What problem does it solve? Verifying whether a published paper's results actually hold up requires careful planning, environment control, and disciplined evidence tracking. This Skill structures the entire replication workflow so claims are tested systematically rather than ad hoc, with every result tied to its script, seed, and command. ## Core Features & Use Cases - Claim Ledger Tracking: Builds a ledger linking each paper claim to its dataset, method, hyperparameters, compute, and observed result, with assessments of aligned, partially aligned, inconclusive, or not attempted. - Environment-Gated Execution: Requires an explicit environment choice (local directory, Git branch, Pixi environment, remote SSH/Slurm, or plan-only) before any packages are installed or experiments run. - Waterology CLI Integration: Uses waterology reproduce, workflow register, and deliverable export commands to restore environments, execute through TORC, and collect run evidence. - Use Case: A researcher wants to verify the headline benchmark table from a recent machine learning paper. The Skill extracts the recipe, checks dataset availability, plans downscaled runs, executes claim by claim after environment confirmation, and produces a report with a reproducible evidence trail. ## Quick Start Ask the AI to plan a replication of a specific paper's main result, starting with a claim ledger and an explicit environment choice before any execution.
