Ensemble sufficiency for stochastic wildfire automata: convergence sweep scripts and derived tables

Code and derived data supporting the article "Ensemble Size Changes What a Fixed-Threshold Agreement Metric Estimates. A Convergence Study of a Stochastic Wildfire Automaton". The record contains the convergence sweep that measures how agreement and burned area behave as the ensemble grows, the per-evaluation timings behind the projected calibration saving, the resampling behind the reported intervals, the reanalysis script that derives every table and every exact permutation p value of the article from the deposited tables, and the derived tables themselves. The README inside the archive lists every script and every table with its purpose. The primary UAECOB incident registry is not redistributed here, because its terms of use rest with the Bogota fire service, and regenerating the calibrated event set from scratch requires it. What changed in version 1.1.0 This version corrects 1 defect, closes 1 reproducibility gap, and adds the reanalysis that the revised framing of the article rests on. No measurement was repeated on the device, because none needed to be. The across-ensemble spread is now computed with divisor J-1, which is what the error definition of the article states. The sweep previously used the NumPy default of J, so every deposited spread was 6.9 per cent below its definition. Since J is 8 in all 96 rows the correction is exact and was applied by conversion rather than by re-simulating. It moves no sufficient size in the power-of-2 sweep and leaves the rank correlation and its p value unchanged. The record now runs. The previous version omitted a module that another script imports at load time, together with the calibrated per-event transition parameters, so 3 of the 5 scripts failed on import. Both are included now, under the same Apache License 2.0, at the path the code resolves. run_reanalysis.py is new and needs neither a graphics processing unit nor any third-party package. It produces the sensitivity of the sufficient size to metric, tolerance and fire subset across all 16 combinations, the fitted scaling of the sufficient size against both observed and simulated extent with confidence intervals, the simulated-against-observed footprint comparison for all 8 events, the drift of the mean of kappa across the sweep, and the log-log decay slopes. Every p value is now an exact permutation value obtained by full enumeration, which with 5 or 6 fires is 120 or 720 relabellings. The t approximation that statistical packages return is unreliable at that sample size and disagreed with the exact value by up to a factor of two. Four artefacts of version 1.0 moved to legacy/, with a note explaining each. Two tables were withdrawn outright, one written by hand and carrying ranges in prose, and one comparing the sufficient size against a prediction anchored at a single event. Two more carry the t approximation and a percentile interval that the article no longer cites, and are superseded by the sensitivity, scaling and jackknife tables. The scripts that produced the retired tables moved with them. Nothing in code/ or data/ now disagrees with the article. 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/. The record is self-contained, since it carries the calibration and simulation module vendored under Apache 2.0, so there is no external repository to fetch first. Licensing, which is split on purpose The source code in code/, together with the vendored modules, is released under the Apache License 2.0, in the file LICENSE. The derived tabular data in data/ are released under Creative Commons Attribution 4.0 International, in the file LICENSE-DATA.txt. The split is deliberate, so that the code can be reused in a downstream product without the attribution obligations of the data travelling with it. Environment Python 3.12. The convergence sweep runs on a graphics processing unit through the batched engine the record vendors. The reanalysis script runs on central processing units alone and needs no third-party package. Random seeds are fixed throughout and recorded in each script, and the replicates are drawn on the device with a counter-based generator whose per-stream independence is what allows the across-ensemble spread to be read as an error. A note on language The scripts carry Spanish-language comments and the tables carry Spanish column names, because Spanish is the working language of the project. The README included in the archive gives the English purpose of every file and a glossary of the recurring column names.