Research
Every claim here was pre-registered with a numeric threshold, graded against a frozen holdout, or both. Two of these studies report refuted hypotheses, and one reports a metric I had to mark instrument-invalidated after tracing it to a fixture defect. Those are in the results tables with everything else.
Published studies
May 2026
A federation of agents on a self-hosted 70B open-weight model produced operations-grade decisions on a contractor's real data with zero fabrications across 51 held-out decisions, and no frontier API anywhere in the decision path.
Reported against me: A divergence-from-control metric came back at 0% and was marked instrument-invalidated. The fixture's feedback hints had been authored to align with the control's default picks, so both arms reached the same answer for the right reason on each side. The metric was measuring the fixture, not the model. It was traced, documented, and given a surgical fix for the next run rather than quietly dropped.
July 2026
Vector-free graph retrieval refused 100% of adversarial distractors. Hybrid retrieval, on the same corpus and the same harness, refused 10%. Two of four pre-registered hypotheses were refuted and published anyway.
Reported against me: Two of the four pre-registered hypotheses were refuted. Both are in the results table with the rest.
July 2026
Graph, vector, and holographic memory in one SQLite store, characterized in vivo over 17 days: 1,976 recall events across 235 sessions at 37.4 ms median, with zero API calls on the hot path.
Evaluation environments
Task environments with hidden verification rather than written studies. The scorer is the primary artifact and the agents are swappable arms behind it.
July 2026
Extended thinking cut LangGraph recall from 1.000 to 0.600 while raising Vercel AI SDK precision from 0.714 to 0.938. Same model, same tools, same frozen data, opposite behavior by orchestration.
Reported against me: The LangGraph arm's 0.000 fabrication rate is structural, not earned. An evidence-check node drops any citation it cannot verify, so the zero holds regardless of what the model proposes. The two autonomous arms earn their zero from the model plus a strict schema, which is a weaker guarantee, and the writeup says so.
July 2026
The naive read says high-adoption accounts churn 30.4 percentage points less. Adjusting for account size puts the true effect at 11.7 points. The correction removes about 88% of the confounding bias, and the generator knows the right answer, so the correction is checkable.