From /simplify codebase sweep (2026-06-10).
pipeline/phases/forward_pass.py:14,79-90: resolve_forward_batch_size first reads config.run.forward_batch_size -- but AppConfig (configs/schema.py:274-283) has no run group at all -- then falls back to data.forward_batch_size, then to a private constant 32, with isinstance(int)/> 0 validation at call time.
Cost: the run.* path is dead-or-undocumented config surface (struct-mode Hydra would reject it anyway); type/range rules live where Hydra validation cannot see them, so errors surface mid-run instead of at config load; the default 32 is invisible to schema docs and can drift from documentation.
Fix: one schema home -- DataConfig.forward_batch_size: int = 32 (or a real RunConfig if a run group is wanted) with validation in the schema layer; delete the dual-location resolver and the phantom path. Pre-1.0, clean break.
From /simplify codebase sweep (2026-06-10).
pipeline/phases/forward_pass.py:14,79-90:resolve_forward_batch_sizefirst readsconfig.run.forward_batch_size-- butAppConfig(configs/schema.py:274-283) has norungroup at all -- then falls back todata.forward_batch_size, then to a private constant 32, withisinstance(int)/> 0validation at call time.Cost: the
run.*path is dead-or-undocumented config surface (struct-mode Hydra would reject it anyway); type/range rules live where Hydra validation cannot see them, so errors surface mid-run instead of at config load; the default 32 is invisible to schema docs and can drift from documentation.Fix: one schema home --
DataConfig.forward_batch_size: int = 32(or a realRunConfigif a run group is wanted) with validation in the schema layer; delete the dual-location resolver and the phantom path. Pre-1.0, clean break.