Acceptance vs delinquency (Fritz BNPL samples)
Local MSE baseline vs challenger
Stored production score (autogluon_v25 in the tables)
Blue = production score already stored on pre_scored_users. Charcoal = the selected package scored locally on the same Fritz users.
Is this the same acceptance metric as Fritz acceptance_class?
Plotly versions of the Rank-1 almost_full error panels, scored on local test_v25 (n≈1978, not the Fritz 4k samples). Hover a point or bin for values. Hurdle modes follow the picker. Packages 18–19 have no cached scores.
Local MSE baseline
Challenger
Side-by-side Plotly error panels on the same local test_v25 holdout (n≈1978). Pick any two scored heads, including hurdle modes. Packages 18–19 have no cached scores. This tab does not use the left package list.