Reproducibility Package for Point-in-Time Validity in Municipal 311 Service-Time Prediction: A Multi-City Benchmark of Outcome Construction, Information Availability, and Temporal Evaluation

This record contains the reproducibility package for the study “Point-in-Time Validity in Municipal 311 Service-Time Prediction: A Multi-City Benchmark of Outcome Construction, Information Availability, and Temporal Evaluation.” The study evaluates how outcome-model specification, point-in-time information availability, and temporal evaluation affect municipal 311 service-time prediction across Austin, Chicago, and New York City. The benchmark crossed closed-only squared-error regression and cutoff-censored Normal accelerated failure-time modelling, creation-stable and hindsight information contracts, and random retrospective and rolling-origin evaluation protocols. The package includes the complete Supporting Information, a machine-readable supporting-data workbook, CSV exports, analysis and audit scripts, dated protocol and environment locks, extraction and cohort manifests, aggregated primary and secondary results, calibration-resolution outputs, model-ranking and practical-tie analyses, sanitized portal metadata snapshots, file manifests, and SHA-256 checksums. The primary benchmark comprises 120 factorial result cells. The secondary and sensitivity analyses comprise 136 additional cells. The package also includes the numerical-integrity audit used to resolve 180 primary calibration-slope estimates. Raw municipal request records, request identifiers, request-level predictions, fitted model objects, feature-table files, databases, virtual environments, and execution logs are not redistributed. The official municipal open-data portals remain the sources of record. Extraction manifests and reconstruction scripts are provided, but later re-extraction may not produce byte-identical source files if the municipal portals revise historical records. The prediction outcome is time to an administrative terminal disposition recorded by the municipal system. It should not be interpreted as verified physical resolution of the reported issue. The outcome and evaluation contrasts compare complete benchmark specifications and should not be interpreted as isolated causal effects of censoring or random splitting. The hindsight information contract is diagnostic and includes mutable current-state fields and near-terminal proxies that were unavailable at request creation.