The complete Gecode before/after report and its verified measurements now live separately from the core solver contribution.
Download gecode-before-after.zip, unzip it, and open gecode-before-after/index.html. All 705 report files and linked data/documentation are preserved byte-for-byte. SHA256SUMS verifies the report archive.
The repository's python3 verify.py command checks the immutable files, 115 frozen models, 356 measured observations, three loader probes, 355 saved witnesses, and 21 sampled brackets without running a solver. python3 -B experiments/optimize/test_final_instances.py -v repeats the generator regressions.
The final study uses one thread and ten seconds per solve. “Original” means the pre-algorithm project facade, not pristine upstream Gecode. Sampled capacity brackets are not speedup ratios or mathematical maximum-size guarantees. Witnesses are independently checked; large-instance optimality remains backend-reported. The study predates the MiniZinc controls and source-only cleanup; it has not been relabeled as a new measurement.