ZUSE Automat Agent: Empirical Law Discovery in Elementary Cellular Automata

Version v1.3 adds the Phase 24 primitive period-8 background oscillator sweep.The experiment covers all 256 ECA rules, 30 primitive length-8 binarybackgrounds, and 502 initial-condition words (3,855,360 runs). It identifies23 new rule/type pairs, oscillator periods up to T=15, and a new glider speedof 2/3 cell per step. A rotation sub-test finds background-phase dependence inall 10 sampled cases. The abstract and Sections 7.7, 9.3, and 10.2 have beenupdated accordingly. Version v1.2 added Section 7.6: a periodic-background oscillator sweep over all256 ECA rules and 15 non-zero periodic backgrounds (1,927,680 runs). Keyfindings included 30 stationary oscillator rules and 36 moving oscillator rulesunder non-zero backgrounds, including period-4 oscillators and speed-0.5 glidersnot present under quiescent zero background. rule_108 persisted under all-onebackground with the same motif. The zero-background uniqueness claims fromv1.1 remained valid and unchanged. We present ZUSE Automat Agent (ZUSE), a deterministic, policy-driven discoveryloop for elementary cellular automata (ECA) that builds an empirical atlas ofcycle laws, world families, and basin fragility without a language model in thediscovery loop. The pipeline combines a fixed seven-law evaluator, adedup-gated observer stack, and persistent multi-seed world records, and isapplied to a 20-world sample spanning ECA rules, Conway's Game of Lifepatterns, and synthetic controls. Key results include a seven-category empirical atlas, a complete fragilityspectrum across measured worlds, rule_108 as the unique stationary localperiod-2 oscillator under the tested quiescent-background protocol, aneight-rule family of minimal moving period-2 speed-1 gliders, and a richerbackground-conditioned oscillator landscape containing period-4 throughperiod-15 oscillators and glider speeds of 0.5, 2/3, and 1 cell per step. The discovery loop itself is fully deterministic. Language-model assistance wasused only after execution for interpretation and documentation, not for worldselection, law evaluation, scoring, or acceptance of empirical results. Every result is reproducible from deterministic scripts with no stochasticcomponents in the discovery loop. Repository:https://github.com/conchaestradamiguelangel-droid/zuse-automat-agent