scenario_id string | scenario_text string | claim string | label int64 |
|---|---|---|---|
train_001 | A server is overloaded. Option A adds more logging. Option B shifts traffic to backup capacity. Option C waits for demand to fall. | Option B is the best stabilizing intervention. | 1 |
train_002 | A patient has low blood pressure and signs of dehydration. Option A gives fluids. Option B increases paperwork. Option C delays review until tomorrow. | Option B is the best stabilizing intervention. | 0 |
train_003 | A project is delayed because no one owns the next milestone. Option A adds more meetings. Option B assigns a single accountable owner. Option C changes the logo. | Option B is the best stabilizing intervention. | 1 |
train_004 | A supply chain has low inventory and one supplier is late. Option A increases marketing demand. Option B activates a backup supplier. Option C ignores the delay. | Option A is the best stabilizing intervention. | 0 |
train_005 | A team is overloaded and errors are rising. Option A adds more urgent work. Option B pauses noncritical tasks. Option C removes all documentation. | Option B is the best stabilizing intervention. | 1 |
train_006 | A database is near storage capacity. Option A archives old data. Option B increases user notifications. Option C changes the dashboard colour. | Option C is the best stabilizing intervention. | 0 |
train_007 | A tenant reports recurring water ingress after rain. Option A paints over staining. Option B investigates external water entry. Option C asks the tenant to open windows only. | Option B is the best stabilizing intervention. | 1 |
train_008 | A payment queue is backing up. Option A disables monitoring. Option B increases queue processing capacity. Option C launches a new feature. | Option A is the best stabilizing intervention. | 0 |
train_009 | A model gives inconsistent answers. Option A checks outputs against source documents. Option B raises confidence wording. Option C hides uncertainty. | Option A is the best stabilizing intervention. | 1 |
train_010 | A machine overheats during peak load. Option A lowers load temporarily and checks cooling. Option B increases load. Option C ignores temperature alerts. | Option B is the best stabilizing intervention. | 0 |
train_011 | A clinic has rising wait times and unused admin capacity. Option A redirects admin support to triage. Option B adds more forms. Option C closes feedback channels. | Option A is the best stabilizing intervention. | 1 |
train_012 | A bridge inspection finds widening cracks. Option A reduces load and orders structural review. Option B increases traffic volume. Option C removes warning signs. | Option C is the best stabilizing intervention. | 0 |
train_013 | A school has repeated timetable clashes. Option A creates a single scheduling owner. Option B asks each teacher to solve it separately. Option C stops publishing timetables. | Option A is the best stabilizing intervention. | 1 |
train_014 | A warehouse has picking errors due to unclear item labels. Option A improves labels. Option B increases speed targets. Option C removes quality checks. | Option B is the best stabilizing intervention. | 0 |
train_015 | A customer support team has a growing backlog. Option A pauses low-priority channels. Option B adds more ticket categories without staffing. Option C deletes old tickets. | Option A is the best stabilizing intervention. | 1 |
train_016 | A network has repeated failures at one router. Option A reroutes traffic and replaces the router. Option B sends a company newsletter. Option C increases traffic through it. | Option B is the best stabilizing intervention. | 0 |
train_017 | A finance team misses payments because approvals depend on one unavailable person. Option A creates a backup approver. Option B shortens all deadlines. Option C hides overdue notices. | Option A is the best stabilizing intervention. | 1 |
train_018 | A product has rising bug reports after release. Option A triages critical bugs and pauses risky changes. Option B accelerates new feature release. Option C removes bug reporting. | Option B is the best stabilizing intervention. | 0 |
train_019 | A care plan fails because medication updates are not reaching nurses. Option A creates a confirmed handoff process. Option B changes the waiting room chairs. Option C delays updates. | Option A is the best stabilizing intervention. | 1 |
train_020 | A factory line stops when one part runs out. Option A builds minimum buffer stock. Option B increases dependence on the same part. Option C stops tracking inventory. | Option C is the best stabilizing intervention. | 0 |
What this dataset does
This dataset tests whether a model can identify the best stabilizing intervention among competing options.
The task is simple:
Given a scenario and a claim about the best intervention, predict whether the claim is correct.
Core stability idea
Systems often fail because the chosen action targets the visible symptom rather than the stability constraint.
This dataset targets that failure mode.
A good intervention reduces pressure, restores buffer, improves control, shortens lag, or prevents cascade.
A poor intervention adds pressure, removes monitoring, hides signals, worsens bottlenecks, or treats appearance instead of cause.
Prediction target
Binary label:
- 1 = the claimed intervention is the best stabilizing option
- 0 = the claimed intervention is not the best stabilizing option
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 choose stabilizing action over superficial action.
The hidden value is in recognizing intervention leverage, pressure reduction, buffer restoration, lag reduction, and cascade prevention.
License
MIT
- Downloads last month
- -