{"id":"def9b3aa-860d-45ea-81e6-5c283a74cf2d","arxiv_id":"2509.16157","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A transaction-level optimization model shows that JIT liquidity providers on Uniswap V3 could raise profits by up to 69% by accounting for price impact, but optimized JIT activity would cut passive LP fee income by up to 44% per trade.","lead":"This paper builds a per-transaction optimization model for just-in-time liquidity providers in Uniswap V3-style pools and finds that real JIT traders currently leave substantial profit on the table. It also estimates that smarter JIT liquidity would reduce trader slippage while eroding passive liquidity providers' fee income.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The empirical profit-gain headline depends on unmodeled inclusion-auction costs: Section 4.2.2 assumes the bid cost v is strategy-independent, while Section 4.1 describes a sealed-bid auction where the winning JIT bid should consume surplus.","rationale":"The reader's weakest assumption—that inclusion-auction costs are strategy-independent and unmodeled—is exactly the load-bearing concern. The paper's own text flags it twice, and the review rule says such self-acknowledged limitations must be weighed. The empirical headline is the part of the paper most likely to be acted upon by practitioners, and it is the part most vulnerable to this concern. The theorem and the transaction-level model have independent value: they provide a parameter-free derivation of the utility decomposition and a formal existence result, and the authors make code available for reproducibility. I also note internal inconsistencies flagged by the reader: Lemma 3.2's statement that q′ is 'strictly decreasing in L' contradicts its own proof sketch and the monotonicity claim; the abstract's 69% gain differs from the introduction's 41%; and Lemma 3.2 is stated with an internal contradiction in the main text. These reinforce the need for revision but are secondary to the auction-cost issue. Because the reader already assigned CONDITIONAL with this concern, my read does not change the verdict; the condition should be that the empirical claims be restated as upper bounds and the model's assumptions made explicit.","tokens_in":23563,"tokens_out":4243,"duration_ms":39608,"concrete_test":"Estimate the distribution of winning JIT bundle bids for the same USDC/WETH pool and January–June 2024 window using Flashbots or MEV-Share auction data, then re-run the Section 6.1.1 counterfactual with net utility U(s) − v, drawing v from that estimated distribution. If median net optimized profit is no longer significantly positive, or the 69% average gain collapses, the headline claims must be restated as gross upper bounds before inclusion costs.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claim—that real JIT LPs could have earned up to 69% more with Algorithm 1 and that the optimized strategy 'never loses money' (Section 6.1.1)—is computed under a cost model that deliberately omits the strategic cost of winning block inclusion. Section 4.2.2 states: 'we assume v to be a constant that does not relate to the choice of liquidity position.' Section 6 (Data Collection) explicitly says: 'we do not simulate different results of bidding auctions among JIT LPs.' This is not a minor technicality: Section 4.1 itself models inclusion as a sealed-bid auction among JIT LPs, where the winning bundle pays a tip bounded below by the second-highest bid. If bids are competitive and consume a meaningful share of the gross surplus F−C, then the 69% figure and the 'never loses money' statement are gross-of-auction upper bounds, not achievable net profits. The existence theorem (Theorem 4.1) is mathematical and survives this concern, but the paper's headline practical contribution—an optimal strategy that JIT LPs should adopt—does not. A secondary, related issue is that Algorithm 1's optimality guarantee is conditional on solving a non-concave univariate optimization to global optimality, but the implementation uses Particle Swarm Optimization and binary search, neither of which carries a global optimality certificate. This does not invalidate Theorem 4.1, but it means the empirical 'optimized' profits are only as good as the heuristic solver; the direction of this bias is unclear a priori.