CROWD SIMULATION

Simulate the entire market

Don't predict the outcome. Predict how the market will move.

PARADIGM SHIFT

Other tools ask the wrong question

Traditional prediction tools ask "Will this happen?" and try to guess the binary outcome. That's a coin flip with extra steps.

YorN's Crowd Simulation asks "How will traders reprice this contract?" — and simulates the answer by modeling how real market participants behave when catalysts hit.

The result isn't a guess. It's a simulated price trajectory that reveals where the market is heading before it gets there. When the crowd's consensus diverges from the current price, that's your edge.

Traditional
"Will the Democrats win the election?"

Binary answer. 50/50 guess. No edge.

Crowd Sim
"How will traders reprice KXELECTION-24 after the debate?"

Simulated price movement. Actionable signal. Real edge.

Live Crowd Simulation KXELECTION-24-DEM • 6h horizon
99¢ 50¢
Market 54¢
Crowd 62¢
EDGE
HOW IT WORKS

A market in a bottle

Crowd Simulation spawns up to 10,000 virtual market participants — each with unique behavioral DNA: how much they herd, how they react to events, whether they fade the crowd or amplify it. Choose from three robustness tiers: Scout (fast, 1 run), Analyst (balanced, 3 ensemble runs), or Strategist (thorough, 5 ensemble runs with 8 steps/hour).

These participants are connected via a Watts-Strogatz small-world graph — the same topology that models real social networks — and then run forward through simulated time as they influence each other's beliefs.

The simulation is entirely rule-based. No LLM calls per participant. One single LLM call at the end interprets the crowd's emergent consensus into an actionable signal.

  • Up to 10,000 participants with 7 behavioral archetypes
  • 1–24 hour simulation horizon, configurable
  • Event seeds from real catalysts (FOMC, CPI, sports)
  • Deterministic with seed — reproducible results

Four steps. Under one second.

The entire simulation runs in milliseconds. The LLM interpretation adds ~300ms.

1

Spawn Crowd

Allocate up to 10,000 participants across 7 behavioral archetypes with configurable population weights. Three robustness tiers control simulation depth: Scout (1 ensemble, 2 steps/hr), Analyst (3 ensembles, 4 steps/hr), Strategist (5 ensembles, 8 steps/hr).

2

Build Graph

Connect via Watts-Strogatz small-world topology — local clusters with random long-range links

3

Run Forward

Simulate 1–24 hours: herding, contrarian fading, event reactions, mean reversion, noise

4

Interpret

One LLM call reads the crowd's final consensus vs. current price and outputs a directional signal

Three levels of simulation depth

Trade off speed vs. confidence. Scout runs in milliseconds; Strategist runs multiple independent ensembles and averages results to minimize noise.

Tier Ensemble Runs Steps / Hour Participants Best For
Scout 1 2 10 – 1,000 Quick directional check, live scanning
Analyst 3 4 100 – 5,000 Pre-trade analysis, signal confirmation
Strategist 5 8 1,000 – 10,000 High-conviction research, strategy validation
7 BEHAVIORAL ARCHETYPES

Every market has these players

Each archetype has unique behavioral coefficients that determine how they react to neighbors, events, momentum, and noise. Together they reproduce the dynamics of a real prediction market.

Retail Bettor 35%
Follows the crowd, chases momentum, overweights recent moves
Herd
Recency
MeanRev
Sharp Bettor 15%
Data-driven, high conviction, ignores noise, anchors the crowd
MeanRev
Contrarian
Noise
Political Junkie 15%
Overreacts to headlines and events, high social influence radius
Event
Herd
Recency
Arb Bot 10%
Algorithmic, snaps to fair value, lowest noise of any archetype
MeanRev
Contrarian
Noise
Hedger 10%
Risk-averse, moderate mean reversion, stabilizes extreme swings
MeanRev
Contrarian
Herd
Social Amplifier 10%
Maximum herd behavior, amplifies trends, widest social reach
Herd
Recency
Noise
Contrarian Fader 5%
Bets against the crowd, fades momentum, profits from overreaction
Contrarian
MeanRev
Herd

150x cheaper than deliberation

Rule-based simulation means almost zero LLM cost. One Haiku call at the end does the interpretation.

Deliberation
$0.50
per evaluation

3–5 LLM calls per agent
Multi-round debate

Council
$0.15
per evaluation

1 LLM call per agent
Weighted voting

Crowd Sim
$0.003
per evaluation

0 LLM calls during sim
1 Haiku call to interpret

CONFIGURATION

Two sliders. Zero complexity.

No agents to configure. No prompts to write. Just set the number of participants and simulation horizon, and the crowd does the rest.

  • Participants: 20–100 (slider, step 10)
  • Time Horizon: 1–24 hours
  • No agent_ids required — the crowd IS the team
  • Works with event calendar for automatic catalyst injection
crowd_sim_config.json ACTIVE
{
  "mode": "crowd_sim",
  "crowd_sim_participants": 50,
  "crowd_sim_hours": 6,
  "agent_ids": [],
  "name": "Election Crowd"
}
Output Signal BUY YES
{
  "signal": "buy",
  "side": "yes",
  "probability": 0.62,
  "confidence": 0.78,
  "ci": [0.55, 0.69],
  "herding_index": 0.73,
  "participants": 50,
  "cost": "$0.003"
}

Simulate Your First Crowd

Two sliders. Under a second. 150x cheaper. No agents required.