Skip to content

One file per guest; the train runs the browser acceptances and says why it waits - #164

Merged
github-actions[bot] merged 7 commits into
mainfrom
claude/project-thread-jbqzew
Nov 16, 2026
Merged

github-actions[bot] merged 7 commits into
mainfrom
claude/project-thread-jbqzew

Conversation

@afflom

@afflom afflom commented Nov 16, 2026 •

Copy link
Copy Markdown
Owner

Guests no longer share a file

Every new guest appended to app/specs/guests/spec.json, app/specs/guests/lock.json and generators/guests/steps.json. The open guest pull requests (#159 to #162) each conflicted with the others there.

  • Each guest is app/specs/guests/<facet>/<id>.json, with its lock beside it as <id>.lock.json. <facet> is the directory under site/ that holds its wasm. spec.json keeps only the registry's name, title and purpose; lock.json is gone.
  • Each guest step is its own file, generators/guests/steps/<facet>.json, and reads app/specs/guests/<facet>/*.json. tools/generators/run.sh reads a program's steps/*.json in name order, beside or instead of its steps.json. A program's pin does not cover its steps directory, as it does not cover steps.json.
  • tools/guests/gen.py reads and writes the per-guest files.
  • The LexLean pin generator (generators/guests, GenGuests) folds over the guest files in the request and no longer over one array. It is rebuilt and pinned, and its acceptance vectors use the new layout.
  • A loader with several guests now lists them in id order, so site/google/guest.mjs, site/holon/guest.mjs and site/read/guest.mjs change in order only. The pinned hashes are the same.

To add a guest now: add app/specs/guests/<facet>/<id>.json; for a new facet also add generators/guests/steps/<facet>.json. An open guest pull request moves its entry out of spec.json, lock.json and steps.json into those files when it next merges main.

The parts, standards and oracle registries have the same shape and still conflict; they are not in this change.

The train runs the browser acceptances on the merged tree

The train ran the fast gates on each pull request merged with main, but the browser acceptances ran only on each pull request alone. #145 and #161 were each green, and main then failed CO-01 and the timing budgets.

After the fast gates pass, the train now:

  • builds a holo from the merged tree (no proof: main's own run binds that),
  • assembles the page as the page job does,
  • runs every test/*/run.mjs and every *.test.mjs once.

It does this in a git worktree, so the train's tree stays what it merges, and in a shell of its own, because bash ignores errexit inside an if. The test job's limit goes from 90 to 240 minutes. This adds about 15 minutes per pull request the train takes.

The train says why it waits

A pull request whose required check ended without passing was skipped by the train without a word (#155: page / gates cancelled at the old time limit, which read as green). The train now comments once for that head and names the check and how it ended.

Both train changes are in the model, app/specs/pipeline/workflows/merge-train.yml.json.

Checks run

  • tools/guests/gen.py check: 18 current.
  • tools/generators/run.sh --check
  • tools/standards/gate.py --check --base origin/main
  • tools/handwritten/gate.py --check --base origin/main: 26,369 hand-written lines, none added or grown.
  • cargo test -p repo-conformance: all pass except router_ru_03 (QR-03 labeling-ms-per-100, 301 against 250). That fails on main too since Read mail headers through the LexLean grammar guest, synchronously #145 and is being fixed in another pull request; this change does not touch it.
  • The train's acceptance step, locally end to end: the holo builds, the page assembles and all 281 tests run in about 13 minutes. The gates' own tests pass in a git worktree.

Not yet run on a runner: the new train step. I will run the train from this branch by hand once main is green, before this merges.

🤖 Generated with Claude Code

https://claude.ai/code/session_013Asa3LuNYtPDBa4FbSEQUZ

Every new guest appended to app/specs/guests/spec.json, lock.json and
generators/guests/steps.json, so guest pull requests conflicted with each
other there. Each guest is now app/specs/guests/<facet>/<id>.json with its
lock beside it as <id>.lock.json, and each guest step is its own file under
generators/guests/steps/. tools/guests/gen.py and the LexLean pin generator
(GenGuests, rebuilt) read those; run.sh reads a program's steps/*.json.
The guests of a loader are now listed in id order.