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies just-in-time (JIT) liquidity provision in concentrated liquidity market makers (CLMMs). It develops a transaction-level model in which a JIT LP chooses a position (L,a,b) immediately before a swap and earns fees minus price impact, subject to a budget constraint. The paper characterizes when price impact is favorable, analyzes how post-swap price depends on injected liquidity, proves existence of an optimal strategy (Theorem 4.1), and proposes Algorithm 1 to compute it. The empirical evaluation on Uniswap V3 USDC/WETH data (January-June 2024) claims that real JIT LPs are suboptimal, could earn up to 69% more with the optimized strategy, that JIT returns are driven mostly by price impact, and that optimized JIT activity reduces passive LP fees and trader slippage.","tokens_in":23951,"tokens_out":6504,"duration_ms":54540,"significance":"If the results hold, this would be a valuable first formal, transaction-level treatment of JIT LP optimization in CLMMs. The theoretical core is largely self-contained: the price-impact characterization and the existence theorem do not rely on fitted parameters, and the empirical study uses real on-chain data. The paper also articulates falsifiable predictions through the three strategic archetypes and the fee-sufficiency condition. However, the headline empirical claims are currently overstated because inclusion-auction costs are assumed away and the numerical optimizer lacks a global-optimality certificate; these issues do not invalidate the existence theorem but do affect the practical contribution.","major_comments":[{"comment":"Lemma 3.2 states that q' is 'continuous, strictly decreasing in L' with lim_{L→∞} q'=q, but with q'(L)<q for finite L, convergence to q from below requires q' to be increasing in L. The proof sketch in Section 3.2 even says 'As L increases, the same ∆x has less price impact, so q'↑,' and the appendix proof does not establish the claimed direction. This is not a cosmetic typo: the sign of the monotonicity is used implicitly in Lemma 3.1 and in the continuity/boundedness argument of Theorem 4.1. Please correct the statement and make the proof consistent.","section":"Section 3.2 / Appendix D (Lemma 3.2)"},{"comment":"The model assumes the auction/bid cost v is 'a constant that does not relate to the choice of liquidity position' (Section 4.2.2), while Section 4.1 describes inclusion as a sealed-bid auction in which the winning JIT bundle pays a tip bounded below by the second-highest bid. Section 6 explicitly does not simulate bidding auctions among JIT LPs. Consequently, the reported 'up to 69% more than current profit' and 'never loses money' (Section 6.1.1) are gross-of-auction upper bounds, not achievable net profits; under competitive bidding the surplus could be consumed by tips. The existence theorem is unaffected, but the practical headline should be either re-derived with an auction model or explicitly labeled as an upper bound.","section":"Section 4.2.2 and Section 6 (Data Collection)"},{"comment":"Algorithm 1's optimality guarantee is conditional on solving a non-concave, non-linear univariate problem (Eq. 19) to global optimality. The implementation uses Particle Swarm Optimization and binary search, neither of which provides a global optimality certificate. The empirical 'optimized' profits, including the 69% figure and the 'never loses money' statement in Section 6.1.1, are therefore only as strong as the heuristic solver. Please either provide a certified global method (e.g., branch-and-bound over the single variable) or report bounds on the suboptimality gap.","section":"Section 5 (Algorithm 1)"},{"comment":"The proof of Proposition 5.1 contains sign errors: it states 'F_m < C_m for all m ⇒ U>0' when U=F−C would give U<0, and it concludes that the derived inequality is 'a sufficient condition for U>0' although the proposition asserts U<0. The proposition itself appears correct after rearrangement, but the appendix proof needs to be fixed to match the statement.","section":"Appendix E (Proposition 5.1)"}],"minor_comments":[{"comment":"The proof divides by sqrt(q)−sqrt(q') and calls it positive, but for q<q' it is negative; the final inequality is correct only after a sign flip. Please correct the sign discussion.","section":"Appendix B, proof of Theorem 3.1"},{"comment":"The data-collection paragraph reports 6,829 JIT swaps, while Section 6.1.1 states 6,884 JIT transactions; please reconcile these numbers.","section":"Section 6"},{"comment":"The bid/inclusion cost is denoted v in Section 4.2.2, v* in Appendix D's proof of Theorem 4.1, and 'v*_j' in Table 1; please unify the notation.","section":"Section 4.2.2 and Table 1"},{"comment":"The axis labels appear garbled ('102 101 100 0 100 101 102 103'); please format the log-scale axis clearly.","section":"Figure 3(a)"}],"recommendation":"major_revision","confidential_remarks":"The core existence theorem appears sound and the paper is a good fit for the journal. The main risk is that the empirical headline (69%, 'never loses money') will be quoted without the auction-cost caveat. I would encourage the editor to request a revision that either models the inclusion auction or clearly labels the empirical results as upper bounds, and that tightens the consistency of Lemma 3.2 and the Appendix E proof."