pi-repro

Read an academic paper and extract its method, datasets, hyperparameters, compute needs, and every reproducible quantitative result into structured claims. Use as the first step of reproducing a paper, after init_reproduction.

Write the final reproduction report for a paper — a per-claim comparison of reported vs reproduced values, overall fidelity, deviations, blockers, and next steps. Use as the last step of reproducing a paper, after claims have been attempted.

Reproduce the quantitative results of an academic paper. Use when asked to reproduce, replicate, or verify the results/numbers/tables/figures of a paper, given a link, PDF, arXiv id, or DOI (optionally with a reference GitHub repo). Orchestrates ingest → gap-analysis → plan → per-claim reproduction → report.

Analyze a paper's reference code repository against the paper to determine what is actually implemented versus missing, partial, or stubbed. Use when a GitHub repo (or local code) was provided for reproducing a paper, before planning the reproduction.