Backtestspydanticrun-20261027-165153#13687
populate_by_name (and implicit validate_by_name when validate_by_alias=False) is ignored by @validate_call
Better
Our patch was judged better than the merged fix.
In this judgment, Patch B = our agent, Patch A = the merged fix.
Both patches fix the root cause the same way in effect: the arguments/arguments-v3 schemas stop embedding the raw `validate_by_name` config key and instead embed the value after the `populate_by_name` and `validate_by_alias=False` compatibility mapping. All eight independent tests, covering both the `populate_by_name` case and the `validate_by_alias=False` variant, fail on base and pass on both patches, so correctness on the issue's scenarios is equal.
Tests separate them. Patch B adds tests for both variants the issue describes (`populate_by_name=True` and `validate_by_alias=False`), parametrized across both `arguments` and `arguments-v3` schema types in the experimental module, plus a `validate_call` test that also checks alias behaviour per config and a negative `populate_by_name=False` test. Patch A only tests `populate_by_name=True`, only for `arguments-v3` in the experimental test, and never exercises the 'worse variant' (`validate_by_alias=False`) that the issue explicitly calls out. A also includes unrelated `dict(...)` to `{...}` churn in three existing tests.
On craft, A has a point in its favour: it reuses `core_config()` as the single source of truth rather than duplicating the mapping logic, whereas B introduces `effective_validate_by_name`, which re-implements the compatibility rules and could drift from `core_config()`. B also widens scope by routing model and dataclass signature generation through the new property. That is a plausible adjacent fix (restoring pre-2.11 signature behaviour under `populate_by_name`) but it is untested and changes behaviour outside the issue's ask, so it earns no credit and carries mild risk.
Weighing in order: correctness is a wash, B's tests are materially stronger and directly cover the second reported failure mode, and the scope/craft trade-offs roughly offset (A's test churn and missing coverage versus B's duplication and untested signature change). B is the better patch.
The judge saw only the two patches, labelled A and B in random order, with issue links and version numbers removed. The reasoning above is shown verbatim.
Solve
11m 55s
Wall-clock time in the container
Full test suite
No new failures
1,079 passed · 36 skipped · 1,115 total
Contamination audit
Clean
No sign the agent saw the real fix
Patch size
62vs 39 merged
Changed lines, ours vs merged