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague — here's my read. The real contribution is the transaction-level formulation: a JIT LP choosing (L,a,b) to maximize fees minus price impact for a single swap in a CLMM. The price-impact threshold (Theorem 3.1) is clean, the three archetypes are useful, and Theorem 4.1's existence argument over a finite set of tick intervals with continuous utility is basically right. They back it with real Uniswap V3 data from USDC/WETH, and the code is public. That is more than most theory papers in this space.\n\nThe soft spots are real but not fatal. First, Lemma 3.2 states q' is strictly decreasing in L; the proof and the limit result require q' to increase toward q as L grows. The direction is simply wrong as stated. It doesn't break Theorem 4.1, which only needs continuity, but it needs a corrected statement. Second, the empirical profit gains are inconsistent across the paper — 69% in the abstract and Section 6.1.1, 41% in the introduction. More substantively, the 69% and the 'never loses money' claim are computed with v (the bid for inclusion) treated as a constant independent of the liquidity position, and with no simulation of the sealed-bid auction they themselves describe in Section 4.1. So those numbers are gross of the strategic cost of winning the block. They are upper bounds, not achievable net profits. The theory survives; the practical adoption claim does not. Third, Algorithm 1's optimality is conditional on globally solving a non-concave univariate problem, but the experiments use PSO and binary search, neither with a certificate. So the 'optimized' profits are heuristic estimates, with unknown bias. Minor: Proposition 5.1's statement says U<0, but the proof's last line says U>0 — a sign typo.\n\nCitation pattern looks fine — they build on [11], [24], [25] and the JIT literature is appropriately credited.\n\nWho should read: anyone working on DeFi microstructure or MEV. The model and existence theorem deserve to be in the literature. The empirical section will need a careful rewrite before I'd trust the numbers. This paper deserves a serious referee — the right outcome is conditional acceptance with revisions that fix the lemma, reconcile the headline figures, and reframe the empirical results as upper bounds.","headline":"Solid first model of JIT LP optimization in CLMMs, with an existence theorem worth taking seriously, but the empirical headline numbers are upper bounds that ignore inclusion auctions and the paper has one genuinely sloppy lemma.","tokens_in":24421,"tokens_out":3575,"would_cite":true,"duration_ms":30573,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A transaction-level model of just-in-time liquidity provision in concentrated-liquidity AMMs proves an optimal single-swap strategy exists, and fitting it to real Uniswap V3 data shows JIT LPs could earn up to 69% more by accounting for…","keywords":["just-in-time liquidity","concentrated liquidity market makers","Uniswap V3","automated market makers","price impact","impermanent loss","MEV","liquidity provision optimization"],"falsifier":"Obtain the complete set of bids from the block-builder auctions for the same six months of USDC/WETH JIT bundles, and recompute the profit gap as $U=F-C-v(\\text{bid})$ with each JIT LP charged its actual inclusion cost; if the gap between observed and optimized profit shrinks toward zero as $v$ rises with position profitability, the 69% estimate is an artefact of the constant-$v$ assumption rather than a real inefficiency.","tokens_in":23314,"feed_emoji":"💧","tokens_out":9019,"duration_ms":76329,"temperature":0.7,"pith_summary":"This paper builds the first transaction-level theory of just-in-time (JIT) liquidity provision in concentrated-liquidity AMMs such as Uniswap V3. A JIT LP watches a pending swap, mints a concentrated position just before it, and burns it immediately after, earning fees while also gaining or losing from the price move the swap causes. The paper proves that for every swap there is a global profit-maximizing choice of liquidity amount and price range, and gives an algorithm that computes it. Applied to six months of real Uniswap V3 JIT transactions, the optimized strategy would have increased profits by up to 69% relative to what the JIT LPs actually earned, because most real JIT LPs ignore price impact and even take trades with negative net utility. The authors also find that if JIT LPs optimized, traders would face lower slippage while