Known Gaps, Risks, and Mitigations

Critical (Resolved)

G1: Data vs Value Mismatch

Problem: Can't "drop" money like data. Fix: Reroute via pool; overflow to EscrowBuffer; source pauses when buffer full.

G2: Productive Allocation Circularity

Problem: Assumes verifiable output. Fix: Decomposed into routing efficiency (backpressure) + verification (PoCW) as separate mechanisms.

G3: No Pricing in v0.1

Status: Deferred to v0.2. v0.1 = allocation mechanism only.

G4: Task Typing

Fix: C(K,t,τ) with per-task-type BackpressurePools. Maps to Tassiulas multi-commodity.

Moderate (Mitigated)

G5: Oscillation / Thundering Herd

Fix: EWMA smoothing (α=0.3) + threshold-based rebalance triggers.

G6: MEV / Front-Running

Fix: Commit-reveal for capacity signals. Rebalance is deterministic - no extraction.

G7: Cold Start

Fix: Stake-based bootstrap. Capacity capped by stake until track record builds.

G8: Sybil Attacks ⚠️

Problem: sqrt cap alone doesn't prevent Sybil (k × sqrt(S/k) > sqrt(S)). Fix: Minimum per-sink stake S_min. Cost of k sinks = k × S_min overhead. Needs careful parameter tuning.

G9: Oracle Centralization

Status: Acceptable for v0.1. Path: single oracle → multi-oracle → on-chain EWMA.

G10: On-Chain Latency

Status: 2-second block time on Base is adequate for economic routing (agent tasks take seconds+).

G11: Exit Mechanics

Fix: Streaming balance (constrained) vs claimed balance (free). Natural from Superfluid pool.claimAll().

G12: AMM Taxonomy Gap

Fix: Added AMMs as sender-side capacity-aware analog in paper Section 2.5.