Skip to content

docs: declare three unbuilt tri commands so the documented-commands gate reads 68 again (Closes #13001) - #13002

Merged
gHashTag merged 1 commit into
masterfrom
claude/doc-commands-declare-13001
Oct 31, 2026
Merged

gHashTag merged 1 commit into
masterfrom
claude/doc-commands-declare-13001

Conversation

@gHashTag

Copy link
Copy Markdown
Collaborator

What

Declares three commands "not built yet" in three live documents, so the check Documented t27c subcommands exist reads exactly 68 dead tri mentions again.

  • .claude/skills/pareto-improvement-loop/SKILL.md: the "New tri commands" heading gains "(proposed, not built yet)" with a one-line pointer to the command contract spec, and the table row for pareto_commands.t27 carries the same words. Excuses 7 x tri pareto.
  • docs/blog/2026-10-27-t27b-backend-shootout.md: the heading "Tri CLI: tri shootout improve" gains "(proposed, not built yet)". Excuses 1 x tri shootout.
  • README.md: the PHI LOOP block line tri spec edit <module> gets back its "NOT BUILT: edit the .t27 spec directly" note, dropped by the 2026-10-31 README rewrite (4cfd2f2). Excuses 1 x tri spec.

Closes #13001

Why

The check is red on master head 21e20c4 (run 41189552897): "77 mention(s), 22 distinct (ceiling 68)". The ceiling was lowered 86 -> 68 in b49f648 with a count of 68; exactly nine undeclared mentions arrived since (the 7 tri pareto from #12914, 1 tri shootout, 1 tri spec edit). The gate fails on every PR built on master until they are declared. This is the route d565036 took for tri viz graph.

How

The checker excuses a mention whose own line, paragraph (up to three lines, stopping at a blank line) or nearest heading above says "not built", "proposed" and similar. Only prose changed: 6 lines added, 4 removed, three .md files. tools/check_documented_commands_exist.py is untouched, the ceiling is not moved, no paragraph is deleted. The other 68 mentions are the recorded ceiling and stay as they are.

The gate fails when the count falls below the ceiling as well as above it, so exactly nine were declared, not more.

Gates

Local run (kept t27c t27c-bc19b54e6, Rust tri binary of 2026-10-27, no build):

  • Before: "dead tri mentions rose 68 -> 77", exit 1 (matches CI).
  • After: "68 mention(s), 20 distinct (ceiling 68)" and "ok: every t27c subcommand a live document names exists, and the 68 dead tri mentions are the recorded ceiling", exit 0.
  • lefthook on commit: foreign-budget, own-language, secret-gate all pass. No --no-verify.

Draft. Merge is by hand on the owner's word; auto-merge is unavailable (GH_AGENT_TOKEN expired). Any other check that is red here and also red on master's own head is not touched by this PR.

🤖 Generated with Claude Code

…ate reads 68 again (Closes #13001)

The check "Documented t27c subcommands exist" fails on master head
21e20c4: 77 dead `tri` mentions against the exact ceiling of 68.
Nine arrived after the ceiling was lowered in b49f648 and none of
them said the command is not built:

- 7 x `tri pareto` in the pareto improvement loop skill (one table row
  and the six-line list). The heading above the list now reads
  "(proposed, not built yet)" and the row carries the same words.
- 1 x `tri shootout` in the heading of the t27b backend shootout post,
  now "(proposed, not built yet)".
- 1 x `tri spec edit` in the README PHI LOOP block; the "NOT BUILT: edit
  the .t27 spec directly" note the line had before 4cfd2f2 is back.

The checker excuses a mention whose own line, paragraph or nearest heading
declares it unbuilt (the route d565036 took for `tri viz graph`). No
ceiling move, no checker edit, no paragraph deleted.

