Interactive Research Experiment · Agent-Based Simulation
Talent is necessary. Luck decides who wins.
1,000 agents · 40 simulated years · multiplicative capital dynamics. The Pluchino–Biondo–Rapisarda model shows how random event paths can generate heavy-tailed outcomes from Gaussian talent. The DCG extension explores what changes when agents can navigate the opportunity field instead of waiting for it.
Research boundary: the Original Model reproduces the published low-agency simulation. The DCG Extension is a separate exploratory scenario for causal geometry and structural agency; it is not a result reported by the original paper.
—
Luck vs Talent
corr ratio
corr ratio
—
Winner Talent
max possible = 1.0
max possible = 1.0
—
Gini Index
from perfectly equal start
from perfectly equal start
—
Agents ending
below starting capital
below starting capital
01
Simulation
Interactive agent-based model
2×
Scenarios
Pick a scenario for a one-click configured experiment — or tune the parameters manually below.
VS
ORIGINAL MODEL
N=1000 agents (fixed) · NE=500 events (random walk) · 201×201 toroidal · 40 years · Δt=6 months
Lucky
Unlucky
Agent
Winner
Max Talent
Step0/80
Time0y
Gini0.000
Max Capital10—
Corr T↔C—
Corr L↔C—
02
Key Metrics
Real-time experimental output
Winner Capital——
Winner Talent——
Corr(Talent, log C)—Pearson correlation
Corr(Luck, log C)—Pearson correlation
Gini Index——
Top 20% Share—Pareto ratio
03
Distribution Analysis
Visual diagnostics
Capital Distribution (log₁₀)
Heavy tail emergent from multiplicative dynamics. Pareto-like power law under paper parameters.
Talent vs Final Capital
Central misalignment: the most successful agent is rarely the most talented one.
Lucky Events vs Final Capital
Strong positive correlation: luck explains more outcome variance than raw talent.
Top 5 Trajectories (log scale)
Capital paths over 40 years. Late exponential explosion from clustered lucky events is typical.
System Evolution — Inequality & Correlation Over Time
Gini rises as multiplicative dynamics concentrate capital, while the luck–capital correlation pulls away from the talent–capital correlation as event history accumulates. The gap between the green and blue lines is the quantitative signature of luck dominance.
04
Research Foundations
Original paper & DCG extension
Original Paper
Talent versus Luck: The Role of Randomness in Success and Failure
The paper builds an agent-based model where N=1000 agents with normally distributed talent (μ=0.6, σ=0.1) interact with random lucky and unlucky events over 40 simulated years. When a lucky event reaches an agent, capital doubles only if a random draw falls below the agent's talent — talent acts as a probabilistic filter on opportunity. Unlucky events halve capital unconditionally.
The key result: despite talent being Gaussian, the final capital distribution follows a Pareto-like power law. The most successful agent typically has mediocre talent (~0.61), while the most talented often ends below starting capital. Across 100 runs, agents with T>0.8 are best performers in only ~3% of cases. The paper concludes that talent is necessary but not sufficient — realized event sequence determines extreme success.
C(t) = C₀ · 2^(S(t) − U(t)) where S = converted lucky events, U = unlucky events
Read on arXiv →
DCG Extension
Dynamic Causal Governance: From Randomness to Structural Navigation
DCG reframes the TvL result through a structural lens: luck is unresolved causal geometry. What appears random from the agent's perspective is, at the system level, a structured opportunity field — a distribution of favorable and unfavorable events across space, time, networks, and institutions.
The extension introduces three levels of agency: (1) Talent Optimization — improve the conversion function ρ(T); (2) Position Optimization — navigate toward regions with higher λ⁺ and lower λ⁻; (3) Field Governance — alter the opportunity geometry itself. The original TvL is a zero-agency baseline where u(t)=0. DCG adds strategic movement, cumulative feedback, and structured opportunity clusters.
d/dt log C(t) = ρ(T)·λ⁺(x,t) − λ⁻(x,t) + κ·Φ(C,x,t) + u(t)·∇Ψ(x,t)
Read the DCG Framework →
How DCG Extends the Original Result
The original paper proves the mechanism: multiplicative dynamics + random events → extreme inequality even from Gaussian inputs. DCG takes this further by asking: if events are not truly random but structurally distributed, can agents who understand the field outperform those who are merely talented?
Original TvL
Agents are fixed in space. Events arrive randomly. Success = Talent × Random Event Path. No strategic agency.
Success = f(T, Random Events)
→
DCG Extension
Agents navigate the opportunity field. Feedback loops amplify. Success = Conversion × Opportunity × Trajectory × Feedback.
Success = ∫ᵧ [ρ·λ⁺ − λ⁻ + κΦ + u·∇Ψ] dt
05
Model Parameters
Adjustable experimental inputs
Original Paper (Pluchino et al., 2018)
DCG Extension
06
Discoveries & Insights
What the experiment reveals
Counterintuitive
The most talented almost never wins
The best performer typically has talent ≈ 0.61, barely above the population mean of 0.6. Agents with T>0.8 are winners in only ~3% of runs. Maximum talent ≠ maximum success.
