{"id":"efe4e58b-b967-473d-b308-ccce03a2be91","arxiv_id":"2506.14768","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Optimistic MEV, mostly non-trading cyclic arbitrage probes, accounts for 51% of on-chain gas on Base and 55% on Optimism in Q1 2025, versus 7% on Arbitrum.","lead":"This paper defines and measures a new category of Ethereum Layer 2 behavior called optimistic MEV, finding that speculative cyclic arbitrage probes consume more than half of all gas on Base and Optimism in Q1 2025 while paying less than a quarter of transaction fees. It matters because it quantifies how much blockspace on major rollups is being taken by low-value automated spam rather than real user activity.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The >50% gas-share claims treat all successful no-trade DEX-touching transactions from cyclic-arbitrage contracts as optimistic probes; without per-transaction trace evidence this label is not established.","rationale":"The paper does what it claims: it operationalizes optimistic MEV, releases the contract list and pipeline, validates contracts through multiple heuristics, and cross-checks Arbitrum behavior with an external dataset. The relative patterns (Base/Optimism high, Arbitrum low, L1 negligible) are robust and likely correct. However, the numeric headline (51%/55% of gas) is sensitive to the interpretation of interaction-success transactions. The reader's weakest assumption identifies exactly this sensitivity; my trace-based test would settle it. Since the reader already issued CONDITIONAL, no verdict change is needed. If the test shows most interaction-success transactions are read-only probes, the paper's central claim stands and the current condition is satisfied; if not, the paper would need to qualify the gas-share statement.","tokens_in":27799,"tokens_out":3569,"duration_ms":35636,"concrete_test":"Sample ~1,000 cyclicArb-interaction-success transactions per network from Q1 2025 on Base and Optimism. Replay full traces and annotate each as (a) read-only probe: every subcall to a DEX pool is STATICCALL/slot0/getReserves and no state-changing subcall occurs besides the fee transfer; (b) state-changing non-trade: writes to storage or calls to non-DEX contracts; (c) precomputed pattern: calldata identical over a contiguous run with no on-chain branching on pool state. Also check whether the bot contract has admin/storage-write transactions that set path/amount parameters off-chain. Recompute Q1 gas share using only category (a). If the adjusted share falls below 50% on either network, the headline claim overstates optimistic MEV; if it stays above 50%, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central result is the Q1 2025 gas share of cyclicArb-interaction-success transactions on Base (48%) and Optimism (52%). Section 4.2 assigns 'interaction' purely from (i) first callee in Cbot and (ii) no Swap event while the trace touches a DEX pool. That is a contract-level or pool-contact proxy, not evidence that the transaction's execution path was an on-chain opportunity search. Appendix E validates contracts (majority of no-trade transactions touch DEX contracts, callers, swap-count percentiles), and Section 5.5's calldata-reuse and short-calldata evidence is aggregate; it is also consistent with a bot that resubmits a fixed calldata and lets the contract terminate when a precomputed path is no longer profitable, or with parameters loaded from storage written by earlier admin transactions. Because interaction-success dominates the gas share on Base and Optimism, this is not a small boundary case: the optimistic-MEV interpretation of the headline numbers rests entirely on this classification. I agree with the reader that this is the load-bearing assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper defines \"optimistic MEV\" as MEV in which opportunity discovery and execution parameters are determined on-chain at execution time, and measures its prevalence among cyclic arbitrage on Arbitrum, Base, Optimism, and Ethereum L1 using Dune data from August 2023 to May 2025. A multi-stage pipeline detects profitable cyclic swaps, validates the involved contracts (Cbot) through manual and heuristic audits (Appendix E), and labels each transaction by purpose, DEX involvement, and execution outcome. The headline results are that in Q1 2025 cyclic-arbitrage contracts account for 51% of gas on Base, 55% on Optimism, and 7% on Arbitrum, with the Base and Optimism shares dominated by successful no-trade \"interaction\" transactions, which the paper interprets as on-chain opportunity probes. The paper