Praxist: From Experimental Artifacts to Solution Lineages
Praxist addresses a critical limitation in autonomous research systems: they generate and score many artifacts but discard intermediate insights that could be recombined later. The system separates the artifact that was evaluated from the claim that should be inherited, representing research as a recurrent transformation from artifact to finding, frontier, agenda, and lineage. Evaluated on all 75 MLE-bench tasks and four open-ended engineering studies, Praxist demonstrates that the reusable unit of autonomous research is not the score-ranked artifact but the artifact-grounded finding.Script
Most autonomous research systems throw away everything except the highest-scoring results. But what if a failed experiment contains the key insight that makes the next breakthrough possible?
Praxist converts each artifact into typed findings that carry the intervention, outcome, evidence maturity, limitations, and inheritance action. A negative finding that rules out an assumption can be more valuable than a positive result with immature evidence.
Each generation inherits a frontier of evidence, constructs deliberately diverse experiments under design contracts, evaluates the artifacts, and synthesizes new findings that update the frontier and shape the next agenda.
On all 75 MLE-bench tasks, Praxist achieved 60 medals including 49 golds, compared with 55 medals and 34 golds for the baseline. It reached this result with approximately 3,000 dollars in model spend versus 38,000 for the baseline.
Four open-ended case studies demonstrate the method's range: 100 percent landing success on a deterministic rocket simulator, 53 percent compounded annual return in walk-forward quantitative trading, 72 percent visual workload reduction in SLAM, and improved common-horizon tracking in tokamak magnetic control.
The reusable unit of autonomous research is not the score-ranked artifact but the artifact-grounded finding. Visit EmergentMind.com to explore the full lineage and create your own research videos from the latest papers.