Datasets:
id int64 1 10 | population_shift stringclasses 3
values | protocol_deviation_rate stringclasses 3
values | site_variance_level stringclasses 3
values | endpoint_fragility stringclasses 3
values | signal stringclasses 10
values | label int64 0 2 |
|---|---|---|---|---|---|---|
1 | none | low | low | low | Population stable with low deviations and uniform sites. Endpoint robust | 0 |
2 | minor | medium | low | low | Small population shift with manageable deviations. Endpoint remains robust | 1 |
3 | minor | medium | medium | medium | Population shift plus site drift creates early endpoint fragility | 1 |
4 | major | high | medium | high | Major population shift with high deviations undermines endpoint validity | 2 |
5 | none | high | high | high | High deviations and site variance fragment endpoint despite stable population | 2 |
6 | major | low | medium | medium | Population shifts but deviations low. Endpoint pressured by site effects | 1 |
7 | major | medium | high | high | All nodes misalign. Endpoint becomes fragile and unstable | 2 |
8 | minor | low | high | medium | Population mostly stable but site variance drives fragility | 1 |
9 | none | medium | high | high | Site variance dominates and endpoint credibility breaks | 2 |
10 | minor | high | low | medium | Deviations rising with mild population shift creates endpoint fragility | 2 |
Clinical Quad Population Shift Protocol Deviation Site Variance Endpoint Fragility v0.2
What this dataset does
It tests whether a model can detect when clinical trial endpoints lose credibility under quad coupling.
Quad nodes
- population_shift
- protocol_deviation_rate
- site_variance_level
- endpoint_fragility
Labels
0 coherent
- Stable population
- Low deviations
- Low site variance
- Endpoint robust
1 tradeoff
- Some drift exists
- Endpoint still usable
- Risk is present but not terminal
2 collapse
- Endpoint credibility breaks
- Drift becomes structural
- Site or deviation pressure overwhelms interpretation
What changed in v0.2
- Version bumped to make scorer changes visible
- New scorer with validation and id-based matching
- Confusion and error sampling added
- Risk score and rule_pred diagnostics added
Files
data/train.csv
data/test.csv
scorer.py
Run scoring
python scorer.py --preds_csv predictions.csv --gold_csv data/test.csv
- Downloads last month
- 15