passive LPs' fee income would drop by up to 44% per trade.","feed_headline":"Just-in-time liquidity providers leave 69% profit on table","feed_subtitle":"A new model finds the optimal single-swap position and shows real JIT LPs ignore price impact.","key_machinery":"The load-bearing object is the tick-decomposed utility identity $U=\\sum_m (F_m-C_m)$, where $F_m=\\delta_m L/(L+P_m)$ is the JIT LP's pro-rata share of the fee accrued in tick $m$, and $C_m$ is the corresponding price-impact term derived as a function of the JIT liquidity $L$, the passive liquidity $P_m$, the swap size, and the market and tick prices. The key structural fact is the price-impact threshold of Theorem 3.1: for a move $q\\to q'$, with $\\hat q$ the projection of the current price onto the chosen range, $C\\le 0$ holds exactly when $\\hat q' \\le (p_x/p_y)^2/\\hat q$ for a price decrease, and the reverse inequality for a price increase. This turns the JIT decision into a finite search over tick ranges that are actually touched as the price travels from $q$ to the no-JIT post-swap price $q^*$, plus a one-dimensional optimization in $L$; Algorithm 1 performs that search, and Theorem 4.1 guarantees the search has a finite global winner.","core_discovery":"On the paper's model, a JIT LP's per-swap utility is $U(s;\\theta)=F(s;\\theta)-C(s;\\theta)$, where $F$ is the trading fee it captures from the swap and $C$ is price impact (the change in dollar value of its position between mint and burn, defined as positive when the position loses value). Theorem 4.1 states that this utility always attains a global maximum over the strategy space of single positions $(L,a,b)$ subject to the budget; the argument is that tick ranges are finite, and on each range $F$ is continuous and bounded while $C$ is continuous in $L$, so the feasible interval for $L$ is compact. Algorithm 1 finds the optimum by enumerating ranges in $\\mathcal{R}(q,q^*)$ and solving a one-dimensional non-concave problem for $L^*$. The paper classifies the resulting optima into three strategic archetypes: overpriced trades (pool price moves away from market price, price impact helps the LP), arbitrageur trades (price moves toward market, price impact hurts), and overshoot trades (price crosses the market price, so the LP should concentrate liquidity only in the segment where $C<0$). Fitting the algorithm to 6,829 real JIT swaps in a USDC/WETH Uniswap V3 pool, the authors find that observed JIT behavior underperforms the optimum by up to 69%, that price impact accounts for roughly 75% of total returns on average, and that optimized JIT participation could cut passive-LP fee income by up to 44% per trade while lowering trader slippage.","pith_inferences":["The 69% figure likely overstates what JIT LPs could retain if inclusion is a competitive auction: the paper fixes the winning-bid cost $v$ independent of the position, so part of the measured gap may be a transfer to block builders rather than profit.","The threshold condition in Theorem 3.1 could be turned into a real-time classifier: label a pending swap by whether $q'$ crosses the boundary, and only dispatch capital when the label is 'overpriced' or the post-crossing segment of an overshoot trade.","Because the strategy space is restricted to single positions, a natural extension is to allow two or more simultaneous positions; the proof that $U$ attains a maximum would need revisiting, but the tick-decomposed utility carries over.","The empirical study uses one pool, and the authors say volatile pools show wider gaps, so the same counterfactual should be re-run on a high-volatility pool to see whether the 69% gap widens."],"forward_implications":["Overpriced trades, which push the pool price away from the market price, are unambiguously favorable: price impact is non-positive and the optimal strategy can deploy the full JIT budget.","Arbitrageur trades, which push the pool price toward the market price, are usually loss-making unless fees cover the price impact, so the optimal strategy is often to skip the trade.","If JIT LPs optimized their positions, passive LPs' fee income would fall by up to 44% per trade, while traders would enjoy lower slippage.","JIT liquidity acts as a preemptive arbitrage force: it dampens price dislocations and reduces the profit available to arbitrageurs who would correct the same mispricing later.","The three-way classification of trades (overpriced, arbitrageur, overshoot) gives a practical participation rule derived from the price-impact threshold, not just a numerical recipe."],"supporting_citations":[{"why":"Supplies the Uniswap V3 CLMM mechanics, tick structure, reserve formulas, and swap algorithm on which the fee and price-impact