# BiasCraft

BiasCraft is an educational AI-bias simulation app. Users adjust hiring-model feature weights, run a deterministic weighted-sum model, and inspect fairness, accuracy, demographic selection rates, and bias alerts.

Machine-readable project graph:
- /graphify.json

Primary API:
- POST /run-model
- Request JSON: {"education": 0.5, "experience": 0.5, "location": 0.2}
- Response JSON includes metrics, alerts, and sample_candidates.

Core formula:
- pred_score = (education / 100) * education_weight + (experience / 15) * experience_weight + location_val * location_weight
- location_val = 1 for San Francisco candidates, 0 otherwise
- selected = pred_score >= 70th percentile threshold

Important dataset note:
- Candidate data is synthetic.
- San Francisco candidates receive a +10 education boost before the ground-truth hiring label is calculated.