Rulesets: main counts the machine user's approval, as the switch of 2026-11-11 left it (#198, record 0191)
This brings config/rulesets.json to the live main ruleset after the switch of "The ruleset switch: main requires the machine user's approval of the most recent commit, and slice 1's event is recorded" (#198), and records the switch as decision 0191. Part of #198; it does not close it.
What GitHub holds
On the Maintainer's choice, "Switch; Factory approves", as the Factory Builder relayed it, the Factory Builder changed ruleset 24886320 through the Maintainer's login at 02:18:03Z on 2026-11-11. Four fields changed and no other:
| Field | Before | After |
|---|---|---|
required_approving_review_count | 0 | 1 |
require_code_owner_review | false | true |
require_last_push_approval | false | true |
dismiss_stale_reviews_on_push | false | true |
bypass_actors stays []. I read the live ruleset after the switch: updated_at is 2026-11-10T20:18:03-06:00 and the four fields read as above. sh tools/rulesets_drift.sh reported those four fields before this change and reports no drift after it.
What was shown before the switch
"AGENTS.md: say what is built; drop the model claim no check holds" (#688), as GitHub records it on 2026-11-11:
| When | Event | Actor |
|---|---|---|
| 00:25:08Z | Approved at head 2a60aef6, by the Factory's approving act | donpetry-bot |
| 01:07:14Z | Added to the merge queue | don-petry |
| 01:21:38Z | Merged | don-petry |
The Factory's queue step did not enqueue it: the queue workflow has no run on this repo. So the Factory's review and approval are shown, and its own enqueue is not. The record says so.
What changes
config/rulesets.json: the four fields, andread_on2026-11-11.tests/test_tools.py:test_main_counts_the_machine_users_approval_as_the_switch_left_itholds the committed record to the switch: the four fields, no required reviewer, no bypass, the ruleset's id, and a read date of the switch day or after.tools/check_rulesets.pyholds how an approval is counted, and still lets through amainthat counts none, so without this test the record could go back with nothing failing.- The Rulesets tests'
run_onnow carries the repo's own.github/CODEOWNERS. The check reads the code owners oncemaincounts an approval, andtest_this_repos_rulesets_agree_with_its_lensesfailed without them.
tools/mutants.py: six mutants ofconfig/rulesets.json, each killed by the new test: each field back, the four together, and the read date back.- Record 0191: the switch, what a merge now needs, what is not shown yet, what it overtakes in records 0018, 0020, 0022, 0026 and 0087, and the undo.
What it does not do
tools/check_rulesets.pyis unchanged: it does not refuse amainthat counts no approval. A check that refused it would refuse the undo's record too.config/AGENTS.mdis unchanged: its row on howmaincounts an approval already reads in the present, and the file is at its 40-line limit.- A comment in
sim/world.pystill speaks of the switch as to come. That file is protected, so it is left for a ruling. - The diff is 12,384 characters and 131 lines, under the Factory's review budget.
- How a fix merges while the Release on
nextis broken (#198's third box) is not shown, and the record says so.
Checks
| Check | Result |
|---|---|
make checks | OK |
tests.test_tools.Rulesets, RulesetsDrift | OK, 25 tests |
Mutants of config/rulesets.json, tools/check_rulesets.py and tools/rulesets_drift.py, run alone on an export | 56 of 56 killed, at 924fb48; the head differs from it in the record's words alone |
sh tools/rulesets_drift.sh | no drift |
Full make verify | Not run locally, by the Maintainer's standing change of 2026-11-09; CI's verify is the answer |
| Reviewer | Head | Unread parts | Blocking | Major | Minor | Info |
|---|---|---|---|---|---|---|
| agy (Carbon) | d207d86 | none | 0 | 0 | 0 | 0 |
| muse | d207d86 | none | 1 | 0 | 0 | 0 |
| astra | a04c419 | none | 0 | 0 | 0 | 0 |
| claude-review | d207d86 | none | 0 | 1 | 1 | 2 |
| claude-security | d207d86 | none | 0 | 1 | 0 | 0 |
The head is 8fc5492. It differs from a04c419, which Astra read clean, in the record alone: its number, and the three rows a review comment on this pull request led to (below). a04c419 differs from d207d86 in the record's words, the test's read-date case and one mutant.
| Finding | What | Disposition |
|---|---|---|
| muse@d207d86 #1 (blocking) | The record was numbered one past a number main did not hold, so make checks failed | fixed: the record is 0191. The Factory Builder ruled that numbers go in order of readiness; the Pause record, whose pull request was split, takes the next |
| claude-security@d207d86 #1 (major), claude-review@d207d86 #1 (major) | The record said the Factory's queue step enqueued #688 with no person's act between; GitHub names don-petry as the enqueuer, and the queue workflow has never run | fixed: the row says who enqueued and that the Factory's enqueue is not shown |
| claude-review@d207d86 #2 (minor) | Records 0018, 0020, 0022, 0026 and 0087 still say main requires no review, and 0191 named only 0092 | fixed: the row "What it overtakes" |
| claude-review@d207d86 #3 (info) | Nothing held read_on | fixed: the test holds it, with a mutant |
cubic on 924fb48, three threads | The record said every merge needs the machine user's approval, where the ruleset asks for a code owner's; the Decision table told the dry run's history; and numbers were said to need spelling out | the first two fixed at 8fc5492: the row names a code owner's approval and when it is the machine user's, and the history is in the Why. The third is no change: the rule is the title and then the number after the word |
| claude-review@d207d86 #4 (info) | The reviewer did not read the live ruleset or the time of the switch | answered above: I read both |
š¤ Generated with Claude Code