Reviewer Guide

Warning

This guide is under construction. The content below outlines what will be available here once the guide is complete.

The Reviewer Guide is aimed at referees and readers who want to reproduce, verify, or challenge the results presented in the Sage paper. The goal is to make independent verification as frictionless as possible.

What Will Be Here

Automated Review Tools

Note

A dedicated reviewer tool suite is in preparation. It will allow referees to reproduce key paper results with a single command — no manual configuration required.

The planned tools will include:

  • Result reproducer — a CLI that reruns the full MLGWSC-1 evaluation from a published checkpoint and prints the sensitive distance and FAR curves used in the paper

  • Waveform validator — compares Sage IMRPhenomD/Pv2 outputs against LALSuite at a set of standard test points and reports mismatches above a threshold

  • Checkpoint inspector — loads a checkpoint and prints a human-readable summary of the model architecture, training config, and training history

Reproducing the Paper Results

Step-by-step instructions for reproducing each table and figure in the paper from publicly available data and checkpoints. Each step will list:

  • Input data (GWOSC segment, injection set, or synthetic)

  • The exact command to run

  • Expected output and how to interpret it

  • Approximate runtime on a reference GPU

Data Availability

  • Where to download the O3b noise data used in training (GWOSC)

  • Where to download the MLGWSC-1 injection set

  • Download instructions for published model checkpoints

Common Reviewer Questions

A curated list of questions that arose during peer review of the paper, with detailed technical answers — covering methodology choices, baseline comparisons, and robustness claims.

Getting Help

If a result cannot be reproduced or a claim seems inconsistent, please open a GitHub issue with the label reproducibility. We treat reproducibility issues as high-priority bugs.