Datasets:
id int64 1 10 | adherence stringclasses 2
values | dose_change stringclasses 4
values | ae_signal stringclasses 3
values | efficacy_signal stringclasses 4
values | narrative_move stringclasses 2
values | label int64 0 2 | signal stringclasses 10
values |
|---|---|---|---|---|---|---|---|
1 | stable | none | low | strong | aligned | 0 | Stable adherence with no dose change. Low AE. Strong efficacy. Narrative matches data. |
2 | stable | reduce | high | strong | aligned | 1 | Dose reduction driven by AE. Efficacy remains strong. Narrative honest about safety tradeoff. |
3 | stable | increase | low | plateau | aligned | 1 | Dose increase justified by plateau with low AE and stable adherence. Narrative acknowledges uncertainty. |
4 | unstable | none | low | weak | aligned | 1 | Weak efficacy plausibly driven by nonadherence. Narrative states adherence uncertainty before conclusions. |
5 | stable | hold | high | moderate | aligned | 1 | Dose hold for AE with partial benefit. Narrative consistent with protocol constraints. |
6 | unstable | increase | moderate | weak | spin | 2 | Dose increase used to mask nonadherence and weak efficacy. Narrative claims dose was main limiter. |
7 | stable | increase | high | weak | spin | 2 | Narrative reframes poor efficacy while escalating despite high AE burden. |
8 | unstable | reduce | high | weak | spin | 2 | Narrative blames disease while AE and adherence drive missed dosing and decline. |
9 | stable | reduce | moderate | weak | aligned | 1 | Reduction for tolerability with weak efficacy. Narrative acknowledges potential exposure loss. |
10 | stable | none | low | weak | spin | 2 | Narrative declares success despite weak efficacy with no operational explanation. |
Clinical Quad Adherence Dose Changes Adverse Events Efficacy Narrative v0.2
What this dataset does
It tests whether a model can classify coherence versus collapse in a clinical decision loop.
The loop couples four operational nodes plus narrative behavior.
Quad nodes
- adherence
- dose changes
- adverse events
- efficacy signal
Narrative node
- aligned means the story matches data and constraints
- spin means the story tries to justify a decision that the data does not support
Labels
0 coherent
1 tradeoff
2 collapse
What changed in v0.2
- Added explicit narrative_move field
- Tightened label meaning
- New scorer with validation, confusion, and error samples
- Order independent prediction matching by id
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
- 10