Check the model (no Julia) | Fit (needs the Julia engine) | After this works | Twin-package flow

What was tested | Run accounting | Prediction interval results | When does the neural component add predictive value? | Inferential results | Package-check scope | Reproducibility and limits

Start here | Common declarations | What pigauto does | How it works (briefly) | Installation | Inspecting the internal pipeline | The AVONET bird data | Step 1: Preprocess | Step 2: Create evaluation splits | Step 3: Fit the BM baseline | Step 4: Build the phylogenetic graph | Step 5: Train the model | Step 6: Predict | Step 7: Evaluate — and read the gate | Step 8: Visualise | Training history | Uncertainty ribbons | Experimental analysis-aware multiple imputation | Phylogenetic tree uncertainty | Active imputation: where to measure next | Advanced controls | Solver, calibration, EM, and refinement | Exact conditional prediction route | Phylogenetic-signal gating | Assembling covariates from public data | Tips and next steps | References

Overview | Synthetic example | Preprocessing | Creating splits | Baseline fitting | Fitting | Prediction and decoding | Evaluation | Experimental analysis-aware multiple imputation

Which imputations can be pooled | Impute | Fit and pool | Phylogenetic regression (nlme) | Mixed models (lme4 and glmmTMB) | Location-scale models (drmTMB) | Multivariate latent-variable models (gllvmTMB) | Other models | Bayesian models: concatenate, do not pool | References

Scope | Supplying posterior trees | Prediction-sensitivity workflow | When the optional GNN is shared | What this does not establish | References

1. End-to-end data flow | 2. The phylogenetic baseline | 2.1 Continuous, count, ordinal, proportion: Brownian motion | 2.2 Binary and categorical: phylogenetic label propagation (LP) | 2.3 Joint MVN baseline (Phase 2) | 2.4 Threshold-joint baseline (Phase 3) | 2.5 OVR categorical (Phase 6) | 2.6 Multi-obs aggregation (Phase 10 + B1 soft) | 3. The GNN: ResidualPhyloDAE | 3.1 Encoder | 3.2 Graph Transformer Block (Phase 9 + B2) | 3.3 Decoder and the gate | 3.4 Loss and three safety regularisations | 3.5 Within-row cross-trait attention (B3, v0.9.3, opt-in) | 4. Calibration: making the gate a real safety floor | 5. Uncertainty quantification | 6. Posterior-tree prediction sensitivity | 7. Putting it together: a worked predictive equation | 8. Where to look in the code | References

1. "impute() stopped because my input is fully observed" | 2. "My ordinal trait predicted 100 % majority class" | 3. "The gate stays closed and the GNN seems to do nothing" | 4. "How do I know if my dataset has enough phylogenetic signal?" | 5. "Predictions are way bigger than anything I observed"