With two model Lenses, the review admitted is the one the evidence names (#855)

Merged
DO
3d ago
+24-8

Part of #855: its third point, which the review of #859 found beside it.

What. With two model Lenses required, kernel/decide.py's _wanted_run admitted the first whatever Lens awaiting_review named. So once the first had reviewed, an item whose second Lens was asked would run the first again, to the Run ceiling.

  • It now admits the review of the Lens the evidence names. Derive names one only from the model Lenses the configuration requires, so no guard is needed past that.
  • The default configuration requires one model Lens, so nothing changes live.
  • #855's other two points, the simulator and the explorer pushing a head that leaves the diff as it was, follow in their own pull request.

Record. None, unless the Factory Builder rules one: the rule, admit the review the State awaits, is unchanged, and this mends its reading for a second Lens.

Tests and mutants.

  • tests/test_decide.py::test_with_two_model_lenses_the_one_awaited_is_the_one_admitted failed first, as the issue says: the evidence named the second Lens, and pr-review was admitted again. It passes now.
  • The mutant "the first model Lens is admitted whatever the evidence names" is new, and "the act ceiling stops a Run" follows the new call.
  • All 105 mutants of kernel/decide.py are killed. tests.test_decide, tests.test_transitions and tests.test_issue pass, and so does make checks.

Readers (head ff5043a1):

ReaderFindingsDisposition
Claude review1 minorTaken in f1b0ee6b: the guard on a Lens the configuration does not require could not be reached through go(), and was tested only by a direct call to the private _wanted_run. Both go, and the test calls the Kernel through go() alone
Claude security1 minorThe same, taken
Carbon (gemini-4-carbon-high)none—
Astra (gpt-6-astra)none—
Kimi (kimi-k3.1-pro)Kimi: 0 findings, 0 taken—

šŸ¤– Generated with Claude Code