_arbiter.py 397 -> 246 + _claims 241
_escape.py 300 -> 230 + _escape_types 82
_claims holds the claim document and everything that decides whether one is
still owned: taking the advisory lock IS the liveness proof, so reading a
claim and reaping a dead one are the same operation and belong together.
_arbiter keeps the policy — who outranks whom, and whether this app may arm.
An AST pass over every module in the package (names loaded vs. names defined
or imported) caught two things a green suite would not have:
- _try_lock and fcntl moved with the locking code, so the tests' patch
targets had to move too;
- OutputSource was left behind by the earlier _outputs split and only
resolved because of `from __future__ import annotations` — a latent
NameError the moment anything evaluated that annotation.
gatelock is now at 0 files over 250 lines: 389 passed, 100% branch coverage,
the same 58 public names, and palette_check still at 0 failing checks.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CFSGeWxHPY5rHUGrPqhcmG
Krzysztof kuhy Rudnicki
f5f9db21dc