Run experiment to see data
Central Result
Luck explains more variance than talent
The Pearson correlation between lucky events and log-capital is systematically higher than between talent and log-capital. Talent raises expected value; luck determines the realized path.
Run experiment to see data
Emergence
Equal start → Pareto inequality
All 1000 agents begin with C=10. After 40 years of multiplicative dynamics, the top 20% controls ~80% of total capital. This Pareto-like concentration emerges purely from random event exposure.
Run experiment to see data
DCG Extension
Navigation triples talent expression
When agents can navigate the opportunity field, talent↔success correlation increases ~3×. Structural navigation allows talent to convert more opportunities. Luck still dominates, but less.
Run experiment to see data
Structural Insight
Talent is an operator, not a stock
Talent should be modeled as T: Opportunity → Gain. It transforms opportunity into capital, but does not generate opportunity. A genius in a low-opportunity field stays invisible. Position matters more than raw ability.
DCG Framework
Three levels of strategic agency
Level 1: Optimize conversion ρ(T) — skill, effort. Level 2: Optimize position — move to higher λ⁺ regions. Level 3: Govern the field — create platforms, institutions, opportunity flows.
Each level structurally dominates the previous. Most people focus on Level 1; the highest leverage is Level 3.
Policy Implication
Naive meritocracy fails
Rewarding past success may reward accumulated luck, not merit. The paper shows egalitarian funding outperforms elitist funding in surfacing latent talent. Random allocation among qualified candidates performs surprisingly well.
Try it: Run Multi-Run 50× and compare winner talent distributions. The most talented agents are rarely at the top.
Empirical Support
Random impact in scientific careers
Sinatra et al. (2016) showed that a scientist's highest-impact paper appears randomly in their publication sequence. Fortin & Currie (2013) found that impact per dollar is lower for larger grants. Both support the TvL mechanism.
Counterintuitive
Late explosion, not gradual growth
The typical winner's trajectory shows low capital for most of their career, followed by a sudden exponential rise from clustered lucky events in the final years. Success timing is unpredictable.
Try it: Watch the Top 5 Trajectories chart during Play mode. Notice how winners often emerge very late.
Mathematical Core
Multiplicative dynamics create power laws
Since C(t) = C₀ · 2^(S−U), log-capital is a random walk. Small differences in event history become exponential differences in outcomes. This is why Gaussian talent coexists with Pareto wealth.
Try it: Increase lucky % to 70 or decrease to 30. Watch how the tail shape and Gini coefficient change dramatically.
DCG Insight
Luck = unresolved causal geometry
What the agent calls "luck" is a structured field of events that can be mapped, navigated, and governed. The DCG extension makes this visible: opportunity clusters glow on the canvas. Agents who navigate toward them convert more.
Try it: Switch to DCG mode and run Play. Watch agents migrate toward the purple opportunity zones and accumulate faster.
Key Equation
Success is an integral over a trajectory
The DCG formulation: Success = ∫ [Conversion × Opportunity − Adversity + Feedback + Navigation] dt. Intelligence is local; causal geometry is global. The global usually dominates unless the agent learns to navigate the field.
Experiment Dynamics
Parameters reveal structural sensitivity
The model's behavior changes qualitatively with small parameter shifts. This is not noise — it reveals the structural sensitivity of multiplicative systems.
Try these: Set σ=0.01 (near-equal talent) — luck dominance becomes extreme. Set % Lucky=80 — inequality drops as opportunity floods. Set N=100, NE=1000 — every agent gets hit, reducing positional advantage.
Live Tracking
Fate of the most talented agent
The cyan T★ marker on the canvas tracks the single highest-talent agent in the population. In most runs they finish mid-pack — sometimes below starting capital. Genius without event exposure is statistically invisible.
Run experiment to see data
Try it: Hover any dot on the canvas to inspect that agent — click to pin it and follow its fate.
Hidden Asymmetry
Most agents end with less than they started
Doubling requires a lucky hit and a successful talent draw; halving is unconditional. This asymmetry means the median trajectory is decline — the spectacular winners are paid for by a silent majority of losers.
Run experiment to see data
Winner Anatomy
Winners are outliers in exposure first, conversion second
Dissecting the winner's history shows they were hit by far more lucky events than average — being repeatedly in the right place precedes converting well. Talent only filters what exposure delivers.
Run experiment to see data
Counterfactual
Re-run the universe, crown a different king
Identical talent distribution, identical rules — a different random seed produces a completely different winner. If success replayed from talent alone, the same agents would rise every time. They don't.
Run experiment to see data
Try it: Change Seed to 7, 99, 2024 and Run Complete each time. Watch who takes the crown.