QJHWC/PaperForge

What it solves

PaperForge is a "Research OS" designed to bring rigor and verifiability to the scientific research process. It solves the problem of fragmented research workflows by unifying research planning, controlled experimentation, paper writing, and publication into a single system where every claim in a paper must be backed by verifiable evidence (code, data, or literature).

How it works

The system operates through a unified runtime (ResearchOSRuntime) and a scientific memory database (SQLite) that tracks the provenance of every assertion. It employs three execution profiles to enforce strict boundaries:

• writing-only: Restricted to reading evidence and writing LaTeX, blocking all compute/experimentation tools.
• research: Adds capabilities for literature review and research planning.
• full: Allows the execution of approved experiments across various backends (Local, Docker, SSH, Slurm, Kubernetes, Cloud SSH).

Experiments follow a mandatory pipeline: Proposal $\rightarrow$ Static Check $\rightarrow$ Mini Experiment $\rightarrow$ Full Experiment. The publication engine then runs a "Claim Gate" to ensure 100% of the paper's sentences are linked to evidence before rendering the final PDF using templates for CVPR, IEEE, or Elsevier.

Who it’s for

It is built for academic researchers and students who want to automate the tedious parts of paper writing and experiment tracking while maintaining a high standard of scientific integrity and reproducibility.

Highlights

• Evidence-Centric Writing: Every public claim in a paper is assigned a claim_id linked to specific source code, logs, or literature.
• Controlled Experimentation: Mandatory approval workflows and multi-stage execution (static $\rightarrow$ mini $\rightarrow$ full) to prevent uncontrolled compute spend and errors.
• Unified Compute Backends: A single interface to manage jobs across local machines, containers, and HPC clusters (Slurm/K8s).
• Publication Pipeline: Automated LaTeX compilation, layout diagnosis, and a final "Release Gate" that scans for secrets and verifies protected hashes before marking a project as completed.
• Domain Plugins: Built-in support for metrics and evidence import for CV, NLP, RL, Bio, Physics, and Robotics.