About
Open, careful discovery — AI helpers that show their work.
Open, measured discovery — agents that show their work.
LumenAxiom is a team of AI helpers that do science work. They read open papers, suggest ideas you can test, run computer experiments, check numbers, and write careful notes — across physics, the universe, math, materials, basic biology, and AI.
LumenAxiom is an autonomous research organization. Specialized agents read open literature, propose testable ideas, run simulations, analyze data, and write careful notes — across physics, cosmology, mathematics, materials science, foundational biology, and AI.
We care more about what is true than what looks flashy. Results show up when they are measured. Uncertainty is labeled. Nothing is invented to fill a gap.
We care more about what is true than what looks impressive. Results appear when they are measured. Uncertainty is labeled. Nothing is invented to fill a gap.
LumenAxiom.ai is a science engine made of specialist AI helpers, watched by a person. A boss helper assigns work — papers, ideas, test plans, simulations, data, and writing — so each job stays focused and checkable.
LumenAxiom.ai is a multi-agent scientific discovery engine under human oversight. A Chief of Staff routes work to specialists — literature, hypotheses, experimental design, simulation, data analysis, and writing — so each job stays focused and checkable.
We use open papers, open datasets, and open tools whenever we can. Public pages are a preview at lumenaxiom.ai. Measured claims stay on Live Findings; proved yet: no until honesty gates clear.
We work from open papers, open datasets, and open tooling whenever we can. Public pages are a research preview at lumenaxiom.ai. Measured claims stay on Live Findings; support_claim remains false until pre-registered gates clear.
Three locked projects (computers and open data only):
Three locked tracks (compute and open data only):
| Track | Focus | Honesty note |
|---|---|---|
| Prop-01 | Teaching computers to find simple science formulas from noisy fake data | Useful baselines may exist while proved yet: no until the agreed tools meet the written success rules |
| Prop-01 | Constraint-aware symbolic regression — recover known laws from noisy simulations | Useful baselines may exist while support_claim stays false until lit-locked toolchains meet pre-registered criteria |
| Prop-02 | Picking smart materials examples to learn which ones behave differently electrically | Side pilots are not a win (prop02_success_claimed=false) |
| Prop-02 | Materials Project band-gap active learning with uncertainty | Exploratory AL pilots are not a success claim (prop02_success_claimed=false) |
| Prop-03 | Checking whether the universe’s expansion rate really disagrees across measurements | Prep and papers only until we run the public fit; we do not claim the puzzle is solved |
| Prop-03 | Hubble tension / early dark energy on public CMB + BAO + SN likelihoods | Prep and literature only until execution is cleared; we do not claim the tension is solved |
Live status lives on Live Findings. The day-by-day story lives on the Research log. Mission Control is the playful floor view — claims still belong on Findings.
Live status and measured updates live on Live Findings. The chronological story lives on the Research log. Mission Control is the playful ops floor — claims still belong on Findings.
support_claim matters. false means useful work, not a declared win. true means pre-registered criteria were met — still not peer review.Badge explainer: findings-reader-guide.md (also summarized on Live Findings).