further compares top bot contracts across chains, reports fee and revert dynamics, and uses OLS regressions to associate cyclic-arbitrage activity with ETH volatility, retail trading, and DEX aggregator usage.","tokens_in":27964,"tokens_out":8849,"duration_ms":81752,"significance":"If the interpretation holds, this is the first systematic measurement of a distinct L2 MEV strategy and a valuable contribution to the MEV and rollup-design literature. The paper ships an open implementation and classification artifact, validates contracts through a multi-stage audit, and produces internally consistent cross-network contrasts—trade vs. interaction shares, revert behavior, calldata reuse, and fee levels—that are informative for sequencer and fee-market design. The finding that a small set of arbitrage contracts can consume the majority of blockspace at very low fees is policy-relevant. The main caveat is that the optimistic label is an interpretation of an aggregate proxy rather than a per-transaction verified property; the significance of the paper therefore depends on the trace-level evidence requested below.","major_comments":[{"comment":"The headline gas-share figures—51% on Base and 55% on Optimism in Q1 2025, with 48% and 52% respectively attributed to cyclicArb-interaction-success—depend on classifying every successful, no-swap transaction that invokes a validated cyclic-arbitrage contract and touches a DEX pool as an on-chain opportunity probe. The assignment rule uses only the first callee and pool contact; it does not verify that the execution path actually performed an opportunity search, such as reading multiple distinct pool states before terminating. The validation in Appendix E is contract-level (majority of no-trade transactions touch DEX contracts, caller diversity, swap percentiles) and is compatible with a transaction being a failed off-chain-planned attempt, a fixed-calldata resubmission, or a maintenance call that happens to read a pool. Because the interaction-success category is the dominant component of the gas share, the central claim that cyclic arbitrage on L2s is \"predominantly executed as optimistic MEV\" rests entirely on this assumption. Please add per-transaction trace evidence, for example the distribution of the number and diversity of pool reads in interaction-success transactions or path variation under identical calldata, or weaken the claim to what the classification supports.","section":"§4.2, §5.2, Appendix E"},{"comment":"The calldata-reuse and short-calldata statistics are presented as the main evidence that \"key parameters, i.e., the path of the arbitrage transaction and the swap amounts, are determined during execution rather than precomputed off-chain\" (Section 3). However, high transactions-per-unique-calldata ratios (e.g., 7,979,548 on Optimism for 0xabf4...017c) and median calldata lengths of 2–6 hex characters are also consistent with a bot resubmitting a fixed trigger while the contract reads parameters from storage updated by earlier transactions, or with repeated attempts at a precomputed path that is no longer profitable. The defining property in Definition 1 is that path and amount decisions are made within the transaction itself; that property should be demonstrated on traces, for instance by showing that identical calldata yields different multi-pool paths across executions, or by reporting what fraction of interaction-success transactions issue multiple distinct DEX staticcalls before terminating.","section":"§5.5, Table 1"},{"comment":"The validation procedure reports a gas-coverage stopping rule (more than 80% of gas from candidate-contract transactions originates from validated contracts) but does not report the number of candidates inspected and removed at each step, nor any precision/recall estimate against a labeled ground truth. Since Cbot is the denominator for all gas-share percentages and for the regression dependent variables, the absence of quantified accuracy leaves the central numerical claims without an error bound. Please report candidate counts and removal counts per validation step and, where possible, compare the final Cbot on each chain with an independent source such as EigenPhi [33] or Entropy Advisors [34,35] rather than using those sources only as a consistency check for revert behavior.","section":"Appendix E"}],"minor_comments":[{"comment":"In item 3 of the cyclic arbitrage detection description, \"the the token bought\" should read \"the token bought\".","section":"§4.1"},{"comment":"The phrase \"out analysis reveals\" should read \"our analysis reveals\".","section":"§1, contribution iii"},{"comment":"The sentence \"This classification stage aimes to identify\" contains a typo; \"aimes\" should be \"aims\".","section":"Appendix D"},{"comment":"The validation text says \"a cyclic arbitrage requires at least two transactions,\" but the surrounding discussion is about swaps per transaction; this should read \"at least two swaps.