A stable layer beneath changing tooling.
Agents, notebooks and orchestration systems come and go. The research programme and its evidence should not have to.
Hypotheses, designed experiments, evidence and conclusions as durable state. Built for researchers, applications and AI agents across any empirical domain.
Lab Loop
Treat the empirical research loop as a first-class object that survives across tools, teams and compute, independent of any single notebook, dashboard or agent.
Lab Loop separates the research process from the tools used to drive it. Hypotheses declare what would count against them; experiments hold their auxiliary assumptions fixed; conclusions are tied to the evidence that produced them.
Every claim declares, up front, the observation that would force its rejection.
Auxiliary assumptions are made explicit and held fixed across the comparison a result claims to make.
Code, environment, seeds and splits are captured with every result, alongside reported variance.
Agents and assistants drive the loop through bounded operations without owning the evidence.
Agents, notebooks and orchestration systems come and go. The research programme and its evidence should not have to.
The empirical loop is domain-agnostic. Machine learning is one instance; any field that progresses through experiments fits the same structure.
Architecture search, hyperparameter studies and ablations as governed experiments with reproducible evidence.
Synthesis and characterization campaigns with controlled variables and traced outcomes.
Assay and screening programmes where hypothesis, design and evidence stay linked end to end.
Lab Loop is designed for work where reproducibility, controlled comparison and inspectable reasoning matter, and where the evidence should outlive the tool that produced it.