HOW WE RESEARCH
AI-first research.
Publicly checkable evidence.
Accelerate discovery, expose the reasoning, and make participation possible with or without AI.
AI throughout the research process.
Our operating method uses AI in ideation, literature review, modeling, code, proof development and review. Authors remain accountable for the claims, sources, assumptions and released artifacts. AI output becomes evidence only when the supporting work can be examined and checked.
Disclose the role of AI and the review performed. Distinguish an independent check of an artifact from repeated agreement among models.
Publish openly. Keep improving.
We favor open-source research and prompt, versioned releases of manuscripts, code and supporting evidence, with DOI publication where available. Institutional peer review is not a prerequisite for our publication route. Public scrutiny, critique and correction remain essential.
A release should expose its scope, source versions, assumptions, reproduction instructions and known gaps. Corrections belong in the version history so readers can identify what changed. Check each artifact’s own license and third-party terms; availability alone does not grant universal reuse rights.
Different claims need different checks.
Formal proofs
Lean checks precisely stated mathematical propositions against their assumptions and dependencies. A checked lemma does not certify a whole manuscript or unstated real-world premises. Readers need the theorem scope, build environment and remaining gaps.
Empirical reproducibility
Experimental claims need methods, conditions, data or saved evidence, code and evaluation limits. Reproducing a calculation does not independently prove its imported premises or establish general real-world performance.
A DOI identifies a persistent research record and version. It is not quality certification, peer review or proof of a result.
Decentralized participation. Practical contributions.
Research benefits from democratic participation: anyone can inspect, question, critique or contribute, with or without AI. Contributions are assessed through their reasoning and evidence.
- Report an unclear assumption, citation error or reproduction problem in a public repository issue. Include the version and enough detail for another reader to check.
- Propose a focused code, documentation or proof change through a pull request, with its rationale and validation.
- Contribute without coding: improve explanations, identify relevant literature, question the causal link to the mission, or offer a concrete use case.
Choose the matching artifact from the research library for other projects. A proposal or public critique is an invitation to review, not a guarantee of acceptance.