Real fairness, calibration, and baseline-gap artifacts from the canonical run - #74

Open

duckyquang wants to merge 15 commits into

mainfrom

feat/eval-artifacts

Commits on Aug 12, 2026

Record pooled rerun results and cohort sensitivity
duckyquang
committed
544229a
Fit the EF<=40 Platt scaler on val predictions instead of in-sample
duckyquang
committed
5a40bcb
Stratify fairness from the joined manifest, per missing-modality condition
duckyquang
committed
ec4c7cd
Add the in-cohort fused vs solo comparison behind the E10 diagnosis
duckyquang
committed
0c0eabf
Let the failure harness read the joined manifest
duckyquang
committed
8d228bf
Record measured fairness, calibration, and gap-diagnosis results in the docs
duckyquang
committed
035a258

Commits on Aug 13, 2026

Commits on Aug 27, 2026

Read the ECG-dropped condition as a below-chance failure
duckyquang
committed
634c2ca
Record how many bootstrap replicates each AUROC interval rests on
duckyquang
committed
d365b07
Name the model behind every failure-analysis artifact
duckyquang
committed
c8d0ec5
Give every artifact one canonical location
duckyquang
committed
37b7d93
Stop two artifacts from degrading quietly
duckyquang
committed
b849306

Commits on Aug 29, 2026

Regenerate canonical result artifacts
kevzho
committed
78dd9ff

Commits on Aug 30, 2026

Match the docs to the regenerated artifacts
duckyquang
committed
251225a