The Reynolds number of an on-chain price

St0kes

Measured on Ethereum mainnet

Abstract. We define an empirical, dimensionless price-speed ratio, named Re by analogy with fluid mechanics, as market price change per minute divided by a scale estimated from oracle updates. In 194,915 asset-minutes of ETH/USD, BTC/USD and LINK/USD, the Chainlink answer differed from the Binance opening price by more than 1% in 1 of 185,370 observations below Re = 1 and 61 of 9,545 at or above it. All 17 departures above 2% occurred at Re ≥ 5.16 under the baseline calibration. Other calibrations change both the scale and the exceptions; Re = 1 is not a guaranteed safety boundary. This is an in-sample association, not a validated forecast or a consequence of the Navier–Stokes equations. Figures 1 and 2 show a separate live estimate using fixed Uniswap pools.

New research prototype: When truth arrives late — the Navier–Stokes observability lab ↗

Keywords. oracles, liquidation, Reynolds number, Burgers vortex, Chainlink, Uniswap v3

1Introduction

In September 2026 a finite time singularity of the three dimensional Navier–Stokes equations under smooth forcing was announced, with a machine checked proof [1]. The mechanism at the heart of that result is old: a column of fluid is stretched along its axis, stretching concentrates its rotation, tighter rotation draws more fluid inward, and the inflow feeds the stretching. The Burgers vortex [2] is the canonical exact solution exhibiting this loop, and Figure 1 draws it.

The fluid Reynolds number compares inertial and viscous effects [3]. Transition depends on the flow and its geometry; there is no universal fluid threshold at one. Here the fluid analogy motivates a visualisation and a dimensionless market statistic, not a physical equivalence or a proof of market instability.

An on-chain price is different. The market for an asset and the oracle that reports its price are separate contracts on the same chain, and each can be read on its own. This paper compares market price change with an empirical scale estimated from oracle history, over roughly 45 days of three collateral feeds. Larger ratios coincide with more frequent large oracle departures in this sample.

inward spiral

axial stretching

ETH/USD now Re —

Figure 1. Streamlines of a Burgers vortex, 142 paths integrated through the velocity field (4)–(6) with a fourth order Runge–Kutta scheme. Warm lines lie inside the core and are mostly stretched; cool lines begin outside it and spiral inward. The axial strain is not fixed: it is driven by the live Reynolds number of the ETH/USD feed through the volume preserving map (7), as an illustration of the statistic; the shape is not a simulation of market dynamics. Drag to rotate.

animates a synthetic pulse peaking at the recorded LINK value, Re = 26.4; this is not a historical time-series replay.

2Two terms

2.1Push

Let P(t) be the reference price at time t: Binance minute opens in the historical analysis, or a fixed Uniswap v3 pool in the live display. Push is the absolute net price change over five minutes, in percent per minute; it does not measure the full path or intrawindow volatility:

(1)

2.2Damping

Chainlink updates are triggered by deviation and heartbeat conditions [5]. We define an empirical scale using a typical observed answer change δ and an inter-update interval statistic τ:

(2)

also in percent per minute. Both quantities are measured from the feed's own history: δ as the median absolute change between consecutive answers, and τ as the 25th percentile of positive inter-update gaps shorter than 3,500 seconds. These are proxies: δ is not the configured deviation threshold, and τ is not measured publication latency or a proven speed limit. Section 5 repeats the measurement under three other choices of τ.

2.3Reynolds number

(3)

Re below one means the measured market change is below the fitted scale; above one it exceeds that scale. Neither value guarantees how accurately the oracle is tracking.

Table 1. The two terms per feed, from 12,000 rounds of each aggregator. Damping is (2). The last column is the largest Reynolds number reached in the 45 day sample.

Feed	δ	τ	κ	max Re

2.4The figure

The Burgers vortex is the steady solution of the Navier–Stokes equations in which an axial strain of rate a stretches a line vortex of circulation Γ, with viscosity setting a core radius r_c. In cylindrical coordinates its velocity field is

(4)

(5)

(6)

Figure 1 integrates (4)–(6) from 142 seeds. To connect the figure to the measurement, the rendered column is deformed by an incompressible axial strain of factor λ,

(7)

which preserves volume, with s ∈ [0, 1] a monotone function of the live Reynolds number of ETH/USD, s = min(1, (Re / 26.4)^0.55). The normalising constant 26.4 is the largest Reynolds number in the sample (Table 4).

3Data

The source cache contains 12,000 rounds from each of three Chainlink aggregators on Ethereum mainnet, spanning approximately 288, 360 and 264 days. These are finite snapshots of aggregator history, not complete proxy histories across all upgrades. The market reference contains 65,000 one-minute Binance bars per asset (ETHUSDT, BTCUSDT and LINKUSDT), ending 15 September 2026. Each opening price is aligned to the latest cached oracle answer at or before the bar timestamp. USD/USDT parity is assumed; stablecoin basis is not removed.

