Labs
These three labs are parts of the same bigger idea: Forecast → Price → Decide.
They are not one fully connected end-to-end pipeline yet.
That is intentional.
Each step needs a different kind of dataset and assumptions, and I don’t currently have one public dataset that realistically supports all three steps at the same quality level.
So instead of forcing a fake “full cycle,” I built and validated each part properly.
What this means in practice
- You can review each module on its own technical merits.
- You can see how the logic connects across the full decision cycle.
- You also get a clear boundary of what is done vs. what is next.
Forecasting Sandbox (Lite)
This module focuses on forecast quality and model governance for SKU portfolios.
It benchmarks models, selects defensible per-SKU choices, and reports transparent evidence: - MAE - Bias - Score = MAE + |Bias| - regime tags (ADI / CV²)
Pricing Decision Lite
This module focuses on pricing under uncertainty, not fragile point estimates.
It produces: - a clear decision state (OPTIMIZE / HOLD / NO-GO) - a robust price recommendation - profit distribution evidence across scenarios
Decision Kernel Lite
This module focuses on final decision selection under risk.
It makes assumptions explicit (actions, scenarios, probabilities, losses) and compares choices using: - Expected Loss - Minimax Regret - CVaR (tail risk)
Current status
- Today: modular components are working and validated.
- Next: integrate into a single connected cycle when a suitable shared dataset is available.
Scope
All labs here are independent portfolio work built with public/synthetic data where applicable.
No confidential or proprietary employer information is included.