Project
Agora
State-first agent-based simulation lab: Markov populations, scored forecasts against target distributions, and a verifier-first shadow brain that stays powerless until held-out gates clear.
Status Active
Agora
Agora is my agent simulation lab: a vault-native ABM (agent-based modeling) OS where populations evolve under Markov rule dynamics, forecasts are scored against explicit target.json trajectories, and every world, rule, run, and channel is plain OKF files.
Full map of what and how: Agent Simulation Lab.
What it does
- Run named worlds (
opinion_diffusion,sir_epidemic,rumor_debunk) under rule chains. - Predict macro-dynamics with scored error against targets — not narrative success.
- Evolve rules toward those targets when the forecast loop fails.
- Shadow-predict with a CompactCTM-mapped brain that has no control authority until an external Brier / log loss gate clears.
What the evidence says
- Calibration Runs — mixed / provisional local demos (opinion diffusion, SIR, small evolve gains).
- CompactCTM Shadow Stage-1 — sealed negative: held-out CompactCTM loses to an empirical prior; not promoted; no Sakana CTM paper parity claimed.
The numbers are small on purpose. They exist so forecasts and rule changes can fail in public.
Authority boundary
Agora owns the simulated world and the measurement harness. Memory Store — the cognitive memory product I build at Julep — owns durable workspace context for real agents. Transient CTM-style thought is working computation, not that product. Source trees stay private; technical reports publish the sealed gates.
Reports
- Report
Agent Simulation Lab: What I Built and How
Agora as a state-first agent-based simulation lab: Markov worlds, scored forecasts, verifier-first shadow brains, and a sealed CompactCTM negative — with authority boundaries that keep Memory Store out of tick-level control.
- Report
Agora CompactCTM Shadow — Stage-1 Prediction Without Control
Held-out CompactCTM shadow predictor fails to beat an empirical prior on Agora opinion-diffusion transitions; claim class is non-parity, control authority false, not promoted.
- Report
Agora Calibration Runs
State-first simulation calibration on opinion diffusion and SIR worlds: scored forecasts and rule evolution against explicit targets, not vibes.