Skip to content

feat(trinity): re-pin the S01 manifest to gHashTag/trinity#1333 and add the test.brain card - #5969

Closed
gHashTag wants to merge 1 commit into
masterfrom
feat/trinity-brain-card-repin
Closed

gHashTag wants to merge 1 commit into
masterfrom
feat/trinity-brain-card-repin

Conversation

@gHashTag

@gHashTag gHashTag commented Oct 4, 2026

Copy link
Copy Markdown
Owner

Summary

gHashTag/trinity#1333 restores the brain test steps that gHashTag/trinity#517 removed from build.zig. Brain CI is green there, but the headless-profile job fails with 11 UNASSIGNED_TARGET violations because no card owns the new targets. The consumer cannot add that card itself. external/t27 is a byte-locked copy of these cards, and the manifest check reads targets at the revision project.t27 pins. So the pin moves here first, and the consumer re-vendors after this PR merges.

This PR:

  • re-pins the S01 manifest from 976df517 to 291ac8b24, the head of trinity#1333;
  • re-measures the counts at that revision;
  • adds specs/trinity/capabilities/test.brain.t27 (S03).

What changed

specs/trinity/project.t27

  • PINNED_REVISION moves to 291ac8b24308ee39703393b6ea5363c166079b79.

  • Changed counts, re-measured by tools/trinity_manifest.py inventory on a full checkout:

    976df517 291ac8b24
    .zig files 2830 2831
    unreachable 2082 2076
    tests 73 79
    steps 66 71
    canonical / mirror / vendored .t27 31 / 1044 / 0 40 / 2291 / 70
    catalog 856 specs from 8 repos 1776 specs from 14 repos
  • Unchanged: executables 51, libraries 6, installs 46 / 3 guarded, registry 29, dependencies, T27_REVISION.

test.brain.t27 (new)

  • Owns step:test-basal-ganglia, step:test-reticular-formation, step:test-locus-coeruleus, step:test-brain, step:test-brain-stress and test:src/brain/*.
  • EVIDENCE is measured, from S3AI Brain CI run 37184075728. That run was on push at 291ac8b24 exactly; all five steps exited 0 (51/51, 48/48, 52/52, 151/151, 261/261).

research.unreferenced-sources

  • Keeps src/brain. The restored steps reach 6 of the 43 files of src/brain that build.zig did not reach at 976df517, and 37 stay unreachable.

The 50 existing card headers

  • They now say "the consumer tree that specs/trinity/project.t27 pins" instead of quoting 976df517.

The 18 measured cards

  • Their EVIDENCE_SOURCE now says the measurement was taken at 976df517 and was not repeated at the new pin. See "Not measured".

conformance/trinity/inventory.json and report.json

  • Regenerated.

specs/trinity/README.md

  • New section for the 291ac8b24 pin. The 976df517 section is kept under it.

Seals

  • 52 seals re-saved: project.t27, the 50 cards and the new card.

Left unchanged

  • S03's build_graph.t27 and every other per-spec pin stay at 976df517.
  • queen_views keeps afc9d384.

Measured locally

The clone was sparse: specs/{api,brain,catalog,fpga,isa,numeric,queen,tools,trinity,ui,vsa}, tools, .trinity/seals, conformance/trinity, docs/system, .github/workflows. The trinity tree was checked out in full at 291ac8b24, with all 2831 .zig files present.

check result
trinity_manifest.py check OK -- 52 capabilities over gHashTag/trinity@291ac8b24, 51 executables / 6 libraries / 79 tests / 71 steps all owned
negative control: test.brain.t27 removed 11 UNASSIGNED_TARGET (the same 11 the trinity#1333 job reports)
negative control: test.graph also owns step:test-brain DUPLICATE_TARGET
trinity_manifest.py --self-check PASS
t27c seal --verify, 52 specs all MATCH
check_seal_currency.py exit 0: 361 current, 0 newly stale; the rest are absent from the sparse clone
check_specs_still_parse.py --base origin/master 52 changed, 0 newly unparseable
check_duplicate_declarations.py exit 0

