Project Delivery Risk Monitor

Projects to review

Sorted by model risk score. Select a project for its underlying status signals.

No projects match these filters.

Scores are model estimates on synthetic data, not calibrated real-world probabilities. “Largest status signal” is a normalized rule-based indicator, not a causal explanation or model attribution.

Inspect a project

Select a row above to see its status.

Evaluation, with context

The simple review rule outperforms logistic regression on accuracy and recall in this split. One stratified holdout, seed 42. Portfolio scores also include training rows; use the split filter to isolate held-out projects. The original synthetic-data generator was not supplied, so target provenance and temporal leakage cannot be independently verified. This is not a production forecast.

From source data to a decision

CSV → validation → engineered features → logistic regression → scored data → dashboard.

The local pipeline regenerates every score and metric shown here. The optional cloud notebook reads S3, writes Delta tables, and logs MLflow artifacts. Streamlit can read the scored and metrics tables through a Databricks SQL warehouse. Cloud execution has not been verified.