The simulator and the explorer push a head that leaves the diff as it was, and the world says a thread dropped (#855)

Merged
DO
3d ago
+420-18

Closes #855: its first two points. The third, the two-Lens admission, merged as #882.

What. The seeded simulation and the transitions explorer now make the two things the fix for #833 added rules for, which only unit tests held before.

  • sim/world.py: a person's push now and then keeps the content. The new GitHub rule a_push_can_leave_the_diff_as_it_was is grounded by trials/P-29.base.json; the statement says why P-29 stands for a push it did not make.
  • The world says a thread dropped only for its own Lens's review on the current content, and a push of new content clears every dropped word.
  • The check on the act takes a dropped thread as it takes an accepted one, and refuses each dropped thread the Kernel may not resolve.
  • I-11 yields only to a gating thread of the Lens's own that asks for the review: not accepted, and not dropped since the push.
  • tests/test_transitions.py: the explorer's push may keep the diff. From addressing it takes the item to verifying, and once the checks pass a gating Lens is asked again on the same content.
  • examples/facts.json and tests/support.py carry the dropped fact, with its counter-example.

Record. None: no Kernel rule changes.

Protected edits. sim/world.py and tests/test_simulator.py: pre-approved by the Maintainer 2026-11-08, ruled by the Factory Builder (assignment 2026-11-14).

Seeds pinned. AGAIN_SEED = 261 and DROPPED_SEED = 46.

Tests and mutants.

  • Eleven mutants are added, all killed at 40 seeds.
  • At the default 300 seeds, on this head: tests.test_simulator, tests.test_transitions, tests.test_adapters, tests.test_decide, tests.test_issue_simulator, tests.test_tools and tests.test_hooks ran 1282 tests with one failure, tests.test_tools.ReadConfinementTrial.test_the_stand_in_timeout_runs_what_gnu_timeout_runs. That one fails the same way on main on the desk's machine, whose /usr/bin/timeout is not GNU's; this change does not touch it. make checks passes.
  • Not held by the world: the Kernel mutant "a thread the Kernel is to resolve asks its review again" is not caught by the simulation, since no seed produces that case. tests/test_decide.py holds it.

Readers (read at 1847ad11, the same diff before a rebase onto main):

ReaderDisposition
Claude reviewIn; findings taken in 6124ca9 and faef2d3 (the dropped fact's counter-example)
Claude securityIn; findings taken in 6124ca9
CarbonIn; findings taken in 6124ca9
AstraIn; findings taken in 6124ca9
KimiKimi: 2 findings, 2 taken (both major, both real)

The per-reader counts other than Kimi's were kept on the machine the work moved from, and are not restated here from memory.

šŸ¤– Generated with Claude Code