Research
We run our research as four programs of inquiry — generative AI stack, personal AI CEO, sovereign AI system, generative culture. Each ships one report per week with the artefacts attached. Replication is the point.
Programs
4 activeArchive
Claude 4.7 wins voice-fit by a 23% margin on first draft. Gemini 3 Pro wins on factual density. GPT-5.1 wins on structural variety. The honest verdict: model choice depends on what you're optimising — and most creators are optimising the wrong thing.
Read →Verifier-loop wins on irreversible work (publish, send, charge) by 38% on quality at ~3× cost. Pipeline wins on multi-format expansion (one essay → eight outputs) by 22% quality at 1.4× cost. Pure swarm loses to a single Sonnet on every metric except ego.
Read →Five incidents, three root-cause categories — silent schema drift, auth-token rotation gaps, transport-layer back-pressure. A 12-line pre-flight check catches the patterns. Most production MCP failures are operational, not architectural.
Read →We are pre-registering this design before the first cohort runs, so the result lands without hindsight bias. Hypothesis published. Method published. Reproduction kit attached. We will publish the result whether it confirms or falsifies the claim.
Read →Cadence
Each report follows one structure: hypothesis, method, setup, results, takeaway, next. Raw artefacts attached. Replication welcome. Members of the Lounge get the report on Monday — twenty-four hours before public.