Reproducibility Package for Screen-or-Escalate N-1 Pipe Outage Screening in Synthetic Water Distribution Networks — V8.8 V7.1.1 Audited Release

This reproducibility package supports the manuscript “Screen-or-Escalate Workflow for N-1 Pipe Outage Screening in Synthetic Water Distribution Networks.” Version 8.8 extends the frozen-method evidence with a strict V7.1.1 pre-submission audit and additional analyses addressing comparator adequacy, practical benefit, decision-threshold sensitivity, unit of analysis, and proxy-defined hydraulic consequences. The package includes the frozen graph-hydraulic screening workflow, 90 synthetic confirmation networks and their EPANET input files, direct WNTR 1.5.0/EPANET 2.2 confirmation evidence, closure- and network-level results, simple-baseline comparisons, 100-seed random-baseline analyses, isolated one-factor-at-a-time sensitivity analyses for all hard thresholds governing eligibility, consequence classification, and candidate inclusion, including local analysis around the 18 m pressure-eligibility boundary, failure-path accounting, proxy/failure-mode analysis, solver-options audit, practical-overhead timing, scalability evidence, WaterGEMS cross-check material, deviation-log and reference-integrity audit artifacts, journal-relative novelty audit material, and SHA-256 integrity manifests. Direct EPANET confirmation comprised 5,700 single-pipe closures and 4,382 consequential hydraulic-threshold events, with no observed false negatives in the 90-network synthetic confirmation corpus. The direct-engine workflow permitted selective screening in 16 of 90 networks, avoiding 167 of 5,700 runs overall (2.93%) and 167 of 570 runs (29.3%) within the eligible networks; the remaining 74 networks conservatively escalated to complete N-1 analysis. Isolated one-factor-at-a-time sensitivity analyses were performed for the hard thresholds governing eligibility, consequence classification, and candidate inclusion. These analyses identify strong efficiency sensitivity near the frozen 18 m pressure-eligibility boundary; the 35% ranked fraction remains a prespecified frozen choice rather than a demonstrated optimum. Probabilistic uncertainty propagation was not performed; robustness was instead examined through threshold sensitivity, scenario/stress analyses, solver-boundary checks, and direct-versus-reconstruction comparison. These findings are reported as limitations and robustness diagnostics rather than evidence of universal threshold robustness or optimality. The evidence remains synthetic and generator-family-bound and does not constitute field validation, a universal outage-detection guarantee, or evidence of universal computational savings.