What this vignette is for | Setup: simulate one predictor, one response — over years | Stage 1: turn a quick frequentist fit into a prior | Stage 2: Bayesian refit with the informed prior | Stage 3: posterior prediction across the forecast horizon | Stage 4: decompose the variance over time | Stage 5: when does the forecast become uninformative? | The whole-pipeline figure | What this example does not show | Session info

Introduction | The simulated dataset | Setup | Exploring the dataset | Visualise the training data | Single-cluster workflow: Cluster A | Step 1 — Elastic net for prior extraction | Step 2 — Extract informed priors | Step 3 — Fit the Bayesian model | Step 4 — Diagnose convergence and model fit | Step 5 — Predict with environmental noise propagation | Step 6 — Uncertainty decomposition | Step 7 — Forecast shelf life | Step 8 — Calibration assessment | Visualisations | Forecast fan chart | Prior versus posterior | Forest plot: Bayesian vs. elastic net | Parameter recovery validation | Multi-cluster grouped workflow | Fitting | Diagnostics | Prediction and decomposition | Shelf life comparison | Calibration and forecast plots | Long-horizon forecasting with GCM projections | Using lm instead of glmnet | Standardising and back-transforming | Summary | Session information