Autonomous Arbitrage: Measure Route Costs, Fragmentation and Settlement
What Matters
An automated arbitrage route succeeds on settled proceeds after all distinct costs, not on a quoted price gap or a network speed ranking. DN's route ledger separates fees already embedded in quotes, external execution costs, failed attempts and cross-chain completion. It makes fragmentation measurable without inventing an agent performance benchmark or promising protection from every ordering risk.
DN Evidence Block
Reviewed: 12 November 2026. Evidence: current official routing, quotation and transaction-submission documentation. Instrument: DN Route Fragmentation and Settlement Ledger. Data status: the arithmetic example and downloadable record are illustrative. No live agent-wallet cohort, latency benchmark, strategy return series or network efficiency ranking was measured for this article.
The Price Gap Is Only the Start of the Trade
An agent may see different prices for an asset on two venues and still have no executable profit. The quoted size may differ. Inventory may sit on the wrong chain. One quote may include swap fees while another reports them separately. A profitable transaction may lose the competition for inclusion. A successful first leg may leave assets awaiting a second leg.
The practical unit of analysis is the complete route: a defined input asset and amount, a final asset and amount, the relevant times and every cost that is not already in those amounts. “Twenty DEXs” describes distribution, not a loss rate. A route with several venues may execute better than a single venue; splitting it can also add costs and failure paths. The answer requires matched observations.
Claim Versus What the Documentation Supports
| Claim | What can be checked | Remaining evidence |
|---|---|---|
| The best displayed price is the best route | Quoted input/output, size, fee inclusion and estimated execution cost | Actual settlement, external costs and failed attempts |
| A fast quote means a fast transaction | Quote response time and selected route | Submission, inclusion, finality and beneficiary availability times |
| Private submission eliminates MEV | The provider's stated submission and protection scope | Trust assumptions, fallback paths, inclusion and exceptional-block behaviour |
| A cross-chain route is atomic | The exact rollback and refund guarantees for each stage | Evidence that every leg shares the claimed failure boundary |
| Agents outperform human routing | A defined cohort and comparable observed costs | Attribution, matched tasks, failed runs and statistical uncertainty |
Uniswap's quote API documentation distinguishes route and quote information, gas estimates and simulation-related fields. Its routing explanation describes why combining pools and paths can improve execution. Neither establishes a permanent best chain or the realised return of a reader's agent.
The DN Route Fragmentation and Settlement Ledger
Record a route identifier and the chain, asset identifier, amount and time at each boundary. For a round trip, express initial and final amounts in the same asset. If the assets differ, preserve the conversion rate and timestamp used for comparison; a dollar label should not conceal an unpriced token exposure.
Keep three views. The quote view records expected amounts and inclusion assumptions. The execution view records what was submitted and confirmed. The reconciliation view records settled balances and unresolved items. A route that has not completed belongs in the unresolved group, not automatically in the profit or loss column.
| Ledger field | Purpose | Double-counting or scope trap |
|---|---|---|
| Input and quoted final output | Define the economic route | Using a mid-market display instead of an executable quote for the actual size |
| Costs embedded in output | Explain the quote's net amount | Subtracting swap fees or price impact a second time |
| External gas, tips and fees | Capture costs outside that quote | Mixing native-token cost with output-token units without a conversion convention |
| Failed-attempt cost | Allocate separately defined failed-run costs | Ignoring unsuccessful attempts or charging them twice |
| Working inventory and residual assets | Show capital not returned to the reference asset | Calling a valued residual token realised cash profit |
| Completion and finality state | Distinguish submitted, included, settled and available | Treating source-chain confirmation as completion on every destination |
A Worked Example Without Invented Gas Prices
Assume a hypothetical round trip starts with 10,000 units of one token. The route quotes 10,025 units back in the same token after its embedded pool fees and price impact. External execution gas, converted under the stated reference convention, costs 8 units; a separate tip costs 3. The operator assigns another 4 units of failed-attempt cost to this successful run under a disclosed allocation rule.
The quoted route surplus is 25 units. The fully allocated illustrative net is 10,025 − 10,000 − 8 − 3 − 4 = 10 units. Pool fees are not subtracted again because the quoted output already includes them. A separate borrow premium, if applicable, would still need inclusion unless it too was already counted. The flash-loan cost worksheet covers that distinct obligation.
If actual settled output is 10,018 with the same outside costs, the corresponding net becomes 3 units. If a second leg is unfinished, neither result describes a completed route. Preserve the residual inventory and mark final profit undetermined until the defined valuation or liquidation convention is applied. A scenario is useful precisely because its assumptions can be changed without pretending that a quote has already executed.
Submission Protection Has a Defined Boundary
Flashbots Protect documents a private submission service and specific protection features. A service's stated behaviour should be recorded for the route that uses it; it should not become a claim that every private RPC eliminates all MEV, all information leakage or all adverse price movement.
On Solana, Jito's low-latency transaction documentation describes bundle execution and also an important exception: transactions from uncled blocks can be rebroadcast outside the expected bundle protections. This is a concrete reason to review the whole submission and rebroadcast path. The provider's documentation is evidence of design and disclosed limitations, not DN's measured inclusion or loss rate.
Record who receives the transaction, what simulation was performed, which deadline applies, whether a fallback can send it elsewhere, and how outcomes are reconciled. An agent should not silently broaden an approved submission policy when its first path fails. A failed route can be an acceptable outcome if the alternative would exceed its permitted cost or authority.
