fix(gates): green doc-refs and orphan-artefacts on main after #510

[gHashTag]

What

Since #510 merged (57de4a1, 2026-10-04 18:42Z), main fails Document references (run 37225448197) and Orphaned artefacts (37225448159). Every PR whose merge commit includes #510 fails those two checks as well. #867 and #870 are examples: their own changes touch neither checker. A PR that also changes a workflow fails gates-can-fail too, because that gate reports both checkers as UNTESTABLE on a red tree. #870 is that case.

#510 was opened on 2026-08-08, before the GF-T → TEF → TNF renames. It added research/ARXIV_GFT16_v2.md, research/arxiv_gft/ and two dated notes, all of which name files by their August names.

This PR changes gate data and one inventory document. It changes no checker code.

failure on main	count	what this PR does
bare-name references to our other repos	3	qualified as gHashTag/<repo>/<path>, using the repo named in the same inventory row. Two of them, PLAN.md and main.tex, had matched unrelated files in tri-net and t27 by name alone
cross-repo exclusions not in tools/doc_refs_crossrepo.txt	9 → 7	the 7 genuine ones are published (tri-net fpga/gft/gft_mul.v, gft_mul32.v, gft_add64.v; t27 docs/arxiv-submission/trinity-gf16.tex). The other 2 are the false matches above
references to files no tree has	6	the old GF-T oracles (gft16_ref.py, gft_ref.py) and an untracked build directory. Added to tools/doc_refs_baseline.txt, where the same names from sibling documents of the same week already are. research/RENAME_MAP.md now says what they became
baseline entries that no longer reproduce	2	removed: #510 added ARXIV_GFT16_v2.md and gft_paper.tex, the files they named
artefacts with no generator	2	the competition table and figure in research/arxiv_gft/. GitHub code search across gHashTag finds no generator for either; it covers default branches only. Listed in tools/orphan_artefacts_baseline.txt

The lists and the doc-refs baseline were written by the checker's own --update-lists and --update-baseline. The diffs are exactly the rows above and nothing else.

Measured

All local runs used the checkers from main on a full export of 57de4a1. Siblings were laid out as tools/materialise_siblings.sh lays them out: an empty file per path at each sibling's HEAD, with private trios-mesh skipped because CI cannot clone it.

doc-refs counts are new dangling references / unpublished exclusions / stale baseline entries.

run	doc-refs	orphan-artefacts
main, CI	FAIL 7 / 9 / 2	FAIL 2
main, local	FAIL 7 / 9 / 2, all 18 lines identical to CI	FAIL 2, same two paths
this branch, local	OK, 251 known, all still reproduce	OK, 8 known
this branch + a planted `tools/x7_negative_control_absent.py`	FAIL 1 new, the planted one	—
this branch + a planted research/x7_negative_control.json	—	FAIL 1, the planted one

Not changed

• The GF-T documents' claims. The accuracy row in research/arxiv_gft/README.md names an oracle that this tree no longer has. Its numbers cannot be re-derived here, and this PR says so in RENAME_MAP rather than pointing the row at a renamed oracle that was never run against them.
• research/block/kurtosis_all.json is still in the orphan baseline, but it is no longer an orphan, because two Markdown files now name it. Unlike doc-refs, the orphan gate does not fail on a stale baseline entry, and it counts any .md that names a file as that file's producer. Both are checker questions and belong in their own PR.
• tools/doc_refs_split.md is a snapshot that no gate reads, and it is not regenerated here.
• Codegen Validation also failed on 57de4a1. That failure is unrelated: the same check passed on 066bfa2, a descendant of main with only workflow changes. It is codegen-corpus: zigTestOk hides why a sampled spec does not compile (phi_utils_multi flake on main) #871. The tool discards the compiler's stderr, so the reason cannot be read from the log.

Not measured

• The CI run of this branch, which is the real test; its checks follow.
• Whether the tri-net fpga/gft/gft_mul.v is byte-identical to the August build copy that ARXIV_GFT16_v2.md names.

Refs #851, #870

🤖 Generated with Claude Code

[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

[tri-claw]

Summary by t27-bees

This PR fixes two failing CI gates (document references and orphaned artefacts) caused by #510, which added historical GF-T documents that reference renamed/moved files and untracked build paths. The fix qualifies bare cross-repo names in the inventory, adds the now-missing GF-T oracle files and a build directory to the doc-refs baseline, declares genuine cross-repo targets in tri-net and t27 as exclusions, and lists two competition artefacts in the orphan baseline because no generator exists in any default branch.

Worth a look

• tools/doc_refs_baseline.txt:+193-221 — Six new baseline entries suppress warnings for gft16_ref.py, gft_ref.py, and trinity-fpga/build/gft_mul8/gft_mul.v, which no tree contains. If any of these files are later added under a different path, the baseline will silently hide the dangling reference.
• tools/doc_refs_crossrepo.txt:+5-12 — Seven new cross-repo exclusions assume the named files (gft_add64.v, gft_mul.v, gft_mul32.v in tri-net; trinity-gf16.tex in t27) stay at those exact paths. A rename or move in either sibling repo will make the exclusion stale and could mask a real broken reference.
• tools/orphan_artefacts_baseline.txt:+5-6 — research/arxiv_gft/competition.json and gft_competition.pdf are declared orphans based on a GitHub code search limited to default branches. If a generator exists only in a feature branch or private fork, the baseline will prevent the gate from flagging it when merged.
• research/GF_INVENTORY_2026-08-08.md:+24-118 — Three bare names (PLAN.md, gf_preprint_v19.tex, main.tex) are rewritten to gHashTag/<repo>/<path>. If the doc-refs checker's upstream resolution expects a different prefix format (e.g., without gHashTag/), these references will become unresolved.
• tools/doc_refs_upstream.txt:+44-47 — Four new upstream entries mirror the qualified names above. They must stay in sync with the inventory; a future edit to the inventory that changes a path without updating this file will create a false "unresolved upstream" failure.

---

_{Read: 6 files, +33 -6, the whole diff. Head 7cf4429. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}

[dmitrii-f-t27]

Your commit 7cf4429 landed on main unchanged, as part of #909 (squash b077048, with you as co-author). #909 also fixes the last red reason on this head: AGENTS.md and CLAUDE.md named agents.json in backticks. That name had matched .trinity/seals/agents.json in t27 until gHashTag/t27#8054 deleted that file. doc-refs and orphan-artefacts were green on #909's head. This PR has nothing left to add to main; closing it is your call.

Dmitrii Fedorov (@dmitrii-f-t27)