ponder-lab / Hybridize-Functions-Refactoring / #2790

84%

main: 84%

push

github

Issue the already-optimal verdict only where it is true (#871)

Implements the emission map settled on [#865](https://github.com/ponder-lab/Hybridize-Functions-Refactoring/issues/865): a considered function either gains a transformation or carries a precondition failure, where failure includes the benign already-optimal verdict that HAS_NO_PRIMITIVE_PARAMETERS expresses for a correctly-hybrid function. The one defective issuance was the axes-blocked reconfiguration: such a function is not already optimal, since an improvement existed and was withheld as unwritable, so the reconfiguration gate splits and that case now reports the unresolved axis alone. The adjudication path and the pre-inference terminal keep the verdict unchanged.

Both variants of the corrected arm are pinned. testReconfigureBlockedAxes is the faithful reproducer, matching the corpus shape: the FUT carries an existing inline input_signature, and the inferred wildcard comes from flow over-approximation via a zero-trip loop's head phi, the one configuration where a supplied signature coexists with a running program (probe-verified: the parameter types as both shapes while only the conforming one flows; the analysis constant-folds reshape's -1, enumerates dataset batch shapes including the remainder batch, and prunes infeasible constant branches). testReconfigureBlockedAxesAdd covers the signature-less add-path variant of the same arm. Two-sided validation: with the fix the full reactor is green; with the fix reverted, exactly these two tests fail on the assertion that the already-optimal verdict must not co-issue.


Claude-Session: https://claude.ai/code/session_01AM8pvSeP5UqDyyTn9h5E2P

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

6 of 6 new or added lines in 1 file covered. (100.0%)

3176 of 3782 relevant lines covered (83.98%)

0.84 hits per line