Differential Conformance Testing of Fast Pokémon Battle Simulators Across Generations and Doubles

Fast battle simulators let a search agent examine more possible actions, but their value depends on the states and outcomes they preserve. We present a measurement case study of differential testing against Pokémon Showdown. The protocol reconstructs positions, checks state conversion, samples reference outcomes for each joint action, and compares them with an enumerating simulator. We reanalyse recorded audits across singles formats and doubles, separating discrete-outcome agreement, modal agreement, damage-roll identity, and converter acceptance. In Gen 9 Ubers, strict deterministic agreement rose from 95.84% to 98.32% on development sample A and from 99.24% to 99.82% on re-drawn sample B after reference-setup repairs. The engine with 17 patches scored 98.32% on its debugging sample and 93.31% on a re-drawn sample, but a changed state interface confounds that contrast. In doubles, the final development run reached 98.77% strict agreement and 97.79% damage-roll identity; the originally planned gate was not met. A later statistical criterion gave 99.95% on development run 4. On 600 sealed fresh doubles positions, strict agreement was 98.23% and damage-roll identity was 98.02%; both original gates remained unmet. A sealed follow-up sampled 5,400 decisions from new recorded singles games and public doubles replays and completed 9 format comparisons with the deployed engines frozen. Strict deterministic agreement ranged from 97.92% to 99.58%. The format-level results include conversion refusals, pair exclusions, and position-clustered intervals. The sealed results measure installed simulators with frozen reconstruction and comparison code on new games. The historical repair samples remain development evidence, not independent tests.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24373867
Primary Topic
Software Testing and Debugging Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Differential Conformance Testing of Fast Pokémon Battle Simulators Across Generations and Doubles

S M Abdulla Al Mamun
Zenodo (CERN European Organization for Nuclear Research)
Software Testing and Debugging Techniques
preprint

Differential Conformance Testing of Fast Pokémon Battle Simulators Across Generations and Doubles

S M Abdulla Al Mamun
preprint en

Abstract

Fast battle simulators let a search agent examine more possible actions, but their value depends on the states and outcomes they preserve. We present a measurement case study of differential testing against Pokémon Showdown. The protocol reconstructs positions, checks state conversion, samples reference outcomes for each joint action, and compares them with an enumerating simulator. We reanalyse recorded audits across singles formats and doubles, separating discrete-outcome agreement, modal agreement, damage-roll identity, and converter acceptance. In Gen 9 Ubers, strict deterministic agreement rose from 95.84% to 98.32% on development sample A and from 99.24% to 99.82% on re-drawn sample B after reference-setup repairs. The engine with 17 patches scored 98.32% on its debugging sample and 93.31% on a re-drawn sample, but a changed state interface confounds that contrast. In doubles, the final development run reached 98.77% strict agreement and 97.79% damage-roll identity; the originally planned gate was not met. A later statistical criterion gave 99.95% on development run 4. On 600 sealed fresh doubles positions, strict agreement was 98.23% and damage-roll identity was 98.02%; both original gates remained unmet. A sealed follow-up sampled 5,400 decisions from new recorded singles games and public doubles replays and completed 9 format comparisons with the deployed engines frozen. Strict deterministic agreement ranged from 97.92% to 99.58%. The format-level results include conversion refusals, pair exclusions, and position-clustered intervals. The sealed results measure installed simulators with frozen reconstruction and comparison code on new games. The historical repair samples remain development evidence, not independent tests.

Zenodo (CERN European Organization for Nuclear Research)
University of Manchester (GB)
Software Testing and Debugging Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon