johanjoensson / impurityModel / 34968324040

94%

Build Type

push

github

Committed by johanjoensson

Commit Message

Resync the generated config doc; make the per-color test derive its own geometry

Two gate failures from the re-measured fanout constant.

doc/configuration.md is generated from the knob registry and drifted when
GF_APPLY_ROW_CHUNKS's docstring changed. Regenerated the tables, keeping
the hand-written preamble (the regen replaces everything from the first
generated heading onward).

test_run_units_distributed_sizes_each_color_on_its_own_rank_count_mpi
hard-coded inherited_cap=200_000, chosen when the fanout term was ~4.7x
larger. Re-measuring moved the window out from under it and the test
stopped discriminating -- it failed on "the 1-rank color must be
tightened" because nothing was tightened any more. It now derives its own
operating point from max_unit_dets_within_budget at 1 and 2 ranks and
asserts the window is non-empty, so a future change to the constant
either keeps the test meaningful or fails it loudly with an explanation
rather than silently voiding it. Also pins GF_APPLY_ROW_CHUNKS, which the
estimate now reads.

Verified it still catches what it was written for: against the pre-fix
gf_units.py it fails with "the 2-rank color must keep the untightened
job-wide cap".

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SJr65wo9HcVx4z7C6fHVFZ

Coverage Stats

12 of 12 new or added lines in 1 file covered. (100.0%)

57 existing lines in 5 files now uncovered.

34201 of 37368 relevant lines covered (91.52%)

0.92 hits per line