Fix BERNN probability and batch mapping

Fix real submission worker multiprocessing

Prepare head sweep tuned defaults

Expose BERNN class triplet option

Clarify BERNN internal monitor MCC

Persist Real leaderboard to HF Dataset

Seed Real leaderboard baseline examples

Score Real leaderboard by conservative MCC

Pass batch metadata to submission predict

Log official test MCC after ensemble

Use bernn 0.6.24

Forward available test labels to fit

Pass external validation splits into BERNN fit

Expose validation labels to fit

Use fixed test set across CV folds

Separate validation and test CV splits

Switch model submissions to fit API

Expose validation split API to submissions

Expose external test groups in BERNN submissions

Externalize BERNN validation splits

Provide default BERNN CONFIG for submissions

Use BERNN 0.6.21 release by default

Pin BERNN sklearn-style fit release

Use sklearn-style BERNN fit workflow

Keep BERNN holdout test split external

Recover BERNN hparam search pipeline on HF branch

src/bernn_tuned_defaults.json

Remove repeating training traceback watchdog

removing log_metrics in tensorboard

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

adding more debug

Update BERNN leaderboard publication

Update BERNN leaderboard publication

BERNN model selection: parameterized baseline + presets + UI builder

bernn baseline: set config.dataset (bernn reads self.args.dataset for log dir; TrainingConfig lacks it)

Sync app to local version: add database/baselines/dataset_info/version_manager modules, update app.py & code_challenge (bernn n_repeats>=3 fix), expand requirements

Add combatR example, store user code, support click-to-fill + public sharing

Add secure code-run real leaderboard and keep CSV practice tab

Deploy MassBench batch effects leaderboard