# Time-series forecasting and handover

Run `timeseries_cases.py` with `.playwright-mcp/ds-timeseries-venv/Scripts/python.exe` from the repository root. The extension environment uses system site packages plus the exact additions in `timeseries-requirements.txt`; it is not an isolated clean-environment reproduction.

The original fixture contains 144 monthly rows: 96 train, 24 validation and 24 test. Demand has authored trend, annual seasonality, planned promotions and a structural break beginning at the test origin. Generator components are audit aids, not model inputs or real demand. Capacity-censored observed sales are retained to teach the difference between demand and sales.

The lab executes seasonal-naive, rolling-origin, metric, decomposition, differencing, ARIMA, seasonal ARIMA, Holt-Winters, interval, regression, lag-tree, intermittent-demand, hierarchy, leakage, structural-break, multi-step, anomaly and handover cases. Its verifier passes seventeen boundary and reconciliation checks and saves the complete outputs to `timeseries-verification.json`. Statsmodels 0.15.0 is local to the extension environment; global NumPy, SciPy, pandas and scikit-learn packages are reused.

Primary references: [statsmodels time-series API](https://www.statsmodels.org/stable/tsa.html), [ARIMA](https://www.statsmodels.org/stable/generated/statsmodels.tsa.arima.model.ARIMA.html), [ExponentialSmoothing](https://www.statsmodels.org/stable/generated/statsmodels.tsa.holtwinters.ExponentialSmoothing.html), [scikit-learn lagged forecasting](https://scikit-learn.org/stable/auto_examples/applications/plot_time_series_lagged_features.html), and [Forecasting: Principles and Practice](https://otexts.com/fpp3/).