Pipeline

Code Audit

Compare paper claims against code for reproducibility and mismatch risks.

Quick reference

Core invocation details and the exact command surfaces users will touch first.

Type
Pipeline
Slash command
/audit
CLI
bohr audit

At a glance

The shortest useful summary of what this workflow is for and what it produces.

  • Links claims in writing to code behavior and highlights mismatches or risks.
  • Built for reproducibility reviews and ML papers with accompanying repositories.
  • Outputs traceable mappings and ranked risk items.

How to run

Start with the slash command in the REPL for interactive work, or call the CLI when you need a shell-native invocation.

In the Bohr research shell (REPL), enter /audit followed by your topic or instructions. From a terminal, invoke the CLI as bohr audit. Refer to the linked documentation for arguments, profiles, and environment setup.

Flow diagram

Stages and agent labels are indicative. Your configuration may swap or rename modules, but the diagram still shows the intended execution shape and handoff order.

Detailed guide

Full notes, usage guidance, and the longer explanation for how this workflow behaves in practice.

Overview

Maps statements in a paper to repository behavior and flags discrepancies—ideal for ML reproducibility reviews.

When to use it

Before release, during replication studies, or when reviewing third-party papers with code.

Outputs

Traceable mappings plus severity-ranked risks.

Practical notes

Exact agents, tools, and stop rules depend on your Bohr configuration. For CLI flags, examples, and edge cases, open the full documentation from the Documentation & links section on this page.

See also