AbMaze

credits spent 150 / 150

Lab notebooks

Seen = the noisy readout the agent got · true = hidden lab average.

Best variant found so far, by credits spent

How the environment works

Five tools share one credit pool; only measure touches lab data.

AGENT LLM or scripted strategy CR9114: 150 creditsCH65: 250 credits TOOL COST describe_system0 predict0.05 measure1 / antigen notebook0 submit ≤3, rankedends call → result rough computer model misses interactions; capped to assay range measured library 65,536 × 3 replicates 1 replicate "no binding" below limit HIDDEN FROM AGENT scoring clean average of the first-ranked variant closeness to best, 0–100% true values
Names are hidden from the agent ("Antigen 1", "option 0/1").

Leaderboard