# Credit-Risk Evidence Verification

## Reproduced results

- Portfolio size: **50,000**
- Synthetic default rate: **13.344000%**
- PD-model test AUC: **0.7007172881**
- AUC rounded: **0.701**
- FICO buckets: **8**
- FICO boundaries: **564, 592, 616, 652, 676, 698, 762**
- Default-rate monotonicity by ascending FICO: **True**

## Evidence interpretation

These results support the claim that the public project is internally reproducible under its documented synthetic-data generation and modelling specification.

They do **not** establish real-world credit performance, production deployment, regulatory validation, institutional impact, or commercial outcomes.

## Correct professional wording

Use:
> Reproducibly demonstrated on a seeded synthetic credit-risk dataset.

Do not use:
> Proven in production.

## Verification controls
- Seed: `42`
- Observations: `50,000`
- Train/test split: `75/25`
- Split stratification: `True`
- Scaler: `StandardScaler`, fitted on training data
- Model: `LogisticRegression(max_iter=1000, class_weight="balanced")`
- FICO buckets: `8`
- Minimum bucket size: `1,000`