When multiple rounds share a timestamp, this revision retains the highest round ID; the previous script retained the oldest answer. The previous analysis also included five zero-filled lookbacks per feed. This revision excludes those 15 observations, retaining 194,915 asset-minutes with an actual five-minute history. The supplied Coinbase cross-check script covers four selected ETH timestamps only; it does not validate all three assets or the entire sample.

Calibration uses the full cached oracle history, including updates after earlier evaluation minutes, so these results are in-sample. Adjacent minutes and simultaneous assets are correlated. A future test needs calibration frozen before the evaluation period and must distinguish contemporaneous association from prediction.

4Results

Table 2 groups observations by Re. There is one departure above 1% below Re = 1: ETH/USD on 11 September 2026 at 14:04 UTC, with Re = 0.887 and departure +1.063%. The event was hidden by rounding its frequency to 0.00% in the previous draft. Event counts are now shown explicitly.

Table 2. Departure of the oracle from the market by Reynolds number, all three feeds pooled, 194,915 asset-minutes. The rule marks Re = 1. "Tore past 1%" is the share of minutes in the band at which the oracle was more than one percent from the market.

Re	Minutes	Median	99th pct.	Tore past 1%

The observed frequency is 1/185,370 (0.000539%) below Re = 1 and 61/9,545 (0.639078%) at or above it, an empirical risk ratio of about 1,185. This estimate rests on a single event in the lower group; it is not a causal effect or an out-of-sample guarantee. Every departure past two percent in the sample, seventeen asset-minutes in all, occurred at Re ≥ 5.16.

5Sensitivity to calibration

Table 3 changes the inter-update statistic τ, using the same gap filter as the baseline. Below Re = 1 there is 1 departure above 1% for the 25th-percentile choice, 7 for the fastest-tenth choice, and 0 for the median and mean. Therefore the original claim of zero exceptions under all four definitions does not hold. These checks vary one modelling choice; they are not independent validation.

Table 3. The measurement of Table 2 under four definitions of the inter-update interval statistic.

τ taken as	Tore below Re 1	Tore above Re 1	Every 2% tear above

6The largest events

Table 4 lists the five largest absolute departures across all asset-minutes, sorted by magnitude. Multiple rows may describe consecutive minutes of the same episode. ETH and LINK show simultaneous departures on 22 August. These price comparisons alone do not demonstrate an executable liquidation, realised profit, or the absence of that price on every exchange.

Table 4. The largest departures in the sample. u and κ in percent per minute; "Oracle" is the feed's answer at that minute, "Market" the exchange opening price.

Time (UTC)	Feed	Re	u	κ	Oracle	Market	Departure

7Live measurement

Figure 2 reads a recent block and the block 25 positions earlier from public Ethereum nodes¹. It reads the pool state at both fixed blocks, even if no swap occurred, and divides the net price change by the actual elapsed minutes. The interval is approximately five minutes and is shown below. Damping uses the full-precision calibration behind Table 1. The ETH/USD estimate is currently Re —.

The live market reference is an estimate in USD assuming USDC = USD, with WBTC used as a BTC proxy. BTC and LINK are converted through the ETH pool at each corresponding block. Pool basis, stablecoin deviations and spot-price manipulation can affect this display. It uses different venues and sampling from the historical study; current RPC reads do not reproduce that study.

Figure 2. Live Reynolds number of the three feeds, requested twelve seconds after each completed read. Marks at 1, 5 and the largest historical Re are descriptive reference points, not safety limits. Failed or stale reads are shown as unavailable.

¹ publicnode, drpc, llamarpc, ankr and cloudflare, with time-limited failover; the last working node is preferred.

8Reproduction

The historical tables are generated by tools/audit.py from the six cached oracle and Binance files. The generated results include SHA-256 hashes of those inputs, unrounded calibration, counts and the below-one exception. The caches are not included in this static site; full independent reproduction requires those inputs. Separately, the live view uses latestRoundData and decimals on each feed, slot0 at both blocks on each pool, and block headers. Addresses:

References

[1]OpenAI. On the Navier–Stokes Millennium Prize Problem. Finite time blow-up of the three dimensional Navier–Stokes equations under smooth forcing. Announced 8 September 2026, with a formalisation in Lean reported by the authors.
[2]J. M. Burgers. A mathematical model illustrating the theory of turbulence. Advances in Applied Mechanics 1, 171–199, 1948.
[3]O. Reynolds. An experimental investigation of the circumstances which determine whether the motion of water shall be direct or sinuous, and of the law of resistance in parallel channels. Philosophical Transactions of the Royal Society 174, 935–982, 1883.
[4]H. Adams, N. Zinsmeister, M. Salem, R. Keefer, D. Robinson. Uniswap v3 core. 2021.
[5]Chainlink Data Feeds and AggregatorV3Interface API. Documentation.