The first rollback and the move forward after it, the fixture alone: release/fixtures/rollback.v0-1-10.json (#197's ruling 9ā², first of three)
This is the fixture of the first rollback and of the move forward after it, alone: release/fixtures/rollback.v0-1-10.json. They are items 4 and 5 of ruling 9ā² on "Release: cut an immutable tag, move next, and roll back in one action" (#197), the last that story's fixture box waits on. It is the first of three; the box is ticked when the third merges. Part of #197; it does not close it.
Model-Lens inputs touched: none. One file is added, under release/fixtures/. It carries no record.
Size. 37,041 characters and 886 lines by git diff; 43,697 characters as the Kernel counts a diff, with its lines numbered, against a budget of 60,000.
Why it is alone, and what follows
With its test and mutants the change is over the budget, as slice 1's event was, so the fixture goes first. Two pull requests follow, each when the one before has merged:
| Pull request | Holds | Why it waits |
|---|---|---|
| This one | The rollback's fixture | ā |
| The second | release/fixtures/cut.v0-1-10.json, and the cut test told of a rollback made after a cut | The cut was read after the rollback and the move, so their two Deployments stand above the cut's own; the cut test reads them from this fixture |
| The third | tests/test_release.py::test_the_first_rollback_moved_next_back_within_its_time_and_a_move_brought_it_forward_again and 24 mutants of this fixture | It holds the v0.1.10 cut's own Deployment to that cut's fixture |
What GitHub answered
On 2026-11-12, while the repo's Pause stood, by two dispatches of release.yml:
| When | What | Read |
|---|---|---|
| 18:37:43Z | The repo's FACTORY_PAUSED set to true | The variable, still true when read after both runs |
| 22:12:32Z to 22:12:48Z | Release run 43317751252, rollback with no to: next from e719181a (factory/v0.1.10) to b61f0ecb (factory/v0.1.9) | The run, its one job, the job's log at the step; Deployment 7751076685 of b61f0ecb, success at 22:12:44Z and inactive at 22:13:30Z |
| 22:13:11Z to 22:13:34Z | Release run 43317854989, move to factory/v0.1.10: next back to e719181a | The run, its job, the log; Deployment 7751092797 of e719181a, success at 22:13:30Z |
| 17:32:49Z | Eval run 43267765670 on e719181a had ended in success, the eval workflow's one run on that commit | The run, and the workflow's runs on the commit |
- The rollback took 16 seconds from its dispatch to the end of its run, inside which the step moved
next, read it back and recorded the Deployment. FR-49 allows 15 minutes. - Every read was made after both runs. Where
nextstood between them is the rollback's own last line and the Deployment it recorded. - The tags are every
factory/tag as the step's port lists them:nexton factory/v0.1.10 again, and every Release where its cut left it. - The Deployments are the newest four of
factory-next, which is what was asked: the move's, the rollback's, and the two cuts' own below them.
The reads went through the Maintainer's login and were trimmed to the fields and headers rollback.refused.json and the cuts' fixtures keep. tools/redact_trial.py redacted them outside the repo, and only the clean output was read. It took out the names of steps[1] and steps[6] of each job, the token's mint and its post step. Of each log's inputs, the token's line, which GitHub masked, and the exchanges' folder are left out. removed says what was taken.
Checks
| Check | Result |
|---|---|
make checks | OK: tools/check_release.py holds each item to its own exchanges, and finds no credential and no header a fixture drops |
tests.test_release | OK, 34 tests; none reads this file yet |
| The third pull request's test and mutants, on the branch that holds all three | OK; 24 of 24 mutants of this fixture killed at 30bc383 |
| gitleaks on the fixture | No leaks |
Full make verify | Not run locally, by the Maintainer's standing change of 2026-11-09; CI's verify is the answer |
Reviews
The five readers read this file with its test, on the branch that then held both. The fixture was rebuilt once after their reads, to add the read of the Pause that Muse asked for; nothing else of it changed.
| Reviewer | Head read | Unread parts | Blocking | Major | Minor | Info |
|---|---|---|---|---|---|---|
| agy (Carbon) | 8105f59 | none of 2 | 0 | 0 | 0 | 0 |
| muse | 8105f59 | none | 0 | 2 | 1 | 0 |
| claude-review | 8105f59 | none | 0 | 0 | 3 | 4 |
| claude-security | 8105f59 | none | 0 | 0 | 3 | 0 |
| astra | d287b60 | none | 0 | 2 | 0 | 0 |
| Finding | What | Disposition |
|---|---|---|
| muse #1 | "Under a Pause" was held by no exchange | fixed in this file: it gains the read of the repo's FACTORY_PAUSED; the third pull request's test holds it |
| Every other finding | A claim the test did not hold: the runs' branch and commit, the other Releases, the reads' times, the Deployments' times, the 15-minute bound's own mutant, the v0.1.10 cut's Deployment | fixed in the third pull request, whose description lists each. None found the fixture wrong; the Claude review checked every run, Deployment, status and tag in it against GitHub live |
š¤ Generated with Claude Code