Cross-Chain Fragmentation Adds Inventory and Completion Questions
One signed intent or one interface action does not necessarily mean one atomic state transition across all chains. A route may use solver inventory, asynchronous messages, refunds or a claim process. Review the actual terms and contracts rather than infer rollback behaviour from the number of clicks.
CoW's cross-chain interface documentation, for example, describes supported source and destination combinations and a bridge-provider dependency. That supports recording an interface route and its dependencies. It does not remove the need to check availability, amounts, destination support and completion for the specific order.
An inventory map should distinguish funds available for immediate use, funds committed to an open order, pending bridge claims, collateral pledged elsewhere and tokens that cannot currently be sold at the reference size. Aggregating all of them into “available capital” overstates what an agent can deploy. Correlated dependence on one solver, RPC or signing service should be recorded separately from the number of venues.
How to Measure Fragmentation Instead of Assuming It
Define a comparison task: same input asset, amount, permitted destination, deadline and execution constraints. Compare a candidate route against a stated alternative using contemporaneous executable quotes and then record outcomes. A simulated comparison remains simulated; a historical replay must explain which information would have been available at decision time.
Capture every eligible attempt, including rejected quotes, submission failures and incomplete routes. Report median and tail costs only after defining sample size, time window and missing-data treatment. A low mean cost with frequent unfinished transfers may be inappropriate for a time-critical strategy. No single efficiency score should hide that trade-off.
Agent attribution is another separate field. Automated transactions do not establish that an LLM or autonomous agent decided the trade. A keeper, fixed-rule bot, human using an API and agent-operated system can produce similar wallet traces. This article does not infer an agent population from transaction frequency or use such an inferred population to claim superior performance.
Permissions, Records and Commercial Resources
Give the execution system explicit spending and approval limits, permitted assets, route constraints and a stop condition for stale data. Separate proposing a route from approving an irreversible transaction where the deployment calls for that separation. Retain redacted evidence references rather than private keys, API tokens or full customer data in public records.
Existing reader resources include CoW Swap for inspecting available interface quotes and Koinly or CoinLedger for evaluating recordkeeping workflows. These retained commercial links may compensate DN. They are not evidence of best execution, current eligibility, strategy profitability or the suitability of a tax treatment. Check the actual product and jurisdiction before use.
Methodology and Limitations
The ledger is an original accounting and evidence framework, grounded in the distinctions exposed by official execution documentation. The example uses arbitrary round numbers in one reference asset; it is not a network fee benchmark. The downloadable fixture is fictional. It contains no wallet attribution, live quotes, credentials or automated data collection.
Refresh after API changes, submission-policy changes, route dependencies, fees or finality assumptions change. Review actual protocol and contract versions for a deployment. The source review date is 12 November 2026; no scheduler, live feed or endpoint is created by the downloadable record.
Frequently asked questions
Does using more DEXs always reduce profit?
No. Additional liquidity can improve the available route, while extra transactions and dependencies can add cost. Compare the complete route under matched constraints.
Should swap fees be subtracted from the quoted output again?
Only if the quote excludes them. Preserve the quote's fee convention; subtracting an already embedded cost twice understates the result.
Does private submission guarantee a profitable trade?
No. Review the provider's exact protection scope, inclusion conditions, fallback paths and exceptional cases. Profit also depends on the actual economic route.
Does the ledger measure agent performance?
No. It records the fields needed for a meaningful study. The article supplies no verified agent cohort or observed performance ranking.
Corrections and Changelog
12 November 2026: Replaced unsupported chain latency, gas and efficiency rankings, an unsubstantiated trade-size formula and blanket MEV-protection claims. The revised article preserves the autonomous execution topic and canonical URL while introducing explicit quote, cost and settlement boundaries.
Report a correction with the affected field, route or document version and a primary source. Existing internal related-reading links are retained separately. This is technical and financial decision support, not a promise of arbitrage profit or a recommendation to automate beyond the reader's tested authority and resources.
Machine-readable review record
The record below is a bounded illustrative example. It is not a wallet scan, current-chain dataset or assessment of a deployed service. The schema identifier is an identifier, not a promised API endpoint. Downloads are local snapshots; no schedule or automatic refresh is active.
Primary sources
- Uniswap quote API — Quote, route, gas estimate and simulation fields have defined scopes.
- Uniswap routing explanation — Multiple pools and routes can change execution; not a measured agent ranking.
- Flashbots Protect overview — Specific private submission features; not universal MEV elimination.
- Jito low-latency transaction documentation — Bundle and uncled-block/rebroadcast boundaries.
- CoW cross-chain interface documentation — Current supported-route and bridge-provider dependencies.
- CoW Swap reference interface — Current official interface evidence for retained commercial link; no performance endorsement.
Related reading:
When Puts & Calls Disconnect: The Quantitative Guide to Volatility Surface Arbitrage
Harvesting Market-Neutral Yields: The Crypto Funding Rate Arbitrage Playbook (2026)
The Quant Trader’s Guide to Grid Bots and Automated Arbitrage
The CEX-DEX Arbitrage Leakage: How Toxic Bots Drain Value From Passive AMM Liquidity
Quant Grid Bots 2026: Unlocking Cointegration Arbitrage on Perp DEXs