[ field_notes // lab_output ]
From the lab
Shorter, more frequent writing than our papers: methods, post-mortems, and opinions formed under adversarial pressure.
Release·2026.10.19·6 min
Launching the Coffee Law: Agent Collectives Have a Knowledge Budget
A team of agents cannot know more than its starting data plus what its tools can carry. Debate and voting add nothing to that budget. Today we are releasing the paper, the proofs, and the scripts that check every number.
Read →Evaluation·2026.09.18·7 min
The Social Intelligence Gap: Benchmarking Agents as Social Actors
Lacuna's research direction on social intelligence in multi-agent systems shows models that win without persuading, cooperate without understanding, and scale without becoming more social. The Bhatt Conjectures explain why.
Read →Evaluation·2026.09.18·6 min
A Score Is Not a Capability: Reading the Bhatt Conjectures
Lacuna's reading of the Bhatt Conjectures gets to the central problem in AI evaluation: passing a benchmark is evidence, not proof, of reasoning or understanding.
Read →