\"","section":"Appendix E, Swaps per Transaction"},{"comment":"The denominator of the \"cum. MEV bot gas (%)\" column is described as \"all identified MEV bots within the dataset C′bots,chain,\" but C′bots,chain is not defined in the main text; please clarify its relation to Cbot.","section":"§5.5, Table 1"},{"comment":"For Base and Optimism, Model 1 has adjusted R² values of 0.0736 and 0.0240, indicating very limited explanatory power for the overall cyclicArb transaction count; the text should avoid presenting these models as establishing drivers of overall cyclicArb activity and should emphasize the trade-count models instead.","section":"§6, Table 2"}],"recommendation":"major_revision","confidential_remarks":"To the editor: this is a competent and potentially influential measurement paper, but the title and abstract assert a behavioral interpretation ('optimistic MEV') that the pipeline supports only indirectly. I recommend requiring the trace-level validation described in Major Comment 1 before acceptance, since a few top contracts' interaction-success volumes dominate the headline numbers. If the authors cannot provide that evidence, they should reframe the contribution as measuring no-trade interaction volume from validated arbitrage contracts rather than as proving optimistic execution. The open-source artifact [63] is a real strength and should be preserved in revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, here's my read on arXiv:2506.14768.\n\nThe paper does two genuinely new things: it formally defines optimistic MEV and it measures it at scale on the three big L2s. The headline finding—that cyclic arbitrage contracts account for over half of gas on Base and Optimism in Q1 2025, mostly via no-swap \"interaction\" transactions—is striking and, if right, important for fee market and sequencer design. The authors also do real work on the pipeline: the cyclic arb detector, the contract validation in Appendix E, the bytecode similarity analysis, and the cross-network contrasts all show care. The open repo and Dune queries are a plus, and the regression section is honestly labeled exploratory.\n\nThe soft spot is the load-bearing classification assumption. Section 4.2 labels any successful transaction from a known cyclic arb contract that touches a DEX pool without emitting a swap as \"interaction,\" and the paper then treats that as evidence of on-chain opportunity probing. That is a reasonable inference, but it is not per-transaction verification. The aggregate calldata reuse and short calldata evidence is consistent with probing, but it is also consistent with bots resubmitting fixed calldata and letting the contract terminate when a precomputed path is no longer profitable, or with parameters loaded from storage. The stress test is right that this is not a minor boundary case: on Base and Optimism, the interaction-success category dominates the gas share, so the central claim rests almost entirely on this label. The paper gives no precision/recall for the classification and no trace-level validation of a random sample.\n\nI would not call this fatal. The definition is useful, the measurement is the first of its kind, and the cross-network differences (Arbitrum's lower share, the revert patterns, the fee dynamics) hang together and support the qualitative story. But the specific >50% optimistic MEV numbers should be read as upper-bound-ish until the authors quantify the classifier's error or validate a sample manually. I'd recommend they release per-transaction labels and do exactly that.\n\nThis paper deserves a serious referee. It is a strong measurement contribution with a central interpretation that needs tightening, not a desk reject. I'd send it out and ask the referees to focus on the classification validity.