Merge train: a pull request whose required check ended without passing
(cancelled at the time limit, say) was skipped without a word. The train
now says so once for that head.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3LuNYtPDBa4FbSEQUZ
…-jbqzew

# Conflicts:
#	app/specs/guests/lock.json
#	app/specs/guests/spec.json
#	generators/guests/steps.json
@github-actions

Copy link
Copy Markdown
Contributor

Merge train notice: the merge train merges this pull request once it is green on current main; don't enable auto-merge.


Generated by Claude Code

The train ran the fast gates on each pull request merged with main, but the
browser acceptances only ran on each pull request alone. #145 and #161 were
each green and main then failed CO-01 and the timing budgets.

After the fast gates pass, the train now builds a holo from the merged tree
(no proof: main's own run binds that), assembles the page as the page job
does, and runs every test/*/run.mjs and every *.test.mjs once. It does this
in a git worktree, so the train's tree stays what it merges, and in a shell
of its own, because bash ignores errexit inside an if. The test job's limit
goes from 90 to 240 minutes.

Tried locally end to end: the holo builds, the page assembles and all 281
tests run in about 13 minutes.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3LuNYtPDBa4FbSEQUZ

afflom commented Nov 16, 2026

Copy link
Copy Markdown
Owner Author

Status of this pull request:

  • The red check is main's, not this change's. page / gates failed on 3a955ee with 1 of 281 browser acceptances failing. Main has been red in the same step since Read mail headers through the LexLean grammar guest, synchronously #145 (QR-03 labeling-ms-per-100, and CO-01); see The live page is behind main: its deploy is blocked #141. This pull request does not touch the reader or the page. It goes green once the fix for main is in and main is merged here.
  • New commit 0f8192d: the merge train also runs the browser acceptances on the merged tree, after the fast gates. Tried locally end to end (holo build, page assembly, all 281 tests, about 13 minutes).
  • Before this merges: once main is green I will run the train from this branch by hand once, so the new step is proven on a runner before main's train depends on it.

Generated by Claude Code

@afflom afflom changed the title One file per guest and per guest step; the train says why it waits One file per guest; the train runs the browser acceptances and says why it waits Nov 16, 2026
#145 more than doubled labeling-ms-per-100 (about 100 to about 240, limit
250) and merged on one lucky sample. A cold start happens once in a process,
so the test times two more in processes of their own and holds the slowest
to the threshold. The quality record says so.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3LuNYtPDBa4FbSEQUZ
…-jbqzew

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3LuNYtPDBa4FbSEQUZ
…-jbqzew

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3LuNYtPDBa4FbSEQUZ
@github-actions

Copy link
Copy Markdown
Contributor

Merge train: f2303ee: not merged, because it conflicts with main (7871266) and the pull requests the train takes before it. Fix it on this branch (merge main first) and the train tests it again.


Generated by Claude Code

afflom commented Nov 16, 2026

Copy link
Copy Markdown
Owner Author

page / gates on 2d9b5db: 278 of 281 passed, two failed, neither in files this pull request changes.

Main itself failed its last finished deploy on 7871266 with QR-30 journeys-broken, which is the class the train step in this pull request closes. Merged main again (85954b6); checks are running.


Generated by Claude Code

@github-actions
github-actions Bot merged commit ee7a2d8 into main Nov 16, 2026
7 checks passed
@github-actions

Copy link
Copy Markdown
Contributor

Live: this change is on the published page, as main ee7a2d81c6b5d2146f7a4199bcbfe0aa2c8b6c85.

The publish job opened the published page in a fresh Chromium and its live checks passed (PG-01, PD-01, PD-02): https://github.com/afflom/me/actions/runs/43816464783

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.

2 participants