Skip to content

Forecast Sandbox Lite — SKU Explorer

Most forecasting demos show one model and one accuracy number.
This one is different: it helps you decide which model to trust for each SKU, based on demand behavior and transparent evidence.

Try the live demo

Embedded demo

What you can do in ~30 seconds

  1. Select a SKU
  2. See the recommended model
  3. Review Score = MAE + |Bias|
  4. Check demand regime via ADI / CV²
    (Smooth / Erratic / Intermittent / Lumpy)
  5. Compare all candidate models in a ranked table
  6. View Actual vs Forecast and download aligned CSV output

Why this matters

  • There is no single model that wins for every SKU.
  • Portfolio performance is often driven by tail mistakes, not average win-rate.
  • When demand structure is unclear, a robust generalist fallback can reduce downside.

This app is built to support defensible model choice, not leaderboard chasing.

What powers the demo

The Space runs on precomputed artifacts so it is fast, stable, and reproducible:

  • data/processed/test.csv — ground truth
  • metrics/best_models.csv — per-SKU recommendation
  • metrics/combined_metrics.csv — model comparison table
  • metrics/demand_profile.csv — ADI/CV² regime tags
  • metrics/*_predictions.csv — model forecasts

Download slides

Slides

This Space is intentionally a decision-evidence viewer, not a model-training environment.

Scope note

This is an independent portfolio artifact.
No proprietary or confidential employer information is included.