Drop the two multi-section CSVs from the viewer configs

Declare dataset configs and frame the card around both versions

the anchor hazard is recoverable by structure, not by target reformulation

carry cohort labels in the verification output

HEWB v2.1: wire in the CLH group, retract the recoverability claim, add the anchor cohort

Correct the event-floor claim: the failure is era transfer, and add calibration

Point the v1.4 card at the v2 reformulation that repairs it

Add HEWB v2 dataset card

Add HEWB v2 survival reformulation

Add HEWB v2 survival reformulation

Add HEWB v2 survival reformulation

Add HEWB v2 survival reformulation

Add approval_age to the per-cutoff baseline table

Correct the baseline: approval age alone beats the learned model

Document right-censoring: recent cutoffs understate the method by construction

Correct population caveat: extras are superseded records, positive class unaffected

Correct the population caveat: export matches the register public search

Extend the mapped boundary to all four gated domains

HEWB v1.4: benchmark definition, result tables, robustness evidence

initial commit