Not measured

  • check_seal_coverage.py and check_ring_spec_drift.py gave no usable result on the sparse clone.
    • Coverage flagged 530 specs, and every one of them is outside the sparse set.
    • Ring drift found no rings.
    • CI is authoritative for both.
  • Seals were saved with t27c 0.4.0 built before master's fb83b54b3.
    • The compiler changes since aca71b5 re-sealed 10 specs that carry structs; none of them is in this PR.
    • If CI's t27c disagrees on any gen hash here, the seal needs re-saving.
  • The 18 measured cards were not re-measured at 291ac8b24. Their evidence strings now say so, rather than claiming the measurement carried over.
  • test.brain's single ACCEPTANCE command, with all five steps at once, was not run. Each step was run on its own in the cited run.
  • The S03 build graph was not re-pinned. A graph run at 291ac8b24 reported only PIN_MISMATCH (edges 252 -> 255, package edges 152 -> 155).
  • The pin is a pull-request head, not a commit on trinity main.
    • If trinity#1333 is squash-merged, 291ac8b24 stays fetchable but is not in main's history.
    • A later re-pin to the merge commit is a separate change.

After this merges, in gHashTag/trinity#1333

  1. Run python3 tools/contracts.py vendor at the new t27 revision.
  2. Add trinity/test.brain to RUN or NOT_RUN in specs/reproduce/capabilities.t27, since the card has an ACCEPTANCE.
  3. Rebind the work report.

Opened as a draft. Merging is the owner's decision.

Refs #3563

🤖 Generated with Claude Code

…dd the test.brain card

gHashTag/trinity#1333 restores the brain test steps that trinity#517 had
dropped from build.zig. The consumer cannot carry a card for them on its
own: external/t27 is a byte-locked vendored copy of these cards, and the
manifest check reads targets at the revision project.t27 pins. So the pin
moves first, here, and the consumer re-vendors after this merges.

- specs/trinity/project.t27: PINNED_REVISION 976df517 -> 291ac8b24 (head
  of trinity#1333). Counts re-measured by tools/trinity_manifest.py
  inventory on a full checkout: 2831 .zig (2076 unreachable), 79 tests,
  71 steps, 40 canonical / 2291 mirror / 70 vendored .t27, catalog 1776
  specs from 14 repositories. Executables 51, libraries 6, installs 46/3,
  registry 29, dependencies and T27_REVISION unchanged.
- specs/trinity/capabilities/test.brain.t27 (new, S03): owns the five
  restored steps and test:src/brain/*; measured by S3AI Brain CI run
  37184075728 on push at 291ac8b24.
- research.unreferenced-sources keeps src/brain: 37 of its files are
  still unreachable; the restored steps reach the other six.
- The 50 card headers name the pin by reference to project.t27 instead
  of quoting 976df517. The 18 measured cards say their evidence is from
  976df517 and was not measured again at the new pin.
- conformance/trinity/{inventory,report}.json regenerated; README section
  for the new pin, the 976df517 bullets kept below it.
- 52 seals re-saved with t27c seal --save; --verify gives MATCH for all.

S03's build_graph.t27 and the other per-spec pins stay at 976df517.

Refs #3563

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

github-actions Bot commented Oct 4, 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 4, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-04 08:37:12 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 45
PRs with All Checks Green 5
READY 1
FAILING 45
PENDING 0
NO CHECKS YET 0

These columns do not partition: 1 + 45 + 0 + 0 = 46, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=46f5599e58f6 != 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 commented Oct 4, 2026

Copy link
Copy Markdown
Owner Author

Closed as superseded by #5965 (squash 39e06aa), which adds trinity/brain.regions owning the same eleven brain targets. Merging both would give DUPLICATE_TARGET. The S01 pin stays at 976df517 instead of moving to 291ac8b24, the head of the still-unmerged gHashTag/trinity#1333, which a squash merge would leave off trinity main. The branch is kept.

@gHashTag gHashTag closed this Oct 4, 2026
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