SimLedger

SimLedger is a web app (with optional desktop runner) for running agent-based models where every run is reproducible, comparable, and reviewable. It targets a real pain in ABM work: results that can’t be traced back to exact parameters, code, random seeds, data snapshots, and environment versions. Users define experiments, sweep parameters, and launch runs locally or on a managed compute pool. Each run automatically logs provenance (inputs, git commit, container hash, seeds), outputs (metrics, time series, artifacts), and a human-readable “run card” that can be shared with reviewers or stakeholders. Built-in comparison views highlight which parameter changes actually moved outcomes, and it flags non-deterministic behavior. This isn’t a toy simulator; it’s the boring, necessary infrastructure layer that teams end up building badly in-house.