# Proof of Work — Portfolio Copy

## Quantitative Risk
Reproducible credit-risk PD modelling and FICO segmentation on a seeded synthetic portfolio of 50,000 borrowers.

## AI Evaluation
A 240-case benchmark with explicit error labels and deterministic quality scoring. Current benchmark accuracy: 0.500 on the synthetic gold labels.

## AI Workflow Automation
A 160-document text pipeline covering classification, extraction and validation. Current synthetic benchmark: 1.000 type accuracy and 1.000 full validation pass rate.

## SQL + BI
A reproducible SQLite decision-intelligence pipeline with 1,200 customers, 3,000 orders, SQL KPI queries and management-oriented outputs.

## Economics & Forecasting
A 60-month revenue backtest with a 12-month holdout and explicit scenario forecasts. Current synthetic holdout MAPE: 2.51%.

## Evidence standard
Each result identifies provenance, test conditions, reproduction method and limitations.