Why Base's Liquidity Graph Leaks Value

Published on Fri Aug 14 2026 00:00:00 GMT+0000 (Coordinated Universal Time) by Jacob Cavazos

The Graph

Base has six major AMM DEXs. Uniswap V2. Uniswap V3. Uniswap V4. PancakeSwap. Aerodrome. SushiSwap. Each maintains its own set of liquidity pools for the same token pairs. USDC/WETH alone has active pools on at least four of these venues. Each pool has different fee tiers. Each pool has different reserve balances. Each pool has a different price quote at any given moment.

Add Curve for stablecoin pairs. Add Balancer for weighted pools. Add the RFQ venues — Bebop, Wintermute, Liquorice. The liquidity landscape for a single token pair can span eight or more distinct venues.

This is fragmentation. Fragmentation is also opportunity. Every price discrepancy between venues is a potential arbitrage. A path through the graph where the output exceeds the input after fees and gas. The question is whether the discrepancy is large enough to capture. The question is whether it persists long enough to execute against.

The Two Kinds of Discrepancies

We have observed two categories of price discrepancies on Base. They behave differently. They require different approaches.

The first category is transient. Momentary price differences caused by normal trading activity. A large swap on Uniswap V3 moves the pool price. For a brief window, Aerodrome’s pool for the same pair is quoting a better price. An arbitrageur swaps on Aerodrome. Captures the difference. The window closes in seconds. Other arbitrageurs notice. Other arbitrageurs act. The next block brings new trades that re-align prices. These discrepancies are small. Five to thirty bps. Brief. Seconds. Highly competitive. Multiple bots are watching. Multiple bots are acting.

The second category is persistent. Price differences that remain open for minutes. Hours. We documented one instance in our edge-persistence article. A LINK/USDC spread on Polygon that sat open for 100 minutes. Peak profit: $97.44. No bot touched it.

The reason is structural. Mainnet MEV whales do not bother with small opportunities on Layer 2s. The gas costs on Ethereum mainnet make $97 opportunities invisible. On Base, where gas is sub-cent, $97 is worth capturing. But the bots that could capture it are not watching Base closely enough. Their thresholds are set too high. They do not have the routing infrastructure to reach the specific pools where the discrepancy lives.

Persistent discrepancies are the interesting category. They suggest that Layer 2 markets are not yet efficient. There are structural inefficiencies that persist because the arbitrage infrastructure has not fully migrated from mainnet. As Base’s TVL increases, these opportunities will shrink. As Base’s activity increases, these opportunities will shrink. But right now, they exist. They are measurable. They are documented.

The Graph Model

Think of it as a directed graph. Tokens are nodes. Pools are edges. Each edge has a cost. The pool fee. Each edge has a conversion rate. Determined by the pool’s reserves and AMM formula. A swap from token A to token B is a path through the graph. An arbitrage is a cycle. A path that starts and ends at the same token. A path where the output exceeds the input.

A direct swap is a single edge. A two-hop route uses an intermediate token. A circular arbitrage is a three-edge cycle. The graph for Base’s major tokens has hundreds of nodes. Thousands of edges. The number of possible cycles grows combinatorially with the number of hops.

The challenge is not finding cycles. The challenge is finding profitable cycles. A cycle is profitable only if the output exceeds the input after all pool fees, gas costs, and price impact. Most cycles in the graph are not profitable at any given moment. The profitable ones are rare. The profitable ones close quickly once someone executes against them.

This is why routing infrastructure matters. A system that only checks direct pairs will miss two-hop and three-hop routes. A system that checks all possible routes but takes too long to execute will find opportunities that have already closed. The value is in finding profitable routes fast enough to capture them. The method is observe, model, simulate, predict, execute. The execution must be fast enough to matter.

What the Existing Posts Documented

The money printer post documented a specific type of persistent discrepancy. Oracle spreads. Oracle-managed reference pools update their prices at discrete intervals. Market pools update continuously via trading activity. When the market moves but the oracle has not updated, there is a window. The oracle pool is quoting a stale price. The market pool is quoting the current price. That window is an arbitrage opportunity.

The edge-persistence post documented that these windows persist much longer on Layer 2s than on mainnet. On mainnet, MEV bots close arbitrage windows in seconds. On Base, we observed a window that stayed open for 100 minutes. The infrastructure to capture these opportunities has not fully migrated to Layer 2. The bots that are watching Layer 2 are calibrated for mainnet-scale profits. They are not calibrated for Layer 2-scale profits.

Why This Matters for Regular Swaps

You do not need to be running an arbitrage bot to care about liquidity fragmentation. Every swap is affected. When you swap USDC for WETH on a single DEX, you are accepting whatever price that DEX’s pool offers. If Aerodrome has a deeper pool for that pair at that moment, you are leaving value on the table. If a two-hop route through an intermediate stablecoin pool produces a better effective price, you are leaving value on the table.

This is the argument for aggregation. An aggregator queries all venues. An aggregator evaluates all routes. Direct and multi-hop. An aggregator selects the path that maximizes output after all costs. The user does not need to know which venue has the deepest pool. The user does not need to know which intermediate token produces the best route. The aggregator handles it.

The 9 bps fee that TVMExecutor charges is the cost of this routing. For most swaps above $100, the routing improvement exceeds the 9 bps fee. The better effective price from finding the optimal venue and route more than offsets the fee. For very small swaps, the fee may exceed the routing improvement. But on Base, where gas is negligible, even small swaps benefit from avoiding MEV extraction through the private mempool.

The Market Will Close

The persistent discrepancies we have documented will not last forever. As more arbitrage infrastructure migrates to Base, the windows will close faster. As more liquidity consolidates into fewer, deeper pools, the fragmentation that creates discrepancies will diminish. This is how markets become efficient. Not through theory. Through the mechanical process of arbitrageurs capturing inefficiencies until the inefficiencies are too small to capture profitably.

We are documenting these observations now because they are interesting. Because they illustrate how Layer 2 markets work in their current, not-yet-efficient state. The physics-based framework we have built — the thermodynamic models, the negentropy metric, the transition path sampling — is our attempt to understand and predict these patterns systematically. Not just observing them after the fact. Predicting them before they occur. Those models are proprietary. The observations about market structure are not. They are visible to anyone who watches the pools.

The line observes. The line models. The line documents. The market will close. The documentation will remain.

Launch Orkid Swap →

Written by Jacob Cavazos

← Back to blog