Reproducibility data and code for factorized Green-column caching in interactive multiclass graph label propagation

This record contains the study-generated reproducibility data and code for themanuscript “Factorized Green-Column Caching for Interactive Multiclass GraphLabel Propagation: Cost Models and Solver Crossovers.” The study develops acumulative selected-column implementation for exact hard-label updates inGaussian-fields-and-harmonic-functions propagation. It maintains raw requestedGreen columns and a dense Cholesky factor of the accumulated constraint block,derives a fixed-batch lifecycle cost model, and gives a calibration-envelopecertificate for sufficient cache selection or rejection under a memorybudget. The deposit includes solver implementations, frozen protocols and launchers,numerical tests, case metadata and query orders, per-process and per-roundrecords, structured failures, aggregate tables, closed-loop/restart/finite-precision audits, manuscript source data, and figure/table generators. Thebenchmark compares seven exact and iterative solver paths and intentionallyretains both favourable and unfavourable cache outcomes. Raw third-party graph and tabular datasets are not redistributed. The archiveinstead records provider URLs, frozen repository revisions, SHA-256 hashes,and deterministic preprocessing instructions. It contains no human-participant, clinical, personal, or confidential data. A stored-byte manifestverifies every archived file after machine-local path sanitization.