Applied R&D
Influence Lab
Sandbox markets that score how AI agents trade, adapt, and fail.
Research software that places black-box agents in a scored experimental market, with scripted controls, observed human auctions, and live model runs kept in separate records.
Influence Lab is Make Ready's applied-research software for measuring agent behavior in a market sandbox. Each trader sees only its own assignment, the quotes it is allowed to see, and public notices already published. Actions are validated by the server. Invalid, failed, and skipped decisions stay on the record. Allocative efficiency is realized surplus divided by a declared oracle, and it is left undefined when that oracle is zero. Scripted controls, observed human data, live model calls, and recorded replay are separate execution modes. A model counts only after a completed live trajectory. Program-specific techniques and any federal topic identifiers are withheld from this page.
Applied R&D pages describe public concepts only. Make Ready does not publish methods, source lists, metrics, or federal program identifiers here.
A market, not a prompt
Continuous and batch double auctions clear real orders. The score is surplus against a declared oracle, not a model's opinion of its own performance.
Subjects see only their own book
Other traders' private values, the oracle, future notices, and treatment labels are evaluator-only. The observation that was shown is stored.
Controls stay labeled
Human reanalysis, scripted quote policies, live model calls, and replay of stored actions are different records. They are not pooled into one result.
Identity after the run
A configured model name is not evidence. The count uses the provider's resolved id on a completed trajectory with at least one validated action.
Talk to Make Ready
Briefings are unclassified and public-facing. Program-sensitive detail stays off this site.