LLM-CA / PCR-Bench: Frozen Synthetic Context-to-Service Simulation Data and Revision Evidence

Frozen synthetic evidence for LLM-CA and the Pervasive Context Reasoning Benchmark (PCR-Bench): 1,440 synthetic context snapshots in smart home, learning, tourism and non-clinical assisted living, with 23,040 recorded baseline and mock verification predictions, raw outputs, metrics, reports, configuration and reproduction source. The revision adds hash-identified clean-test and matched robustness tables and an actual recorded rule-baseline input/output trace. Reference targets are provisional generator judgments; the rule baseline shares the generating function and nearest-neighbor cases share templates. Historical automated annotation passes are correlated and excluded from scientific findings. Independent human review remains pending. No usable new LLM predictions, personal sensor data, clinical judgments or live service outcomes are included. Creators: Konstantinos Michalakis (ORCID 0000-0002-5943-6613) and George Caridakis (ORCID 0000-0001-9884-935X), Department of Cultural Technology and Communication, University of Aegean, Mytilene, Greece. No external funding. Source: https://github.com/kmichalakis/LLM-CA . Original source version: v0.1.0; revision tools are separately identified.