Background and Related Work¶
Backpressure Routing¶
Tassiulas and Ephremides (Tassiulas & Ephremides, 1992) introduced the max-weight scheduling algorithm for multi-hop wireless networks. At each time slot, the scheduler computes the differential backlog across each link and activates the schedule that maximizes the weighted sum of link capacities and backlog differentials. This achieves throughput optimality: any arrival rate vector strictly inside the network capacity region can be stabilized. Neely (Neely, 2010) extended this framework via Lyapunov optimization, introducing the \(V\)-parameter tradeoff between queue backlog (delay) and utility maximization. Georgiadis et al. (Georgiadis et al., 2006) provide a comprehensive treatment of cross-layer resource allocation using backpressure principles.
Network Pricing and Congestion Control¶
Kelly et al. (Kelly et al., 1998) showed that network rate allocation can be decomposed into a system problem (shadow prices on link capacities) and user problems (utility maximization subject to prices). The shadow prices emerge as Lagrange multipliers on capacity constraints—an insight that directly bridges network routing and economics. Srikant (Srikant, 2004) formalized TCP congestion control as distributed optimization, with AIMD (Additive Increase Multiplicative Decrease) serving as a primal-dual algorithm. Jacobson (Jacobson, 1988) established EWMA for round-trip time estimation in TCP, providing the practical precedent for our capacity smoothing mechanism.
Token Engineering¶
Zhang and Zargham (Zhang & Zargham, 2020) take a dynamical systems approach to token economy design, modeling token flows as differential equations and using the cadCAD framework for simulation. Superfluid (Superfluid, 2024) provides on-chain streaming payment primitives: Constant Flow Agreements (CFA) for point-to-point streams and General Distribution Agreements (GDA) for one-to-many proportional splits. BPE adds a flow-control layer above these streaming primitives.
Demurrage Economics¶
Gesell (Gesell, 1916) proposed demurrage (a holding tax on currency) to increase monetary velocity. Fisher (Fisher, 1933) formalized this in Stamp Scrip. Both are stock-based velocity mechanisms: they penalize holding money. BPE is fundamentally different—a flow-based allocation mechanism that routes payments to where capacity exists. The two approaches are orthogonal and composable: demurrage encourages spending velocity, while BPE ensures efficient allocation of the resulting flows.
Sender-Side Capacity Awareness¶
Automated market makers (AMMs) like Uniswap (Adams et al., 2021) implement sender-side price discovery: the bonding curve encodes available liquidity, and senders observe the price impact of their trades. This is sender-side capacity awareness. BPE provides receiver-side capacity awareness: sinks (service providers) signal their capacity, and the protocol routes payments accordingly. The distinction is critical for service economies where the capacity constraint lies at the receiver, not the liquidity pool.
AI Agent Payment Protocols¶
The 2024–2026 wave of agent payment protocols addresses authorization and identity: Google’s AP2 (Google, 2025) handles inter-agent authentication, Coinbase’s x402 (Coinbase, 2025) uses HTTP 402 for single-shot payments, OpenAI–Stripe’s ACP (OpenAI & Stripe, 2025) provides billing and invoicing, and Visa’s TAP (Visa, 2025) extends card-network authorization to agents. Cha et al. (Cha et al., 2025) study LLM inference pricing via self-play. None of these systems address flow control—the dynamic rerouting of payment streams based on real-time capacity. BPE fills this gap.