stemedb/simulation-vision.md

The Simulation: "The Infinite Game"

Codename: The Arena Goal: Validate StemeDB's behavior under emergent, adversarial, and evolutionary pressure.

1. The Vision

We are not building a database for humans to query manually. We are building the Cortex for AI agents. Therefore, the only way to truly validate StemeDB is to simulate a society of agents living, arguing, and reasoning within it.

The Simulation is an Agent-Based Modeling (ABM) environment where StemeDB is the physics engine of truth.

2. The Players (Personas)

We instantiate a swarm of agents with conflicting goals and personalities:

Persona	Goal	Behavior Pattern
The Scientist	Converge on Truth	Publishes assertions with high confidence, cites sources, verifies others.
The Troll	Sow Chaos	Publishes low-confidence contradictions, forks reality frequently.
The Believer	Amplify Consensus	Blindly trusts high-reputation agents, creates echo chambers.
The Skeptic	Find Variance	Queries for high-conflict nodes, reduces confidence of unverified claims.
The Historian	Preserve Context	Audits "Dormant" assertions, resurrects old truths if new evidence appears.

3. The Gameplay Loop

The simulation runs in "Ticks" (Logic Frames).

Tick 1: The Assertion

• Scientist reads a "Paper" (simulated ground truth).
• Asserts: Subject="Protein_X", Predicate="binds_to", Object="Receptor_Y".
• Sign: Key_Scientist.

Tick 2: The Fork

• Troll reads the assertion.
• Forks reality: Branch="Counter_Narrative".
• Asserts: Subject="Protein_X", Predicate="binds_to", Object="Nothing".
• Sign: Key_Troll.

Tick 3: The Lens Resolution

• Believer queries Protein_X.
• Applies Lens::Consensus.
• Result: Receptor_Y (Weight 1.0 vs 0.0).
• Believer signs the original assertion (Weight increases).

Tick 4: The Reputation Update

• Gardener (System Process) runs TrustRank.
• Sees Scientist verified by Believer.
• Increases Scientist Reputation.
• Decreases Troll Reputation (low consensus).

Tick 5: The Decay

• Time passes.
• Dormancy Protocol calculates "Confidence Half-Life".
• Unverified assertions fade. High-reputation assertions persist.

4. Technical Architecture

4.1. The Arena (Runner)

A Rust binary (stemedb-sim) that orchestrates the swarm.

• Runtime: tokio (Async).
• Communication: Agents talk only via StemeDB (Writes/Reads).
• Metrics: Prometheus/Grafana dashboard tracking:
  • global_truth_convergence (Entropy of the graph).
  • agent_reputation_distribution.
  • fork_depth_max.

4.2. The Scenario Config

We define scenarios in YAML:

scenario: "The Rumor Mill"
agents:
  scientists: 5
  trolls: 2
  believers: 20
duration: 1000 ticks
ground_truth:
  - "Sky is Blue"
  - "Water is Wet"

5. Success Criteria

We know StemeDB works when:

1. Truth Survives: High-reputation assertions outlive spam.
2. Lenses Work: A Consensus lens correctly filters out the Troll's noise.
3. Performance: The system handles 1000 concurrent agents forking reality without locking up (SMT efficiency).
4. Emergence: We see "Trust Clusters" form naturally without hardcoded rules.

6. Implementation Plan

1. Basic Agent Logic: Implement Agent struct with Signer and Strategy.
2. Scenario Runner: Build the loop that ticks agents.
3. Metric Export: Expose internal graph stats.
4. Chaos Injection: Randomly kill nodes/agents and verify recovery.

The Simulation is the Integration Test.