Methodically built from peer-reviewed science.
Our engine assembles the applicable context, orchestrates a mixture of frontier models, and operates them in agents that think as scientists do.
It reasons across hours of compute and checks its own conclusions.
Closing the Empirical Loop: Autonomous AI Agents Conduct End-to-end Research With Human Participants
Wehr · Rideaux · Fox · Lightfoot · Tangen · Mattingley · Ehrhardt · Advanced Science (2026), e76675
Made for doing science
Multi-model orchestration
Explorer One works alongside frontier models, with every sub-task routed to whichever handles it best. Results are cross-checked across model families, removing single-model bias by construction.
The scientific tech stack
The stack operates the full scientific method autonomously. Verification checks each source and tests novelty against the field.
Dynamic retrieval
Live
The system retrieves and indexes literature at run time, never from a frozen training set.
DOI verification
Per-citation
The system resolves each reference to a live source before it enters an output.
Novelty assessment
Corpus-grounded
The system checks novelty against the published corpus, not the model’s prior.
Memory systems
Hours of context
The memory systems keep coherence across hundreds of agents and hours of compute.
Agents that think like scientists.
Multiple AI agents work together on research tasks. They spot patterns, break problems down, check their own work, and know when they're finished. The system adjusts in real time.
AgentField-IngestReads literatureAgentResearch-DesignCritiques designAgentMethods & StatisticsAudits analysisAgentPower-AnalysisSpawned · n=12AgentInterpretive-RigourChecks claimsAgentCitation-HealthDOI verify · noveltyAgentAdversarialWrites case againstAgentSynthesisFinal outputAgentField-IngestReads literatureAgentResearch-DesignCritiques designAgentMethods & StatisticsAudits analysisAgentPower-AnalysisSpawned · n=12AgentInterpretive-RigourChecks claimsAgentCitation-HealthDOI verify · noveltyAgentAdversarialWrites case againstAgentSynthesisFinal outputAutonomous end-to-end scientific discovery
A single agentic system operates all steps in the scientific method.
Frame the question.
Map the literature around itHypotheses, power, protocol. Pre-registered.
Run, log and verify at the source.
Statistics checked. Models stress-tested.
Draft, figures and a manuscript marked.
Reviewed, Calibre scored, novelty checked.
Journal fit, integrity screen, amplify.
Measuring the quality of science
Each manuscript review ends with a Calibre score, shown as a tier chip, but the number is only the entry point. Below it, the review gives the reasoning criterion by criterion: which parts are sound, which are not, and why.
The review ranks the issues by criticality, so the largest problems come first. Each issue has a specific fix that you can apply, which makes the score a map for the next revision.
Flawed Tier
Paper Score
Fundamental problems, stated openly.
0-9
Platinum Tier
Paper Score
Rigour at the level that demanding venues expect.
85-100
Gold Tier
Paper Score
Strong. Findings of the review are refinements.
70-84
Silver Tier
Paper Score
Sound core, named gaps to close before submission.
60-69
Bronze Tier
Paper Score
Mostly sound, with issues that weaken the claims.
50-59
Cobalt Tier
Paper Score
Notable weaknesses. Review is a map for the revision.
30-49
Tin Tier
Paper Score
Several central claims do not have sufficient support.
10-29
Flawed Tier
Paper Score
Fundamental problems, stated openly.
0-9
Platinum Tier
Paper Score
Rigour at the level that demanding venues expect.
85-100
Gold Tier
Paper Score
Strong. Findings of the review are refinements.
70-84
Unlocking scientific discovery.
No card required.