\n\nFor me: I'd bring it to reading group, and I'd probably cite it for the definition and the measurement, though with a caveat on the headline numbers.","headline":"First large-scale measurement of optimistic MEV on L2s is valuable and mostly convincing, but its central gas-share claims depend on an unvalidated classification of no-trade DEX interactions as on-chain probing.","tokens_in":28544,"tokens_out":1853,"would_cite":true,"duration_ms":16824,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Cyclic arbitrage on Ethereum Layer 2s has become speculative on-chain probing that dominates blockspace.","keywords":["blockchain","MEV","optimistic MEV","cyclic arbitrage","Layer 2","Ethereum","DEX","rollups"],"falsifier":"Pick a random sample of cyclic-arbitrage interaction transactions on Base and Optimism, replay their traces, and count how many use verbose calldata that encodes a specific path or amount or appear in bursts immediately after a single user trade; if a substantial share of so-called probes look like off-chain-planned back-run attempts that simply failed, the optimistic-MEV interpretation of the numbers would be falsified.","tokens_in":27531,"feed_emoji":"⛽","tokens_out":8781,"duration_ms":74116,"temperature":0.7,"pith_summary":"This paper defines and quantifies a class of blockchain value extraction it calls optimistic MEV, in which a bot decides inside the transaction itself whether an arbitrage opportunity exists, rather than checking off-chain before submitting. Its central finding is that on the rollups Base and Optimism, transactions from cyclic-arbitrage bots accounted for about 51% and 55% of all on-chain gas in Q1 2025, and that most of those transactions were interaction probes that touch DEX pools but never execute a swap. The paper reads this as evidence that cheap Layer 2 blockspace is being consumed by speculative machine activity, which is why blocks on those networks stay persistently full even when organic user demand is low. It further shows that the same behavior is far weaker on Arbitrum, and argues that design choices such as fee floors, transaction ordering, and block interval explain the difference. If true, the result changes how one thinks about Layer 2 capacity, fee markets, and MEV mitigation.","feed_headline":"Arbitrage probes eat over half the gas on Base and Optimism","feed_subtitle":"A Q1 2025 measurement finds arbitrage bots mostly probing pools without trading, so cheap blockspace is dominated by speculation, not users.","key_machinery":"The central object is optimistic MEV, defined as MEV whose opportunity existence and execution parameters are resolved by on-chain computation at execution time rather than verified off-chain before submission; by construction such strategies produce many speculative transactions that terminate without trading. The argument is carried by a three-stage classification pipeline: a cyclic-arbitrage detector that reconstructs token paths from swap events, drops router and aggregator calls, and keeps only cyclic, non-negative-profit transactions to seed a validated set of bot contracts; a per-transaction classifier that marks the first callee as cyclic-arb or other, the DEX activity as trade, interaction, or residual, and the outcome as success or revert; and an inference layer that uses calldata-reuse ratios, median calldata lengths, bytecode similarity, and regression to show that parameters are not precomputed off-chain.","core_discovery":"On its own terms, the paper claims that cyclic arbitrage on Layer 2 rollups has flipped from a precomputed, off-chain-verified strategy into an optimistic one: bots submit high-frequency transactions that begin with read-only calls into DEX pools, decide inside the Ethereum Virtual Machine whether a profitable cycle exists, and usually abort without any swap. The supporting measurements are that on Base and Optimism only 6.3% and 12% of cyclic-arbitrage DEX-touching transactions produce a trade, that the vast majority of these transactions terminate without a swap and without reverting, that top bots reuse minimal calldata tens to millions of times so paths and amounts cannot be precomputed off-chain, and that this activity is what fills post-Dencun blocks after Base and Optimism raised their gas targets. The paper contrasts this with Ethereum Layer 1, where 95.8% of cyclic-arbitrage DEX-touching transactions execute trades, and with Arbitrum, where a 0.01 GWei fee floor, first-come-first-served ordering, and roughly 250ms blocks coincide with a 7% gas share and a 52.6% trade share.","pith_inferences":["A testable extension would apply the same classification to other low-fee rollups such as zkSync, Scroll, or Unichain; the paper's mechanism predicts the highest probe shares on chains that combine very low fees with multi-second block intervals.","The calldata-reuse statistics hint at a concentration the aggregate figures understate: 50 identical bytecode clones on Base suggest that a handful of operators, not a competitive crowd, may be generating a large fraction of the activity, which