DrudgeCAS / gristmill / 32003292152

92%

push

github

fix: Verify a result whatever its external symbols are called

`verify_eval_seq` subtracted the produced definition from the reference as
the two stood.  The optimization canonicalizes the external indices of a
result onto the leading dummies of the range, so they need not be the symbols
the caller wrote, and the subtraction read that renaming as a difference.  A
correct answer was rejected for every choice of external symbols but the
canonical one, with a message naming the result and giving no hint that the
indices were the reason.

Line the two up first, by instantiating the produced definition at the
reference's indices.  That goes through the definition's own indexing rather
than a bare substitution, so the summations inside cannot be captured.  This
is not hypothetical: for a target written over `c, d` the optimization
produces `res[a, b] = u[a, b] + sum_{c} X[a, c]*tau^0[c, b]`, and
instantiating it at `c, d` correctly renames that summation to `a`.

Closes #49

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

5 of 6 new or added lines in 1 file covered. (83.33%)

1859 of 2030 relevant lines covered (91.58%)

0.92 hits per line