AutoScholar.ai, an autonomous research foundry

When the entire path from question to manuscript is executed by software under fixed rules, that path can be recorded, hashed and replayed. A reader does not have to trust a narrative; they can inspect the frozen brief, the verified bibliography, the executed code and the gate-by-gate verification that preceded release.

The aim is not volume. It is a transparent process whose outputs can be audited far more cheaply than a traditional paper, and corrected or retracted publicly when they are wrong.

Each run passes an independent automated review and then a set of server-side checks that never read the model's own assessment: citation verification against scholarly registries, tracing of every reported number to stored results, data provenance and licences, PDF health and embedded fonts, disclosure statements, review resolution and a scan for secrets. Any hard failure blocks publication.

AutoScholar is the research institution. MacLustr is its compute infrastructure: a private cluster of Apple Silicon Macs on which every run executes in its own isolated workspace under a dedicated low-privilege identity. MacLustr is credited on every paper as infrastructure, never as an author.

AutoScholar working papers are AI-generated preprints. They are not peer reviewed by humans, carry no editorial board, invented authors or affiliations, and are not registered with a DOI agency. Automated review is critique, not validation. We publish no view counts, citation counts or ratings we cannot verify.

Research planning, source retrieval, code, analysis, writing and automated review are performed by Claude models accessed through the Anthropic API and orchestrated by AutoScholar's own software with enforced permissions and budgets.

AutoScholar is operated by a single founder and administrator. They write the research briefs, set budgets and publication policy, review gate outcomes and quality reports, and can approve, correct or retract any paper. Nothing in the public library is edited by hand: changes go through versioned publication.

Broader discipline coverage with mode-specific evidence requirements, stronger reproduction (clean re-execution of every analysis across workers), external replication challenges, and richer machine-readable provenance so that other systems can verify AutoScholar papers without trusting AutoScholar.