Backtestspydanticrun-20261022-024030#13687
populate_by_name (and implicit validate_by_name when validate_by_alias=False) is ignored by @validate_call
Our patch was judged better than the merged fix.
In this judgment, Patch A = our agent, Patch B = the merged fix.
Both patches fix the root cause identified in the issue: the raw `validate_by_name` config key being embedded into `arguments`/`arguments-v3` core schemas, bypassing the compatibility mapping done in `core_config()`. The independent tests confirm both make all seven scenarios pass, including the `validate_by_alias=False` variant.
Patch B's approach of reading the value from `core_config(title=None)` is arguably more elegant because it reuses the single source of truth rather than duplicating the mapping logic. Patch A duplicates that logic in a new `effective_validate_by_name` property, with a docstring explaining it mirrors `core_config()`, which carries a small drift risk but is clearly documented.
Where A pulls ahead is tests and scope discipline. The issue explicitly calls out two behaviours: `populate_by_name=True` and the "worse variant" `validate_by_alias=False`. Patch A adds regression tests for both in `test_validate_call.py` (including asserting the alias is correctly rejected under `validate_by_alias=False`), and parametrizes the experimental arguments-schema test across both `arguments` and `arguments-v3` schema types and all three config spellings. Patch B only tests `populate_by_name` on `validate_call` and only the `arguments-v3` schema type, leaving the `validate_by_alias=False` variant and the legacy `arguments` schema path untested. Patch B also includes unrelated churn, rewriting `dict(...)` to `{...}` in three existing tests, which is noise in a bug-fix diff.
Patch A's extension to model and dataclass signature generation is beyond the issue's ask and earns no credit on its own, but it is low-risk, consistent with the fix, and covered by a new test, so it does not count against it.
On balance A is the better fix due to materially stronger test coverage of the behaviours the issue names, with no unrelated churn.
Four judges ruled on this ticket in parallel, and the verdict comes from counting their votes. Each saw the issue and the two patches, labelled A and B in random order (two of them saw ours as A), with issue links and version numbers removed. Where tests could check the fix, they also saw the results of tests a separate agent wrote for the issue, run on both patches. None was told which patch was ours. The reasoning above is one judge's, shown verbatim.
Solve
6m 18s
Wall-clock time in the container
Full test suite
No new failures
1,083 passed · 36 skipped · 1,119 total
Contamination audit
Clean
No sign the agent saw the real fix
Patch size
75vs 39 merged
Changed lines, ours vs merged