Millions of Agents.
One Semantic World Model.

Week 1 (Apr 6–10) — The opening note. Every signal the model watches turns positive at once — equities, semis, commodities, even crypto. Inside the simulation, the policy class lights up: the agents shaped like central bankers, treasury officials, and IMF directors surge to the top of the population, as if the entire manifold has tuned itself to listen for Washington. Institutional agents anchor themselves deep in conviction territory, near the boundary of the Poincaré ball where regimes harden irreversibly. Retail begins to drift toward them, but cautiously. The simulation reads this as a coordinated risk-on regime, broadly believed — yet it already detects a quiet asymmetry: the elite class is far more committed than the household one.

Week 2 (Apr 13–17) — A break in the chord. The commodity signal flips overnight from full positive to full negative, while crypto stays at the ceiling. To the model, this looks like the fingerprint of a discrete policy headline — something out of Washington that catches supply expectations off guard. The interesting thing is what doesn't move: the institutional agents stay long commodities, refusing to unwind. The Treasury and Fed archetypes remain the loudest voices on the manifold. The simulation reads this as the smart money absorbing a policy event before retail registers the change at all.

Week 3 (Apr 20–24) — The dangerous calm. Market signals quiet down across the board. Nothing dramatic on the surface. But inside the simulation, something subtler is happening: agents are drifting further toward the edge of the manifold, the geometric region where conviction becomes irreversible. No new direction is being chosen, but the population is becoming more rigid in whatever it already holds. The model has seen this shape before — it usually precedes a regime transition, not a continuation.

Week 4 (Apr 27 → May 1) — The holding state. Signals turn mixed: equities cool, commodities flip up again, rates push higher. The population's collective conviction barely moves. The Treasury and Fed Chair archetypes are still the loudest voices, but their readings barely shift. The model interprets this as policy uncertainty without a dominant direction — the population is waiting. The first faint signal of retail decommitment appears: household-shaped agents begin, almost imperceptibly, to step back from the edge of the ball.

Week 5 (May 4–8) — The retail exodus begins in earnest. While the surface markets continue to rally, something quiet and unmistakable is happening inside the simulation: the retail agents — the part of the population shaped like households, day-traders, and ordinary savers — start systematically decommitting from the manifold's edge. They aren't selling loudly; they're just letting their conviction drift back toward the center. The institutional agents stay welded to the boundary. The gap between the two cohorts widens visibly for the first time in this window.

Week 6 (May 11–15) — The widening gap. From its April peak — when nearly four in five household-shaped agents stood past the model's commitment horizon — retail conviction has fallen by more than a third, and now sits at about half. Meanwhile every institutional category — banks, hedge funds, pensions, central banks — remains essentially fully committed. Inside the model, this is the widest household-versus-elite gap in the window the simulation has been tracking in detail. Two new things appear: the Activist Investor archetype rises into the top voices for the first time, and the agent population's collective sentiment on the dollar reaches its most bearish reading of the period. The model is describing a world where executive-branch policy is the price-forming entity, where households have stepped back, and where corporate boardrooms are being primed for public political action they cannot yet take. The predictions the model issued at the open of the week — short small caps, long energy, long Treasury yields rising, long semis — produced a positive week.