-rust wave, plus S07-S10 seals made with a pinned binary on
2026-10-01. 245 specs changed after sealing: 154 last edited by bee commits
that implement bodies and never reseal, 91 by repairs on master.

336 specs are resealed with the t27c built from master (compiler sources
unchanged since 756bcff): the 307 currency-stale ones and 29 more the
coverage gate flags, 34 of them hollow seals whose specs now generate.
specs/ar/ternary_logic.t27 still generates nowhere and is resealed with
--force, on the record as hollow. tri seals sync-twins: all consistent.

Currency stale 544 -> 14 and coverage 604 -> 14; the 14 are the seven
specs/ml/ seals #5575 reseals. Hollow 65 -> 31. A reseal records what the
compiler produces, not that it is correct, as in #3440 and #2909.

Refs #5497

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

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-02 11:08:56 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 46
PRs with All Checks Green 4
READY 3
FAILING 46
PENDING 0
NO CHECKS YET 0

These columns do not partition: 3 + 46 + 0 + 0 = 49, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=b7d5cc5c4cf1 != 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).

@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@github-actions

github-actions Bot commented Oct 2, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-02 11:16:02 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 46
PRs with All Checks Green 4
READY 3
FAILING 46
PENDING 0
NO CHECKS YET 0

These columns do not partition: 3 + 46 + 0 + 0 = 49, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=b7d5cc5c4cf1 != 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).

@dmitrii-f-t27
dmitrii-f-t27 merged commit d04bf14 into master Oct 2, 2026
32 of 35 checks passed
This was referenced Oct 2, 2026
Closed
Merged
This was referenced Oct 2, 2026
Closed
Open
Merged
Open
gHashTag added a commit that referenced this pull request Oct 2, 2026
@gHashTag @claude
fix(seal): seal --save runs the spec's tests and refuses on FAIL (refs …
3355ca8 
…#5577) (#5605)

* fix(seal): seal --save runs the spec's tests and refuses on FAIL (#5577)

#5578 resealed 336 specs with `t27c seal --save`, which refused only when a
backend failed to generate and never ran a test. 13 class-(c) specs with
failing tests got fresh seals, and the hash-only seal-coverage gate reported
them as holding.

- seal --save now runs the same machinery as `t27c test-report`.
  FAIL: refuse, name every failing test, exit 1, write nothing.
  BLOCKED (e.g. no zig): seal with a notice; blocked is not failing.
  pass: seal, printing the measurement.
  --force: seal anyway, with the failures recorded in the seal.
- The seal records a `tests` object; twins get the same record.
- check_seal_coverage.py reads `tests.failed > 0` as a new kind,
  `tests-fail`, so a forced seal is on the record instead of holding.
  Self-check gains three controls.
- tools/seal_baseline.txt: the 26 seal files of the 13 specs from #5577
  are ledgered as known-broken (`stale`, the state #5580 left them in),
  with the failing tests named in the detail. Nothing was resealed.
- test_report's scratch dir is per-process, so concurrent runs don't collide.

Refs #5577

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

* docs(now): the ledger lines cite #5577 rather than name the tests

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag gHashTag mentioned this pull request Oct 3, 2026
Merged
gHashTag added a commit that referenced this pull request Oct 3, 2026
@gHashTag @claude
fix(gft): sadd orders before it subtracts, underflow is zero; seal cu…
34849f5 
…rrency reads the stale ledger (#5737)

* fix(gft): sadd orders before it subtracts; magsub underflow is zero (Fixes #5506)

emit-bitexact's Zig arm (-OReleaseSafe) trapped on two defects that C and
Rust computed wrongly without trapping:

- sadd called magsub(ma, mb) before ordering the operands, so magsub shifted
  by a negative count (UB in C, unchecked in Rust -O, a trap in Zig).
- below the smallest binade the normalisation floors and mant goes negative,
  so (off << 9) | mant is no GF-T16 encoding (sadd(256, 66047) = 0xFFFFFEFF);
  C/Rust wrap on `as u32`, Zig's @intcast traps. Such a difference is now 0,
  as enc() already makes anything below 2^-40.

The identical body lives in 31 specs (the duplicate-body ledger keeps the
group whole); all are fixed and the 30 sealed ones resealed with their tests
passing. The Python model gets the same rule, a self-test and a planted
mutant. The gate's comparison is unchanged.

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

* fix(spec-guards): seal currency reads the stale ledger it was blind to (Refs #5577)

check_seal_currency failed master on 30 seals (15 specs whose own tests
fail, #5577) that #5580 deliberately left on their pre-#5578 hashes and that
tools/seal_baseline.txt already records as kind `stale`. It never read that
ledger, so it was red by design and a NEW stale seal hid behind the 30.

It now reads the ledger through check_seal_coverage.baseline() (one ledger,
one reader) and forgives a stale seal only if the ledger says `stale` AND
its spec_hash still disagrees with the spec. Unmoved-spec drift and any
unledgered stale seal still fail; a ledgered seal that holds again is a
NOTE to shrink the ledger. The self-check plants each condition away.

Resealing the 15 would certify broken contracts; the repair is fixing each
spec, then `t27c seal --save`. spec-guards now also triggers on the ledger
and its reader.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@gHashTag gHashTag Awaiting requested review from gHashTag gHashTag is a code owner

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@dmitrii-f-t27