Demand forecasting, made clear.
Compare transparent time-series models, benchmark accuracy against historical data and explore what future demand could look like.
From signals to smarter decisions.
One workspace for rolling backtests, model selection, scenario planning and a readable audit trail. Built for clarity, not black-box claims.
History and near-term outlook
Monthly volume, selected forecast and illustrative uncertainty envelope
Model leaderboard
Rolling-origin one-step test · lower MAE is better
Operating implications
A simple capacity-planning perspective
Trustworthy analysis
What this studio does and does not do
Each test point uses only values that existed before it.
Compare naive, average, smoothing, trend and seasonal methods.
Bands are heuristic residual-based ranges, not calibrated probability intervals.
Forecast Lab
Choose a transparent forecasting model, vary future demand and compare possible operating environments.
Simulation controls
Settings apply to future periods only; historical observations stay unchanged.
Scenario presets
One click changes demand shift and shock assumptions.
Projected demand curve
Historical actuals, simulated forecast, heuristic interval
Monthly forecast schedule
Demand and capacity comparison, including uncertainty proxies
| Month | Projected demand | Indicative low | Indicative high | Capacity gap |
|---|
Data Registry
Review, edit and archive fictional monthly observations, with safeguards for chronology and data integrity.
| Period | Demand | Status | Actions |
|---|
Archived observations are preserved. The forecast uses only active months, sorted chronologically. Missing months in the active series are rejected for time-series calculations.
Model Audit
Inspect chronological holdout tests, compare model errors and see exactly how a selection is made.
Backtest rankings
Models are scored on the same one-step test periods
Evaluation rules
Transparent definitions and caveats
All model outcomes
Equal chronological comparison for each supported method
| Method | MAE ↓ | RMSE ↓ | sMAPE ↓ | Test months |
|---|
A strong historical ranking can still fail under abrupt shifts. No statistical model is guaranteed to generalize.
Method & Backup
Clear formulas, editable local data and portable backup files. No external dependencies.
Forecasting methods
Deliberately simple baselines for intelligible comparison
Projects the most recent observation into future months.
Uses the latest three observations, with recursive updates for future steps.
Uses alpha = 0.35; the fitted level carries forward.
Ordinary least squares against month index; projected values cannot be negative.
Repeats the value from 12 months earlier, including recursively after one year.
Portable local workspace
Keep your work offline and take copies you control
Save all monthly values, archived records, and planning preferences.
Checks version, length, month format, uniqueness, value bounds and preferences before replacing local data.
Restore the original fictional dataset without changing any online account.
Important methodological limits
Responsible interpretation of a toy forecasting system