Researching the
boundaries of
autonomous
financial
intelligence.
A research initiative by Beancount.io
CURRENT FOCUS
ACTIVEACTIVE SINCE JAN 2026
"Can a language model reason over double-entry bookkeeping with the reliability of a trained accountant — and produce a verifiable audit trail?"
RESEARCH LOG
View all entries →Loading…
RESEARCH PRINCIPLES
01
Plain-text first
All experiments use real Beancount ledgers. No synthetic data, no proprietary formats. If it cannot be expressed in plain text, it is not in scope.
02
Reproducible by default
Every result is reproducible from the published ledger data and code. Negative results are published with the same rigour as positive ones.
03
Open findings
Research is published openly on the Beancount.io blog. The community is the peer review process.
FOLLOW THE RESEARCH
Findings and experiment write-ups are published on the Beancount.io blog.
BEAN LABS · BEANCOUNT.IO · 2026