would also change how the regression drivers should be read.","A way to stress-test the optimistic mechanism is to raise fees on a test network while holding block time fixed; if probe volume stays high, the cheap-fee explanation would be incomplete and the probes would be carrying some other value, such as information about pending pool states.","The results imply that any rollup introducing an express lane or priority ordering should expect the speculative probe pattern to adapt to that ordering rather than disappear, since deterministic inclusion lowers the risk that a probe arrives after the opportunity has faded."],"forward_implications":["A network that cuts costs and raises its gas target, as Base and Optimism did after the Dencun upgrade, hands most of the new capacity to speculative probes rather than to user traffic.","Fee floors or minimum transaction costs become the most direct policy lever: Arbitrum's 0.01 GWei floor and short blocks coincide with a 7% gas share, versus over half on the OP-Stack chains.","MEV auctions and revert protection do not transplant cleanly to Layer 2s, because high-frequency probing could flood an auction with noise and short block intervals leave little scheduling room.","Capacity planning and gas-target setting for rollups must separate this speculative probe baseline from genuine user demand, or risk inflating the apparent organic load.","Searcher economics change: bots compete by the cost of repeated probes rather than by priority-fee bidding, which is why these transactions pay less than a quarter of fees despite consuming most of the gas."],"supporting_citations":[{"why":"Documents front-running and priority-gas-auction MEV on Ethereum; the baseline against which optimistic MEV is contrasted.","marker":"[26]"},{"why":"Quantifies MEV on Ethereum L1, including cyclic arbitrage, supplying the high-success-rate Layer 1 baseline.","marker":"[67]"},{"why":"Provides the cyclic-arbitrage detection heuristic that the paper's pipeline adapts to Layer 2.","marker":"[80]"},{"why":"Earlier comparative measurement of MEV across Ethereum and major rollups, which this paper extends to the optimistic variant.","marker":"[75]"},{"why":"Theoretical result that arbitrage profits scale with the square root of mean inter-block time, used to compare Arbitrum with the OP-Stack chains.","marker":"[54]"},{"why":"Extends the inter-block-time profit result to deterministic block intervals, matching the regular 2-second blocks of Base and Optimism.","marker":"[58]"},{"why":"Supplies the intraday volatility estimator used as a regressor in the OLS analysis.","marker":"[42]"},{"why":"Supplies the bytecode similarity method used to detect cloned and shared arbitrage-bot contracts.","marker":"[49]"}],"fun_headline_variants":["L2 arbitrage bots waste half the gas on empty probes","Optimistic MEV: bots fill blocks without trading","Base and Optimism blocks full of nontrading arbitrage probes","Arbitrage probes dominate L2 gas, rarely trade","L2 arbitrage bots: pay little, hog gas, rarely trade"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The headline gas-share figures assume that every no-trade transaction sent to a validated cyclic-arbitrage contract is an on-chain probe, even though that label is assigned by contract address and DEX-pool touch rather than verified per transaction.","fun_headline_variants_meta":{"raw":{"variants":["L2 arbitrage bots waste half the gas on empty probes","Optimistic MEV: bots fill blocks without trading","Base and Optimism blocks full of nontrading arbitrage probes","Arbitrage probes dominate L2 gas, rarely trade","L2 arbitrage bots: pay little, hog gas, rarely trade"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001231,"raw_usage":{"total_tokens":5150,"prompt_tokens":1134,"completion_tokens":4016,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":750,"completion_tokens_details":{"reasoning_tokens":3931}},"tokens_in":750,"tokens_out":4016,"duration_ms":28442,"temperature":1.0,"reasoning_tokens":3931,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T00:09:51.818564+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Pick a random sample of cyclic-arbitrage interaction transactions on Base and Optimism, replay their traces, and count how many use verbose calldata that encodes a specific path or amount or appear in bursts immediately after a single user trade; if a substantial share of so-called probes look like off-chain-planned back-run attempts that simply failed, the optimistic-MEV interpretation of the numbers would be falsified.","supporting_citations":[],"review_version":1}