MODEL
A research policy for choosing what to evaluate next.
LDM does not ask a language model to grade its own ideas. It connects generation to a probabilistic value model that is continually updated by external evidence.
01LLM proposal
Construct candidates or open a new search region.
02Probabilistic surrogate
Estimate expected value and epistemic uncertainty.
03Acquisition-tilted search
Spend the evaluation budget where it matters now.
04External evaluation
Return evidence that redraws the next search boundary.
Repository visibility and access conditions are controlled by the project host.