Khaldoun is right: Uniswap v3 pool data cannot identify a Black-Scholes implied volatility. She says as much in the abstract and formalizes it in a proposition. The practitioner's reason to spend an hour on the paper is the 1276-day ETH/USDC table. It reports one result her discussion passes over, alongside another whose level depends on a window length we could not find printed anywhere in the text. Khaldoun uses the table to argue that the empirical evidence sharpens the identification gap rather than closing it. The evidence supports her.
The fee stream covers one leg
Uniswap v3 allows a liquidity provider to place capital inside a finite price band. The position absorbs swap flow and earns the pool's fee tier while spot remains in range. Once price exits, the LP holds one asset and earns nothing. At the narrowest band the protocol permits, the payoff begins to resemble a short-dated option: the LP carries short convexity in exchange for a stream of fees. Panoptic, an oracle-free perpetual options protocol built on top of Uniswap v3, treats the resemblance literally, streaming premium through time instead of collecting it upfront.
The LP therefore carries two legs. Fees equal feeRate times the volume crossing the range. The second leg is the cost of negative convexity, realized through arbitrage trades that bring a stale AMM price back toward the outside market. The paper writes this cost in its standard continuous-time form, loss versus rebalancing at an instantaneous rate of sigma-squared over eight, and cites a Panoptic blog post for the expression. The economic case comes from the predictable-loss paper of Cartea, Drissi and Monga, cited elsewhere in the text. Khaldoun puts the separation cleanly: even a pool paying a thousand times more in fees faces the same hedging cost for a given realized price path.
Where the closed form gets its units
The derivation starts with Black-Scholes theta. As maturity approaches zero, theta concentrates at the strike and converges in distribution to a mass of K-squared sigma-squared over two. The appendix proves this by substitution, with the exact integral carrying a factor exp(sigma-squared t) before the limit. Khaldoun then spreads that mass evenly over the narrowest band allowed by the protocol, K times tick spacing over ten thousand, producing a rectangular premium height.
She matches this height to observed fee intensity, feeRate times Volume over active liquidity at the tick. Both sides are scaled by the price level K to reconcile their units. Khaldoun calls the match a heuristic bridge. We regard the K scaling as ad hoc; her description says that scaling by K places the two sides on a common footing as price-denominated intensities. The result is
sigma_fee = 2 x feeRate x sqrt(Volume / L_tick).
Linearity in the fee tier follows only because Uniswap's standard tiers happen to set tick spacing at roughly two times ten thousand times feeRate. The tiers {0.0005, 0.0030, 0.01} pair with spacings {10, 60, 200}, producing minimal range widths near 0.10%, 0.60% and 2.02%. Khaldoun explicitly describes that proportionality as a design artifact of the standard tiers rather than a universal law. The closed form consequently depends on a configuration table. Increasing the fee tier raises sigma_fee even when the underlying price process stays unchanged, a weakness the paper identifies itself.
Proposition 2 supplies the formal statement, and it comes close to an identity. If Fees are alpha times the full short-option cost, sigma_fee squared equals alpha times structural volatility squared. Alpha belongs to (0,1] as an accounting matter. It is not a pool state variable, and the map remains one-to-many.
Nobody should be surprised.
The empirical question is whether alpha behaves like a slowly moving constant.
1276 days, one ETH/USDC pool
The sample covers one pool, the 30bps ETH/USDC, observed daily from January 2023 to July 2026 and compared with realized volatility for ETH spot. Correlation is 0.87. Mean sigma_fee is 57.06%, versus mean RV of 53.53%; their standard deviations are 30.12% and 28.83%. The means differ by 3.53pp, while the standard deviations fall within 1.3 points.
Those figures exceed what the derivation seemed likely to produce, especially after the units were reconciled by scaling both sides by K, an ad hoc step in our view. Khaldoun properly calls the section illustrative rather than a general characterization. It covers one pool and one tier. She discusses the mean premium and interprets it as a volatility risk premium. The close match in dispersion appears only in Table 1 and receives no attention in her prose.
The empirical capture ratio is the squared ratio of sigma_fee to RV. Its mean is 1.33, its median 1.18, its IQR [0.92, 1.53], and its maximum 11.33. Table 1 shows the ratio exceeding 1 on 68.26% of days for the ETH/USDC 30bps pool from 2023 to 2026. Khaldoun defines this quantity relative to realized volatility rather than structural sigma_*. If RV is biased, she argues, the empirical ratio has no reason to obey the same (0,1] bound. She declines to treat it as a direct test of that bound, correctly.
Its variation still matters. The median is 1.18, the IQR [0.92, 1.53], and the maximum 11.33. A 1st-to-99th percentile trimmed mean of 1.30 shows that values above-1 describe the typical day rather than a handful of spikes. Khaldoun favors the volatility risk premium interpretation while conceding that a mechanical explanation can do the work alone. RV lags during sharp moves, and squaring an already-lagging ratio creates precisely this right skew. Her strongest observation follows: the ratio becomes least stable when the underlying market moves most, exactly when any useful capture ratio would need to hold together.
How much depends on the window?
sigma_fee scales with the square root of volume over a window, while average active liquidity does not increase with that window. The resulting quantity therefore grows roughly with the square root of window length. Longer windows print higher volatility.
We did not find a numerical observation window or realized volatility window anywhere in the text. This omission affects the reported results unevenly. A constant multiplier leaves correlation unchanged, so the level ambiguity does not directly disturb the 0.87, although changing the window would also resample and smooth both series. The 3.53pp mean spread, the 68.26% proportion of days above RV, and the full capture-ratio distribution all depend on a choice the reader cannot see.
We could not test these claims ourselves. The formula requires pool-level swap volume and active liquidity at the current tick. Our crypto data consists of exchange price and volume bars, without chain state or tick liquidity. L_tick is the denominator that gives the measure its meaning, and bar data offers no substitute.
Beyond the theorem
Khaldoun's conclusion defends a narrower claim directly: this framing remains powerful and provides an oracle-free state variable tailored to Panoptic's premium mechanics. She gives four uses: scaling premium accrual, comparing relative fee intensity across pools, informing collateral or margin buffers, and detecting bursts of activity relative to liquidity. Her argument is that each use depends on the ranking and dynamics of sigma_fee rather than numerical equivalence with structural volatility. We accept the first and last.
We would skip ranking across pools. sigma_fee is mechanically linear in the fee tier, leaving a 1% pool and a 5bp pool on different scales. Khaldoun's own comparability condition calls for similar fee tier, user base, arbitrage efficiency and trading microstructure. In practice, that condition reduces the exercise to comparing a pool with its own history.
The paper offers no strategy or backtest. It reads as a definition paper with one descriptive comparison, presented through two figures and a table.
The sample finishes in July 2026, the same month as the paper's date. Nothing in the study is out of sample.
A second pool at a different tier over the same window would change our view of the comparison. Khaldoun promises such an extension, leaving a systematic study across pools, fee tiers and volatility regimes, together with a forward-looking benchmark such as an options implied volatility index, for future work. The requested evidence therefore follows her own agenda. Printing the window lengths is our addition.
The acknowledgments thank the Panoptic team and Guillaume Lambert, the protocol's inventor. Even so, the paper concludes that the fee-implied number is not a CEX-style implied volatility. A paper that thanks the inventor and still refuses the CEX label deserves to be read.