scenario_id string | scenario_text string | claim string | label int64 |
|---|---|---|---|
train_001 | The report is based on 4 survey responses from a team of 300 people. | High confidence is justified. | 0 |
train_002 | Three independent sensors recorded the same pressure spike within the same minute. | High confidence is justified. | 1 |
train_003 | One customer said the checkout page was confusing. | High confidence that the checkout is confusing for most users is justified. | 0 |
train_004 | The same error appeared in logs across 800 failed requests after deployment. | High confidence that the error is real is justified. | 1 |
train_005 | A small uncontrolled trial showed improvement in 6 of 8 patients. | High confidence that the treatment works is justified. | 0 |
train_006 | The contract states the fee is due within 14 days, and the invoice was received on 1 June. | High confidence about the payment deadline is justified. | 1 |
train_007 | Two witnesses gave conflicting accounts of who entered the room first. | High confidence about who entered first is justified. | 0 |
train_008 | The monitoring system recorded database downtime from 10:05 to 10:19. | High confidence that downtime occurred is justified. | 1 |
train_009 | Sales rose after a new advert launched, but price discounts also began that week. | High confidence that the advert caused the rise is justified. | 0 |
train_010 | The lab test was repeated three times and returned the same abnormal result each time. | High confidence that the abnormal result is present is justified. | 1 |
train_011 | A user stopped using the app after a redesign, with no exit feedback. | High confidence that the redesign caused the user to leave is justified. | 0 |
train_012 | The system accepted 0 successful logins during a 20-minute outage window. | High confidence that login failure occurred during the outage is justified. | 1 |
train_013 | The team missed a deadline after two members were off sick and a supplier was late. | High confidence that one single cause explains the delay is justified. | 0 |
train_014 | The receipt shows payment was made at 14:32 on 7 March. | High confidence that payment was made at that time is justified. | 1 |
train_015 | The model produced one correct answer on one test example. | High confidence that the model is reliable is justified. | 0 |
train_016 | A temperature probe and backup probe both recorded 39.1°C within 10 seconds. | High confidence that the temperature reading is accurate is justified. | 1 |
train_017 | A social media post received more likes than usual after the caption changed. | High confidence that the caption caused the increase is justified. | 0 |
train_018 | The delivery tracker, driver log, and customer confirmation all show delivery at 16:10. | High confidence that delivery occurred around 16:10 is justified. | 1 |
train_019 | A single analyst estimated future demand without showing assumptions or data. | High confidence in the demand forecast is justified. | 0 |
train_020 | The report includes raw counts, method notes, and matching totals across tables. | High confidence in the reported totals is justified. | 1 |
What this dataset does
This dataset tests whether a model can judge when high confidence is justified.
The task is simple:
Given a scenario and a confidence claim, predict whether the evidence supports high confidence.
Core stability idea
Reasoning fails when confidence rises faster than evidence quality.
This dataset targets that failure mode.
High confidence is justified when evidence is direct, repeated, independent, or clearly documented.
High confidence is not justified when evidence is weak, single-source, conflicted, uncontrolled, or causal without support.
Prediction target
Binary label:
- 1 = high confidence is justified
- 0 = high confidence is not justified
Row structure
Each row contains:
- scenario_id
- scenario_text
- claim
- label
Files
- data/train.csv
- data/test.csv
- scorer.py
- README.md
Evaluation
Create a predictions CSV with:
scenario_id,prediction
test_001,1
test_002,0
Run:
python scorer.py --predictions predictions.csv --truth data/test.csv
The scorer reports:
accuracy
precision
recall
f1
confusion matrix
Structural Note
This dataset is intentionally small.
Its purpose is to test whether a model can keep confidence proportional to evidence.
The hidden value is in detecting when repeated evidence, independent confirmation, conflict, sample size, and causal ambiguity change the justified confidence level.
License
MIT
- Downloads last month
- 22