Skip to content

test: pin the #286 akm_slopes cell against fake convergence - #397

Merged
schroedk merged 1 commit into
mainfrom
test/fake-convergence-cell
Sep 23, 2026
Merged

schroedk merged 1 commit into
mainfrom
test/fake-convergence-cell

Conversation

@schroedk

Copy link
Copy Markdown
Collaborator

Stack 4/7 (test), based on main. Lands the #286/#290 regression cell before the exit rework in #388, so the rework visibly keeps it green.

  • tests/test_fake_convergence.py solves the akm_slopes shared_year cell (100k rows, mobility 0.05, tol 1e-12) single-threaded, where the recurrence collapse reproduces, and requires converged with every group mean of the demeaned response below 1e-6.
  • The Store and expose build_duration in Preconditioner #276 wheel reported this cell converged at a group-mean error of 18. Main passes it (1 passed, 4.2 s).

The shared_year cell (100k rows, mobility 0.05, tol 1e-12) once drove
the Schwarz factor near singular, and the collapsed LSMR recurrences
reported it converged at a group-mean error of 18 (#290, #276 wheel).
Solved single-threaded, where the collapse reproduces, it must converge
with every group mean of the demeaned response below 1e-6.
@schroedk
schroedk added this pull request to stack #398 September 23, 2026 15:06
@schroedk
schroedk merged commit 0a850a9 into main Sep 23, 2026
5 checks passed
@schroedk
schroedk deleted the test/fake-convergence-cell branch September 26, 2026 06:58
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant