{"id":"dd89e2e4-f461-4ed6-b146-c8d969629266","arxiv_id":"2509.02941","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"High-frequency Uniswap v3 crypto data show non-quadratic potentials, with double-well shapes at short sampling frequencies that the author links to market stress.","lead":"This paper estimates the drift and volatility of crypto prices on a Uniswap decentralized exchange and derives the shape of an underlying potential function. It reports that the potential is non-quadratic and sometimes double-well, which the author interprets as evidence for non-linear market dynamics and meta-stable states.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Double-well potentials are reported without error bars or null-model tests; the apparent multi-stability may be an artifact of regime pooling or estimation noise, undermining the central meta-stability claim.","rationale":"The reader's CONDITIONAL verdict is appropriate. I identify a different weak point than the reader's weakest_assumption but one that is equally load-bearing: the lack of uncertainty quantification and null-model comparison for the double-well topology. The paper's core contribution is the empirical observation of non-quadratic potentials, especially double-well shapes, but this observation is presented as visual inspection of drift/potential plots without error bars or statistical tests. This is a serious omission given the small effective sample sizes after subsampling and the well-known sensitivity of nonparametric drift estimators to bandwidth and boundary effects. My proposed test directly addresses whether the reported double-well is real. The verdict should remain CONDITIONAL because the concern is addressable with additional analysis, not fatal: if the null-model test shows the double-well persists only in real data, the claim is strengthened; if not, the central claim is an artifact. I do not agree fully with the reader's weakest assumption: the time-step conversion (Eq. 4) is a technical concern but less likely to change the qualitative topology, whereas regime mixing/estimation noise can directly create the double-well. Thus partial agreement.","tokens_in":7425,"tokens_out":7270,"duration_ms":96802,"concrete_test":"Simulate three null/alternative processes calibrated to the observed USDC-WETH series: (a) a linear Ornstein-Uhlenbeck process, (b) a single-well nonlinear diffusion (e.g., cubic drift with one stable root), and (c) a two-regime OU process (two slowly alternating linear regimes). For each, generate the same number of observations with the same irregular 100s spacing, apply the identical trimming, subsampling at 10 min–3 h, and kramersmoyal pipeline, and compare the distribution of estimated potential topologies (number of wells) against the paper's results. If (a) or (c) produce double wells with frequency comparable to the paper's, the empirical double-well evidence is an artifact. Additionally, compute bootstrap confidence bands for the drift at 10-min lag and test whether the number of root crossings is significantly different from zero or one.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim hinges on the multi-stability (double-well) of the estimated potential, seen in Figs. 1 and 4 for two-month windows at short lags. But the paper provides no uncertainty quantification, no hypothesis test, and no comparison against a null model. The drift is estimated by pooling all increments over an entire window (Sect. 2.2, Eq. 4), so if the market switches between two regimes (e.g., high and low ETH/USD levels) within a window, the conditional average E[Δx|x] can exhibit multiple roots even when the instantaneous drift in each regime is linear. This is a classic aggregation artifact, not necessarily a genuine bistable potential. The author's dismissal of the one-month single-well result as 'data insufficiency' (Sect. 3.1) is unsupported; an equally plausible explanation is that short windows better approximate a single regime, while long windows mix regimes and create spurious double wells. Additionally, finite-lag bias in the Kramers-Moyal estimator at τ=10 min (vs. τ=3 h) can produce artificial non-linearities; the observed weakening of non-linearity with lag is exactly what would occur for any nonlinear diffusion and is not diagnostic of scale-dependent dynamics. Without a null model or bootstrap confidence intervals, the paper's headline finding is not statistically secured.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes high-frequency Uniswap v3 log-price data (USDC-WETH and WBTC-WETH on Arbitrum) using a non-parametric Kramers–Moyal expansion. From estimated conditional drift and volatility, it derives effective potentials U(x) for lags from 10 minutes to 6 hours and across two-month and one-month windows. The central claim is that the drift is strongly non-linear and the potential is non-quadratic, sometimes double-well, indicating metastable market states. The paper interprets double-well potentials as evidence supporting the author's earlier theoretical framework of non-linear market dynamics and instanton transitions.","tokens_in":7730,"tokens_out":3324,"duration_ms":39291,"significance":"If substantiated, the finding would provide empirical support for non-linear drift and metastable potentials in intraday crypto markets, which would be of interest for AMM microstructure, market-stress detection, and model specification. The non-parametric approach and the use of granular decentralized-exchange data are appropriate and commendable. However, the paper currently lacks the statistical machinery needed to distinguish genuine double-well dynamics from estimation artifacts, regime pooling, or finite-lag bias. The claimed evidence is therefore not yet secured, although the questions addressed are significant.","major_comments":[{"comment":"The conversion K1(x)/Δt and K2(x)/(2Δt) uses a single average Δt for data that are irregularly spaced and averaged over 15 transactions, with typical spacing of 100 seconds. For non-equidistant observations, this estimator is not justified and has unknown bias. Since the potential U(x) is obtained by integrating μ(x), every reported well and barrier depends on this conversion. Please either use an estimator that accounts for irregular sampling, or provide a quantitative bound on the bias, or validate the procedure on simulated diffusions with known drift and irregular sampling.","section":"Sect. 2.2, Eq. (4)"},{"comment":"The double-well topology appears only for windows of at least two months and disappears for one-month windows. The paper dismisses the one-month result as 'data insufficiency in the tails' without supporting evidence. An equally plausible explanation is regime pooling: E[Δx|x] estimated over a long window mixes distinct market regimes, and a mixture of linear drifts can produce multiple zero crossings even when each regime's drift is linear. A null-model test using regime-switching linear diffusions, or a bootstrap that resamples within shorter sub-windows, is needed to show that the double wells are not an aggregation artifact.","section":"Sect. 3.1, Figs. 1 and 2"},{"comment":"No uncertainty quantification is provided. The drift, volatility, and potential curves are shown as single lines, and the 'small local minimum' in Fig. 4 could easily be within sampling noise. Confidence bands from bootstrap or sub-sampling should be reported, and the barrier height separating a local from a global minimum should be compared with the estimation uncertainty. Without this, the metastable-state interpretation is not statistically supported.","section":"Figs. 1, 3, 4; Sect. 4.0.1"},{"comment":"The observed weakening of non-linearity with increasing lag is not by itself evidence of scale-dependent dynamics. For any diffusion with non-linear drift, finite-lag conditional moments generally become less non-linear as the lag grows, because the transition density is smoothed by diffusion. To support the multi-scale claim, the paper should compare the estimated lag dependence against that predicted by the estimated infinitesimal drift on simulated data, or against an explicit finite-lag formula.","section":"Sect. 4.0.1"}],"minor_comments":[{"comment":"The captions refer to 'UBTC-WETH' while the text discusses 'WBTC-WETH'. Also, Fig. 4 is described in Sect. 3.3 as the USDC-WETH 2025 pool, but the caption says 'UBTC-WETH 2025'. Please correct the naming to avoid confusion.","section":"Figs. 3 and 4 captions"},{"comment":"Typo: 'onbtained' should be 'obtained'.","section":"Sect. 3.1"},{"comment":"The statement that each observation is the log-price 'averaged over 15 consecutive transactions' is ambiguous. Clarify whether this is an average of log-prices or of transaction prices, and how the timestamp for the averaged observation is defined.","section":"Sect. 2.1"},{"comment":"The conversion to 'annualized' rates is not described. Please give the scaling factors used and clarify whether the same scaling is applied at all lags.","section":"Sect. 2.2"},{"comment":"The figures contain many overlapping curves. The paper would benefit from clearer line styles, error bands, or separate panels so that the reader can identify individual lags and windows.","section":"Figs. 1–4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is positioned as validation of the author's own theoretical program, and the selection of two-month windows and the interpretation of short-window results lean heavily on that prior narrative. This is not in itself disqualifying, but it raises the bar for the empirical analysis: the double-well claim needs to survive null models and uncertainty quantification. I would also encourage a data/code availability statement, since the paper does not provide reproducibility details."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nShort version: this is a legitimate empirical test of a specific theory, applied to a new public dataset, and it produces a clear qualitative result—nonlinear drifts and non-quadratic potentials are the norm in intraday Uniswap v3 crypto data. But the paper's flashiest claim, the double-well (meta-stable) potentials, is built on a comparison that could easily be a regime-pooling artifact, and the absence of any uncertainty quantification makes it hard to know what to trust.\n\nWhat's genuinely new: Kramers-Moyal analysis of high-frequency AMM data, using public transaction data from Uniswap v3 on Arbitrum. Unlike Wand et al. (2024) on stocks, who found support for quadratic drift but not cubic/quartic, this paper finds strong nonlinearity in the drift at short lags, and the weakening with sampling frequency is a nice empirical regularity. The author is also honest that one-month windows give single wells, and he flags the stochastic-drift alternative in Section 4.0.2. That's the kind of engagement you want.\n\nThe soft spots are real. First, no error bars anywhere. The drift and potential curves are presented as exact functions, and the difference between a single well and a double well can be a few points in the tails. A simple bootstrap over blocks or a null model—for example, simulate a linear drift with regime switching and run the same estimator—would test whether the double-well is real or an artifact of pooling heterogeneous periods. The paper does neither. Second, the window-length dependence: two-month windows show double wells, one-month windows show single wells, and the dismissal of the one-month result as 'data insufficiency' is not supported. The stress-test's alternative—that shorter windows are better approximations of a single regime—is equally plausible and never tested. Third, the conversion of Kramers-Moyal coefficients to physical drift using a single average dt for irregularly spaced data is a crude approximation; the paper doesn't discuss its effect on the shapes of the potentials.\n\nNone of this is fatal to the program. The non-quadratic, single-well potentials are already a solid empirical contribution. But the meta-stability (double-well/instanton) conclusion is not established. Who should read it: anyone working on market microstructure or stochastic models of crypto prices, and anyone who wants an example of how easy it is to over-read a non-parametric estimate. I'd send it to a serious referee—the data are public, the method is standard, and the weaknesses are addressable in revision—but my own verdict would be conditional, not accept.\n\nBest,\n[You]","headline":"Worth a serious look, but the meta-stability claim is not yet statistically secured: the paper shows clear nonlinear drifts in Uniswap v3 data, but the double-well potentials appear only in 2-month windows and come without error bars or a null-model test.","tokens_in":8172,"tokens_out":2916,"would_cite":false,"duration_ms":33587,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62P05","60J60","91G80"],"pacs":[],"model":"deepseek-v4-flash","headline":"Using high-frequency crypto data, this paper claims log-prices follow strongly non-linear drifts that produce non-quadratic potentials, switching between single-well and double-well shapes with sampling frequency and market regime.","keywords":["non-linear drift","double-well potential","Kramers-Moyal expansion","cryptocurrency market microstructure","automated market maker","metastable dynamics","high-frequency finance","Langevin equation"],"falsifier":"Run the identical estimation pipeline on synthetic data generated from a linear Ornstein-Uhlenbeck process with the same irregular transaction timestamps and sample sizes; if linear inputs routinely produce comparable non-quadratic drifts or double wells, the reported nonlinearity is an artifact of the estimation procedure rather than genuine market structure.","tokens_in":7349,"feed_emoji":"📊","tokens_out":7245,"duration_ms":73518,"temperature":0.7,"texified_at":"2026-08-05T20:22:50.737453+00:00","pith_summary":"This paper tests a long-standing hypothesis that financial prices are not driven by linear diffusion but by non-linear drifts that can be read directly from data. It estimates the drift and volatility of crypto log-prices non-parametrically with a Kramers-Moyal expansion, then integrates the drift to recover an effective potential for the price. The central finding is that the drift is strongly non-linear at intraday sampling, the potential is non-quadratic, and its shape flips between a single well (stable regime) and a double well (metastable, stress-like regime) depending on the look-back window and sampling frequency. If correct, this gives direct empirical support to the author's earlier theory that market dynamics live in non-quadratic potentials, and offers a data-driven way to detect market stress from the topology of the recovered potential.","texify_model":"deepseek-v4-flash","texify_usage":{"total_tokens":2883,"prompt_tokens":683,"completion_tokens":2200,"prompt_tokens_details":{"cached_tokens":0},"prompt_cache_hit_tokens":0,"prompt_cache_miss_tokens":683,"completion_tokens_details":{"reasoning_tokens":1607}},"feed_headline":"Crypto prices reveal double-well 'energy' landscapes at intraday scales","feed_subtitle":"At sub-hour sampling, drift turns non-linear, so the potential flips between single- and double-well shapes.","key_machinery":"The Kramers-Moyal expansion, which estimates the first two jump coefficients $K1(x)$ and $K2(x)$ from empirical time series without assuming a parametric model. These are converted to drift $\\mu(x)=K1(x)/\\Delta t$ and diffusion $\\sigma^2(x)/2=K2(x)/(2\\Delta t)$, then to the potential $U(x)$ by numerically integrating $\\mu(x)=-\\partial U/\\partial x$. The shape of $U(x)$, single-well versus double-well, is the diagnostic that carries the argument.","core_discovery":"The paper claims that model-free estimates of the drift of crypto log-prices reject a linear force and instead show cubic-or-higher non-linearities, so the implied potential $U(x) = -\\int \\mu(x) dx$ is genuinely non-quadratic. In two-month windows of a dollar-anchored pool, the potential frequently has a double-well shape at sub-hour sampling, while shorter one-month windows show single wells, which the paper attributes to data insufficiency in the tails. Non-linearity weakens as sampling frequency drops, and a crypto-cross pair tends to single-well but still non-quadratic potentials. The double-well cases are interpreted as metastable dynamics: the 'particle' log-price sits in a local minimum","pith_inferences":["The conversion of irregular transaction intervals into a single average time step in Eq. (4) is a place where the pipeline could manufacture apparent non-linearity; re-estimating with arrival-time-weighted schemes or testing on synthetic data would clarify this.","Because the double-well structure appears in specific two-month windows, a natural next step is to align those windows with identifiable market-stress events such as volatility spikes or liquidity crunches in the underlying crypto market.","The same non-parametric machinery could be applied to intraday equity or FX data to see whether non-quadratic potentials are a general market feature or specific to automated-market-maker microstructure.","If the instanton interpretation is right, short-window options on the underlying should show related signatures, such as bimodal return distributions, offering an independent test of the metastable picture."],"forward_implications":["Intraday models that assume linear drifts miss the dominant source of non-linearity; state-dependent mean reversion matters at sub-hour scales.","The same market can appear stable at hourly sampling and metastable at ten-minute sampling, so potential topology is inherently time-scale dependent.","A double-well potential implies metastable quasi-equilibria whose lifetime is set by barrier height; rapid transitions between wells may be misattributed to elevated Brownian volatility if ignored.","The single-versus-double-well diagnostic works without parametric assumptions and can be applied to any high-frequency financial series.","Shorter observation windows (one month or less) may under-detect double wells due to tail insufficiency, so the choice of window drives qualitative conclusions."],"supporting_citations":[{"why":"Supplies the data source and its concentrated-liquidity microstructure, defining the transaction stream used for estimation.","marker":"[1]"},{"why":"The classic linear-diffusion baseline (geometric Brownian motion) whose adequacy the paper questions.","marker":"[2]"},{"why":"Provides the Kramers-Moyal expansion and the relations between its coefficients and physical drift and diffusion.","marker":"[4]"},{"why":"Supplies the software implementation of the Kramers-Moyal coefficient estimation used in the analysis.","marker":"[6]"},{"why":"The core theoretical framework being validated, proposing non-linear drifts originating from money flows and metastable dynamics.","marker":"[7]"},{"why":"Extends the non-linear model to multiple assets with phase transitions, framing the potential shapes the paper seeks.","marker":"[10]"},{"why":"Introduces the 'marketron' picture of self-propelling markets and metastable dynamics used to interpret the double-well potentials.","marker":"[11]"},{"why":"Prior intraday empirical test of quartic versus cubic potentials that motivates the model-free approach adopted here.","marker":"[13]"}],"fun_headline_variants":["Crypto data reject linear drift: double-well potential at sub-hour scale","Metastable crypto markets show flipping potential wells in high-frequency data","Sub-hour crypto drifts are non-linear, implying double-well energy landscape","Crypto log-price potential flips from single to double well at sub-hour scale","Non-quadratic potential confirmed in crypto market maker data"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"Everything rests on treating the log-price as a one-dimensional Markov diffusion whose drift depends only on the current price, and on converting irregular transaction timings into a single average time step when turning estimated coefficients into physical drift and volatility.","fun_headline_variants_meta":{"raw":{"variants":["Crypto data reject linear drift: double-well potential at sub-hour scale","Metastable crypto markets show flipping potential wells in high-frequency data","Sub-hour crypto drifts are non-linear, implying double-well energy landscape","Crypto log-price potential flips from single to double well at sub-hour scale","Non-quadratic potential confirmed in crypto market maker data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001173,"raw_usage":{"total_tokens":4665,"prompt_tokens":697,"completion_tokens":3968,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":441,"completion_tokens_details":{"reasoning_tokens":3884}},"tokens_in":441,"tokens_out":3968,"duration_ms":34632,"temperature":1.0,"reasoning_tokens":3884,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T11:14:24.755002+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the identical estimation pipeline on synthetic data generated from a linear Ornstein-Uhlenbeck process with the same irregular transaction timestamps and sample sizes; if linear inputs routinely produce comparable non-quadratic drifts or double wells, the reported nonlinearity is an artifact of the estimation procedure rather than genuine market structure.","supporting_citations":[{"cited_title":"Uniswap v3 Core","cited_arxiv_id":null,"evidence_quote":"Supplies the data source and its concentrated-liquidity microstructure, defining the transaction stream used for estimation."},{"cited_title":"The Pricing of Options and Corporate Liabilities","cited_arxiv_id":null,"evidence_quote":"The classic linear-diffusion baseline (geometric Brownian motion) whose adequacy the paper questions."},{"cited_title":"Gardiner, Handbook of Stochastic Methods, Third Ed., Springer (2004)","cited_arxiv_id":null,"evidence_quote":"Provides the Kramers-Moyal expansion and the relations between its coefficients and physical drift and diffusion."},{"cited_title":"kramersmoyal: Kramers--Moyal coefficients for stochastic processes","cited_arxiv_id":"1912.09737","evidence_quote":"Supplies the software implementation of the Kramers-Moyal coefficient estimation used in the analysis."},{"cited_title":"Quantum Equilibrium-Disequilibrium: Asset Price Dynam- ics, Symmetry Breaking, and Defaults as Dissipative Instantons","cited_arxiv_id":null,"evidence_quote":"The core theoretical framework being validated, proposing non-linear drifts originating from money flows and metastable dynamics."},{"cited_title":"Phases of MANES: Multi-Asset Non-Equilibrium Skew Model of a Strongly Non- Linear Market with Phase Transitions","cited_arxiv_id":null,"evidence_quote":"Extends the non-linear model to multiple assets with phase transitions, framing the potential shapes the paper seeks."},{"cited_title":"Marketron games: Self-propelling stocks vs dumb money and metastable dynamics of the Good, Bad and Ugly markets","cited_arxiv_id":"2501.12676","evidence_quote":"Introduces the 'marketron' picture of self-propelling markets and metastable dynamics used to interpret the double-well potentials."},{"cited_title":"Estimating Stable Fixed Points and Langevin Potentials for Financial Dynamics","cited_arxiv_id":"2309.12082","evidence_quote":"Prior intraday empirical test of quartic versus cubic potentials that motivates the model-free approach adopted here."}],"review_version":1}