The facts name the issue the Factory opened a pull request for, a fact the schema does not require (#823, first of two)
A pull request's facts now say which issue the Factory opened it for, or that it opened it for none. Nothing reads the fact yet. First of two for "The author Role's fix is admitted only on a pull request the Factory opened: the reach of fix-findings (slice 2, S9b)" (#823); it does not close it.
Why two. The whole change came to 63,281 characters as the Kernel counts a diff, over the review budget of 60,000, so it is split as the Factory Builder ruled on 2026-11-13: the fact first, with no record, then the Kernel's rule under record 0216.
Model-Lens inputs touched: none. Protected paths touched: none. The change goes by the Factory's route.
Size. 18,192 characters and 312 lines by git diff; 20,190 characters as the Kernel counts a diff, with its lines numbered.
What changes
| Where | What |
|---|---|
schemas/facts.schema.json | opened_for: the number of the issue the Factory opened the pull request for, a whole number from 1, or null |
adapters/acceptance.py | opened_for gives what bound gives: the issue the branch factory-work/issue-N-K names, only where the App of the Role that opens pull requests authored the pull request on a branch of that shape in the repo itself, and every opening Run was read and none names it for another issue, or twice. Anything else is null, and what bound cannot tell is null too, never its False |
adapters/facts.py | Every snapshot holds the fact |
examples/ | The example holds it; the counter-example is issue 0 |
tests/support.py | The tests' pull request is opened for no issue |
The fact is the one the schema does not require. A fact newly required needs a row in harness/eval/__init__.py for the held-out set, which predates it (decision 0105), and that file is a model Lens's input, approved only by a person's hand. Left optional, facts made before it are read as they stand, and the port's test holds that every snapshot gives it. The Factory Builder ruled it so; requiring it later is mechanical.
Where the binding cannot tell, complete.acceptance is already false from the same read, so the Kernel defers.
Today no Role opens pull requests, so the fact is null in every snapshot.
What is not held here
- Nothing reads the fact yet, and the simulator does not give it. The simulator's two lines, a protected edit the Factory Builder approved, were in this half at first; three readers found that no test could hold them here, so they moved to the second pull request, beside the rule that reads the fact and the mutant that holds them. "The simulator has a pull request the Factory did not open wait for its author, before author_role is named (slice 2, after S9b)" (#824) then has the simulator tell the Factory's pull requests from the rest.
- Until the writer of "Acceptance: a pull request the Factory opens is bound to its issue by the Run that opened it" (#342) lands,
boundtakes the branch alone where no opening Run names the pull request. Decision 0097 tightens that with the writer.
Checks it adds
tests/test_adapters.py::test_the_facts_name_the_issue_the_factory_opened_a_pull_request_for: the issue the branch names, of two branches, with a Run naming it or none yet; none for another account's pull request or while no Role opens; none, and neverFalse, for each thing the binding cannot tell.tests/test_adapters.py, two snapshot tests: every snapshot holds the fact, asked of the pull request as first read with the opener and the opening Runs' records.tests/test_tools.py::test_the_issue_a_pull_request_was_opened_for_is_a_number_from_1_or_null_or_left_out: what the schema takes and refuses.- Ten mutants, each killed by the test it names.
Checks
| Check | Result |
|---|---|
make checks | OK at f54932b |
tests.test_adapters, tests.test_tools.Examples, tests.test_decide, tests.test_transitions, tests.test_controller, tests.test_evals | OK |
Mutants of every file the change touches, and of kernel/decide.py, run alone on an export | 310 of 310 killed at d13a973, the first eight new ones among them; the two added since were applied by hand at f54932b, and each is killed |
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) | d13a973 | not run (quota), owed | ā | ā | ā | ā |
| muse | d13a973 | none | 0 | 1 | 0 | 0 |
| claude-review | d13a973 | none | 0 | 0 | 1 | 2 |
| claude-security | d13a973 | none | 0 | 0 | 0 | 1 |
| astra | ba78a0d | none | 0 | 1 | 0 | 0 |
Astra read the head after the others' fixes. Its finding was a gap in a test, so no one read again. Before the split, Muse, the Claude review and the security pass also read this content as part of the whole change, at c21829f on the second pull request's branch; what they found there is in that pull request's body.
| Finding | What | Disposition |
|---|---|---|
| muse #1 (major), claude-review #1 (minor), security #1 (info) | The simulator gave every pull request an issue, the Maintainer's and the dependency bot's among them, and nothing in this half read the fact, so no test held the two lines | fixed here by taking them out: ba78a0d. They are the second pull request's, with their mutant; that the simulator's worlds are all the Factory's until #824 is the Factory Builder's ruling |
| claude-review #2 (info) | The simulator's comment claimed more than its worlds hold, and named this issue where #824 mends it | gone from this half with the lines; the second pull request's comment says they stand in until #824 |
| claude-review #3 (info) | The schema's description said an absent fact is read as null, which nothing here does | fixed: the sentence is gone; the second pull request's record says how the Kernel reads absence |
| astra #1 (major) | Every case gave issue 36, so a port that gave one number for every pull request the Factory opened would pass | fixed in f54932b: a second branch, of another issue, and two mutants |
š¤ Generated with Claude Code