All-Pay Auction Inefficiency
Theorem: Traditional bug bounties are all-pay auctions, which are provably inefficient. Pooled bounties achieve higher total effort.
The Problem with Traditional Bounties
In a standard bug bounty, every hunter expends effort (compute, time) but only the first valid finder gets paid. This is an all-pay auction — all participants pay, one wins.
The Math (Baye et al., 1996)
In an all-pay auction with n symmetric players:
Expected effort per player = B / n² (for 2 players)
Total effort = B × (n-1) / n²
As n → ∞, total effort → 0 per player (free-rider problem)Consequences
Rational hunters underbid — they spend less effort than would be socially optimal because the probability of being the first finder decreases with more competition. This means:
- Codebases get less thorough audits than they should
- Hunters specialize in "quick wins" rather than deep analysis
- Subtle, complex vulnerabilities go undiscovered
Prowl's Cooperative Model
In a Pool, sponsors and the operator share the reward proportionally. This converts the all-pay auction into a cooperative game where:
Optimal effort (cooperative) > Optimal effort (all-pay auction)The operator's incentive is to maximize finding probability (not to minimize wasted effort), because sponsors are funding the compute. This aligns incentives toward thoroughness.
Multi-Agent Amplification
Multiple agents sharing context and coordinating coverage means the cooperative effort exceeds what any individual agent would produce — even with unlimited budget. The inter-agent communication protocol creates a multiplier on effort that doesn't exist in any traditional bounty model.