Perimeter-ring tautology diagnostic: scripts and derived cell-to-cell tables for the Eastern Hills of Bogota, Next Day Wildfire Spread and WildfireSpreadTS

Code and derived data supporting the article "Contact With the Front Is Nearly Sufficient: An Ablation Baseline for Cell-to-Cell Wildfire-Spread Classification, and a Corpus That Fails Its Own Diagnostic". The record contains the scripts that run every reported experiment and the derived tables they produce. The continental replications run end to end from their public sources. The Andean scripts read two cell-to-cell tables built from the UAECOB incident registry, which is not redistributed here because its terms of use rest with the Bogota fire service, so they regenerate the Andean numbers once a requester has obtained that registry and rebuilt those tables with the tabulator included here. What changed in version 1.1.0 The main result of the article moved from a comparison between a trivial rule and a classifier to an ablation of the neighbour count, and this version carries what that move needs. Three new experiments. run_ablation_neighbours.py and run_ablation_continental.py train the same classifier with every feature, without the neighbour count and with the neighbour count alone, on all four evaluation settings. run_saturation.py tabulates the label rate against the neighbour count, which is where the contact mechanism is read. Every interval now resamples groups, and every draw is shared by the four arms, so a difference between two of them is a paired difference and its interval does not carry the between-scene variance that pairing cancels. The Andean ablation bootstraps the eleven leave-one-event-out folds. The continental ablation repeats the grouped 80/20 split ten times and resamples the test scenes or fires two hundred times inside each. Both report the Matthews coefficient alongside kappa, on the same draws. The Andean tabulator is deposited. tabulador_andino.py built the two Andean tables and is copied from the working tree without changing its body, so the defect that Section 3 of the article declares can be read line by line in the code that produced it. How to run it Unzip the archive and run any script from code/. Paths resolve against the record itself through code/rutas.py, so nothing points at the machine where the work was done. Figures are written to figures/. Environment Python 3.13 with xgboost 3.3, scikit-learn 1.9, scipy 1.18, numpy and pandas. The Andean tabulator additionally needs rasterio, geopandas and shapely. No commercial component. The Andean experiments run in under 10 minutes on a laptop and need no graphics processing unit. The continental replication was run with one GPU for the gradient-boosted fit on 2.5 million rows and completes on CPUs alone at a longer wall-clock cost. Licences Source code in code/ is licensed under the Apache License 2.0. Derived data in data/ are licensed under CC BY 4.0.