scenario_id string | scenario_text string | claim string | label int64 |
|---|---|---|---|
train_001 | A delivery route was optimized for shortest distance, but drivers now miss time windows during peak traffic. | The system shows over-optimization. | 1 |
train_002 | A delivery route was optimized for shorter distance and still meets all time windows. | The system shows over-optimization. | 0 |
train_003 | A model was tuned to maximize benchmark score, but performance dropped on real user prompts. | The system shows over-optimization. | 1 |
train_004 | A model improved benchmark score and real user prompt performance also improved. | The system shows over-optimization. | 0 |
train_005 | A warehouse minimized staffing costs so tightly that one absence now causes shipment failure. | The system shows over-optimization. | 1 |
train_006 | A warehouse reduced staffing costs while preserving backup cover and shipment reliability. | The system shows over-optimization. | 0 |
train_007 | A clinic maximized appointment volume by removing buffer time, causing delays after one complex case. | The clinic shows over-optimization. | 1 |
train_008 | A clinic increased appointment volume while preserving buffer time for complex cases. | The clinic shows over-optimization. | 0 |
train_009 | A server configuration maximized throughput but left no capacity for traffic spikes. | The server shows over-optimization. | 1 |
train_010 | A server configuration improved throughput while preserving surge capacity. | The server shows over-optimization. | 0 |
train_011 | A project plan removed all slack to meet the earliest possible deadline. | The project shows over-optimization. | 1 |
train_012 | A project plan improved delivery time while preserving contingency buffer. | The project shows over-optimization. | 0 |
train_013 | A finance team minimized cash reserves to improve reported capital use, then struggled with one delayed payment. | The organization shows over-optimization. | 1 |
train_014 | A finance team improved capital use while preserving adequate reserves. | The organization shows over-optimization. | 0 |
train_015 | A machine was tuned for maximum output and now overheats under normal variation. | The machine shows over-optimization. | 1 |
train_016 | A machine increased output after cooling capacity was improved. | The machine shows over-optimization. | 0 |
train_017 | A support team reduced average call time by scripting every response, but complex cases now escalate more often. | The support system shows over-optimization. | 1 |
train_018 | A support team reduced average call time while complex-case resolution stayed stable. | The support system shows over-optimization. | 0 |
train_019 | A school optimized test scores by teaching only likely exam items, and broader understanding declined. | The school shows over-optimization. | 1 |
train_020 | A school improved test scores while broader assessment results stayed strong. | The school shows over-optimization. | 0 |
What this dataset does
This dataset tests whether a model can distinguish compensation from recovery.
The task is simple:
Given a scenario and a compensation claim, predict whether the claim is supported.
Core stability idea
A system can remain functional without actually recovering.
Compensation occurs when visible performance is maintained through temporary workarounds, manual effort, hidden strain, deferred cost, or masking.
Recovery occurs when the underlying instability is reduced and normal operation resumes without extra strain.
Prediction target
Binary label:
- 1 = the system is compensating rather than recovering
- 0 = the system is recovering or stable without compensation
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 detect maintained function that hides unresolved instability.
The hidden value is in detecting masking, workarounds, borrowed capacity, manual strain, and unresolved root causes.
License
MIT
- Downloads last month
- 14