Measured with the kept t27c (t27c-bc19b54e6) and the Rust tri binary of
2026-10-27: before, "dead `tri` mentions rose 68 -> 77", exit 1; after,
"68 mention(s), 20 distinct (ceiling 68)" and "ok: ... the 68 dead `tri`
mentions are the recorded ceiling", exit 0. The gate fails when the count
falls below 68 as well as above it, so exactly nine were declared.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-31 10:47:48 UTC

Summary

Status Count
Total Open PRs 39
PRs with Failing Checks 39
PRs with All Checks Green 0
READY 0
FAILING 39
PENDING 0
NO CHECKS YET 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e75b498e5e79 != 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 marked this pull request as ready for review October 31, 2026 11:36
@gHashTag
gHashTag merged commit b0809d1 into master Oct 31, 2026
32 of 33 checks passed
gHashTag added a commit that referenced this pull request Nov 7, 2026
57d63db (merged through #12997) also undid the skill hunk of #13002,
so the six `tri pareto` lines count as dead commands again and the
documented-commands gate reads 0 -> 9. This re-applies that hunk.
The other three dead mentions (shootout, spec) come from elsewhere.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Nov 7, 2026
…13308)

* fix(crons): reuse pick_may_run for fix_may_start again (Refs #12999)

PR #12997 landed an old commit (57d63db) that put the copied body of
pick_may_run back into fix_may_start, undoing #13000. Every PR built on
master since then fails duplicate-bodies with a NEW group. This
re-applies #13000's change: fix_may_start calls the night loop's
pick_may_run. python3 tools/dupe_scan.py: ok, 689 groups, none grew.

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

* docs(skills): mark tri pareto as not built yet again (Refs #12999)

57d63db (merged through #12997) also undid the skill hunk of #13002,
so the six `tri pareto` lines count as dead commands again and the
documented-commands gate reads 0 -> 9. This re-applies that hunk.
The other three dead mentions (shootout, spec) come from elsewhere.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Nov 7, 2026
…s 0 (Refs #13204)

The documented-commands gate holds dead tri mentions at exactly
MAX_DEAD_TRI = 0. Two "not built" markers that #13002 added were lost
on master after the ceiling went to 0, so every PR merge commit reads
0 -> 2. Restore them: README.md `tri spec edit` and the shootout blog
`tri shootout improve` heading. Prose only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Nov 7, 2026
…#13204) (#13313)

* spec(t27os): T27Lagger, the cost-ordered ladder law (free tier first, paid only with a grant)

Measured 2026-11-05 on this Mac: LM Studio on 127.0.0.1:1234 answers 200 in
9.4 s with no key (the only free path alive), the z.ai coding pool is 429/1310
once and 429/1113 four times, ollama cloud is 402, the openrouter free id is
401 (stale key), nvidia is not measured. The path table carries that reading
with its date; the walk visits every free path before any paid one, a paid
path needs the owner's grant, an unmeasured cost is the most expensive class,
the request shape is per (path, model) -- z.ai glm-5.3-flash omits the
thinking field (400/1210 otherwise) -- and the 503 names every tried path
with its cost class.

Closes #13204

(cherry picked from commit 9e9029cbdf, local branch feat/lagger-cost-free-first)

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

* docs: re-declare two unbuilt tri commands so documented-commands reads 0 (Refs #13204)

The documented-commands gate holds dead tri mentions at exactly
MAX_DEAD_TRI = 0. Two "not built" markers that #13002 added were lost
on master after the ceiling went to 0, so every PR merge commit reads
0 -> 2. Restore them: README.md `tri spec edit` and the shootout blog
`tri shootout improve` heading. Prose only.

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

* data(t27b): add the ledger row for specs/t27os/lagger.t27 (Refs #13204)

t27b-native-ratchet names lagger.t27 UNLISTED: the reference passes it,
t27b passes it, and docs/reports/t27b_expectations.json has no row. A spec
PR carries its own row (specs/tri/t27b/ledger_steward.t27). The ratchet's
other findings (golden, shootout codegen_audit, three t27os specs) are
master's and are red on master 8839ccc too.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

docs: declare three unbuilt tri commands so the documented-commands gate reads 68 again

1 participant