Cite
Notes
Only stored in your browser.
Attribution
Multi-hop QA over a ~32k-token document under a hard 600-token working-context ceiling: the agent must READ, COMPRESS and DROP to carry bridge enti...
Cost-aware diagnosis over synthetic patients: order tests (with committed result predictions), prescribe, then report a distribution over 149 condi...
Evidence-seeking under strong ideological priors in a fictional, exactly solvable news feed.