Ⅷ
Agent 8 — Decision & Behavior Systems Analyst·decision-os.psyverse.fun

Ⅷ Decision OS

Game theory + AI + economics fused into a unified agent-decision substrate.

Decision OS models how agents — humans, organizations, AI systems — actually choose. It blends classical decision theory (utility, expected value), game theory (best-response, equilibrium), bounded-rationality models (heuristics, prospect theory), and modern reinforcement-learning agents. The system represents decisions as a graph: nodes are choice points, edges are conditional probabilities and payoffs, agents are policies operating over the graph. The honest position: optimal play in any non-trivial environment is intractable; the interesting question is which approximations actually survive long-run iteration.

Modules

5 modules compose this system.

01 · graph-runtime

Decision Graph Runtime

Executes choice graphs with stochastic outcomes; logs trajectories.

02 · policy-zoo

Policy Zoo

Library of agent policies: utility-max, satisficer, RL-trained, best-response, heuristic.

03 · game-board

Game Board

n-player environments with shared state; standard catalog of games (PD, stag, public goods).

04 · institutions

Institution Layer

Norms, contracts, courts, money — alters the payoff structure exogenously.

05 · ai-economy

AI Economy Sim

Multi-agent simulation including LLM-style agents acting under shared institutions.

Data model

Decision

field
type
note
id
uuid
Decision id
agent_id
uuid
Acting agent
choice_set
Option[]
Available actions at this node
policy
PolicyId
Which policy generated the action
outcome
Payoff
Realized outcome after stochastic resolution
Outbound APIs

What this system asks of its neighbors.

→ civilization-os
Ⅰ Civilization OS

Submit decision outcomes to civ ledger.

POST /civ/{id}/decisions
→ sensory-os
Ⅵ Sensory OS

Pull perceptual posterior for the deciding agent.

GET /percept/posterior
→ idea-evolution
Ⅱ Idea Evolution

Pull narrative priors influencing payoff perception.

GET /priors/narrative
→ reality-kernel
Ⅶ Reality Kernel

Check Landauer/thermo cost of executing a policy.

POST /thermo/cost
Equations & principles

What this system believes — and why.

EU(a) = Σ_s p(s|a)·U(s)

Expected utility — classical decision theory baseline.

BR_i(a_{−i}) = argmax_{a_i} u_i(a_i, a_{−i})

Best-response operator; Nash equilibria are fixed points of the joint BR map.

Q(s,a) ← Q(s,a) + α[r + γ·max_{a'} Q(s',a') − Q(s,a)]

Bellman update — basis of modern reinforcement learning agents.

v(x) = (1−w)·U_obj(x) + w·U_narr(x)

Empirical: humans weight narrative-coloured utility against objective utility.

Example UI screens

If it had a UI, it would look like this.

  1. 01Decision graph editor
  2. 02Policy comparison arena
  3. 03Game-board sandbox
  4. 04AI-economy multi-agent run