decompositions are built.","marker":"[1]"},{"why":"Provides the on-chain USDC/WETH pool dataset from which the 6,829 JIT transactions and the empirical counterfactual are drawn.","marker":"[5]"},{"why":"Prior market-level model of JIT liquidity and adverse selection that motivates the need for a transaction-level optimization model.","marker":"[11]"},{"why":"Describes the bundle-auction and block-builder inclusion mechanism that underlies the model's treatment of JIT competition and the constant-cost assumption.","marker":"[17]"},{"why":"Defines loss-versus-rebalancing, the long-run analogue of the price-impact measure the paper uses at transaction level.","marker":"[19]"},{"why":"Provides the game-theoretic LP model and the observation that JIT transactions in recent data use a single position, justifying the strategy-space restriction.","marker":"[24]"},{"why":"Empirically characterizes JIT LP activity on Uniswap and supplies the mint-burn sandwich identification criterion used to label JIT swaps.","marker":"[25]"},{"why":"Describes just-in-time liquidity attacks and sandwich bundles on Uniswap V3, grounding the mint-swap-burn modeling and the strategic archetypes.","marker":"[27]"}],"fun_headline_variants":["JIT LPs leave 69% on table by ignoring price impact","Optimal JIT strategy could boost LP profits by 69% and cut passive fees by 44%","Model shows JIT LPs underperform by 69%; optimal moves slash passive LP fees by 44%","JIT LP model: 69% profit gain but 44% cut to passive LP earnings","Strategic JIT LP model reveals 69% underperformance, 44% passive LP loss"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central empirical claim assumes the JIT LP's cost of winning the right to include its sandwich bundle is the same no matter which liquidity position it chooses; if that cost is actually a competitive bid that rises with the profit the position can extract, the measured 69% gain would be eaten by the auction.","fun_headline_variants_meta":{"raw":{"variants":["JIT LPs leave 69% on table by ignoring price impact","Optimal JIT strategy could boost LP profits by 69% and cut passive fees by 44%","Model shows JIT LPs underperform by 69%; optimal moves slash passive LP fees by 44%","JIT LP model: 69% profit gain but 44% cut to passive LP earnings","Strategic JIT LP model reveals 69% underperformance, 44% passive LP loss"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001138,"raw_usage":{"total_tokens":4840,"prompt_tokens":1177,"completion_tokens":3663,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":793,"completion_tokens_details":{"reasoning_tokens":3543}},"tokens_in":793,"tokens_out":3663,"duration_ms":20149,"temperature":1.0,"reasoning_tokens":3543,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:49:32.866130+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Obtain the complete set of bids from the block-builder auctions for the same six months of USDC/WETH JIT bundles, and recompute the profit gap as $U=F-C-v(\\text{bid})$ with each JIT LP charged its actual inclusion cost; if the gap between observed and optimized profit shrinks toward zero as $v$ rises with position profitability, the 69% estimate is an artefact of the constant-$v$ assumption rather than a real inefficiency.","supporting_citations":[{"cited_title":"Uniswap v3 core.https://uniswap.org/whitepaper-v3.pdf,","cited_arxiv_id":null,"evidence_quote":"Supplies the Uniswap V3 CLMM mechanics, tick structure, reserve formulas, and swap algorithm on which the fee and price-impact decompositions are built."},{"cited_title":"Allium - enterprise blockchain data platform.https://www","cited_arxiv_id":null,"evidence_quote":"Provides the on-chain USDC/WETH pool dataset from which the 6,829 JIT transactions and the empirical counterfactual are drawn."},{"cited_title":"Game Theoretic Liquidity Provisioning in Concentrated Liquidity Market Makers.Proceedings of the ACM on Measurement and Analysis of Computing Systems, 9(1):1–45, 2025","cited_arxiv_id":null,"evidence_quote":"Provides the game-theoretic LP model and the observation that JIT transactions in recent data use a single position, justifying the strategy-space restriction."},{"cited_title":"Just-in-Time Liquidity on the Uniswap Pro- tocol.SSRN 4382303 Electron","cited_arxiv_id":null,"evidence_quote":"Empirically characterizes JIT LP activity on Uniswap and supplies the mint-burn sandwich identification criterion used to label JIT swaps."},{"cited_title":"De- mystifying just-in-time (JIT) liquidity attacks on uniswap v3","cited_arxiv_id":null,"evidence_quote":"Describes just-in-time liquidity attacks and sandwich bundles on Uniswap V3, grounding the mint-swap-burn modeling and the strategic archetypes."}],"review_version":2}