fix(gates): green doc-refs and orphan-artefacts on main -- #872 plus agents.json after gHashTag/t27#8054

[dmitrii-f-t27]

Closes #908

doc-refs and orphan-artefacts are red on main 09662bb and on every PR built on it (#904, #907 among them). This PR makes both green by fixing references. It changes no checker code.

Two commits

1. 617fdb8 = fix(gates): green doc-refs and orphan-artefacts on main after #510 #872's commit 7cf4429, unchanged, author kept (Dmitrii Vasilev). It fixes the gft_mul: the product wire is 32 bits and GF-T32 needs 52 #510 debt: repo-qualified references, the published cross-repo list, and the orphan baseline for the two GF-T competition files. The measurements and reasoning are in fix(gates): green doc-refs and orphan-artefacts on main after #510 #872. It is carried here because fix(gates): green doc-refs and orphan-artefacts on main after #510 #872 alone stays red (next item). After this merges, fix(gates): green doc-refs and orphan-artefacts on main after #510 #872 has nothing left to add.

2. dbe7885: agents.json in AGENTS.md and CLAUDE.md. The sentence lists file formats that outside tools expect (llms.txt, agents.json, ai.txt, A2A agent cards, ...). check_doc_refs.py reads any backticked *.json as a path. Until t27b vacuous: delete 22 empty generated spec stubs (Closes #8052) t27#8054 deleted .trinity/seals/agents.json on 2026-10-09, this name matched that unrelated t27 file by basename only. That is the same false-match class fix(gates): green doc-refs and orphan-artefacts on main after #510 #872 found for PLAN.md and main.tex. The sentence never referred to that file. The fix:

   • drop the backticks around this one format name;
   • remove the two stale rows from tools/doc_refs_crossrepo.txt.

   No baseline entry is added.

Checks run locally

I ran these on a full-tree export of this head. Siblings were laid out by tools/materialise_siblings.sh; trios-mesh is private and was skipped, as in CI.

run	result
python3 tools/check_doc_refs.py	OK: no new dangling references (251 known, all still reproduce), every exclusion published
same, with the backticks put back in AGENTS.md (negative control)	FAIL: 1 NEW dangling reference(s), AGENTS.md: names \agents.json`, which does not exist`
python3 tools/check_orphan_artefacts.py	OK: no new orphaned artefacts (8 known)

Not in this PR

• The run-syntax self-test. It reads two inputs from e6eac090, which is no longer an ancestor of the rewritten main. That fix is a separate PR.
• Validate VIBEE Codegen. Its sample moves from run to run (codegen-corpus: compile sample moves with the run, so one commit is red and green (#851 W15) #885). The fix is fix(codegen): sample the corpus compile check from the baseline #886.

Dmitrii Fedorov (@dmitrii-f-t27)

[tri-claw]

Summary by t27-bees

This PR fixes two failing checks (doc-refs and orphan-artefacts) by correcting false-positive matches, adding repo-qualified cross-references, and updating baseline allow-lists for known dangling references. The agents.json backtick removal in AGENTS.md/CLAUDE.md eliminates a basename-only collision with a now-deleted t27 file; the cross-repo and upstream tables now declare the expected external locations for GF-T competition artefacts.

Worth a look

• AGENTS.md:346 / CLAUDE.md:301 — Removing backticks from agents.json stops the checker from treating it as a path. If an actual agents.json is added to this repo later, it will not be flagged as a reference (the checker only scans backticked names). This is intentional per the negative-control test.
• tools/doc_refs_crossrepo.txt — New entries declare that gft_add64.v, gft_mul.v, gft_mul32.v, trinity-gf16.tex (from GFT_DEFECTS/GF_INVENTORY) and gft_mul32.v (from arxiv_gft/README.md) resolve in tri-net. If tri-net lacks these files at the expected paths, check_doc_refs.py will fail.
• tools/doc_refs_upstream.txt:44-47 — References from GF_INVENTORY_2026-08-08.md to gHashTag/arith2027-goldenfloat/PLAN.md, gHashTag/goldenfloat-preprint/gf_preprint_v19.tex (twice), and gHashTag/paper3-methodology/main.tex are now declared upstream. Same dependency: those repos must exist and contain those paths.
• tools/doc_refs_baseline.txt — Six new baseline entries (e.g., ARXIV_GFT16_v2.md naming trinity-fpga/build/gft_mul8/gft_mul.v, GFT_DEFECTS_2026-08-08.md naming gft_ref.py, GF_INVENTORY_2026-08-08.md naming gft16_ref.py/gft_ref.py, arxiv_gft/README.md naming conformance/gft16_ref.py) explicitly allow known-dangling references. These are technical debt (gft_mul: the product wire is 32 bits and GF-T32 needs 52 #510) — if the referenced files are ever restored, the baseline must be updated or the check will silently ignore a now-valid reference.
• tools/orphan_artefacts_baseline.txt:5-6 — research/arxiv_gft/competition.json and gft_competition.pdf are added to the orphan allow-list. If these files are later referenced from tracked documentation, they will not be flagged as orphans (the checker only reports files not in the baseline and not referenced).

---

_{Read: 8 files, +35 -10, the whole diff. Head dbe7885. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}

[github-actions]

X Brain Health Check

Score: 100.0/100
Status: 🟢 HEALTHY
Threshold: 80/100

X Brain is above merge threshold

[github-actions]

🧪 Brain Stress Test Results

Score: not measured (no stress test prints a Score: line)
Status: 🟢 PASS
Threshold: 270/300

X Stress test passed