The scoring mode every entry point ships. Named rather than repeated as a
literal in the CLI arg, three pyo3 signatures and the Python layers: those
copies are what would make a future default change land in some entry points
and not others.
v5 re-calibration under per-file admission (4x3 grid, v1 calibration
manifest, held-out validated): winner (tau, cbf) = (0.05, 0.4) at
min(per_benchmark file_recall) = 0.6815 vs 0.6684 for the old point.
Confirmed on the test splits (500x3): file_recall parity with the old
default, contextbench file_precision +0.041 (CI excludes zero), and on
dcbench-372 paired nontrivial recall within noise at -4% tokens. The
weak stop only ships together with the admission gate (#65): without it,
tau=0.05 re-admits the diffuse tail the gate exists to block.
Hard ceiling for a single git cat-file blob read. Sized with headroom
above max_changed_file_size (5 MB) so legitimate large changed files still
load, while a pathological multi-hundred-MB blob is drained and rejected
instead of allocated up front.