← armsys.xyz

Does the pool beat simply holding?

Every ARMSys pool, measured daily against the rebalancing portfolio — the same basket, marked to market, without providing liquidity. Fees are compared to LVR, not to zero. Published as measured, including the days we come out behind.

updated 11 Aug 2026, 17:58 UTC
daily at 08:00 UTC

Fees vs LVR

Rolling last 24 hours from the chain head. 95% interval by bootstrap over the individual swaps; “inconclusive” means it covers zero, and then the point estimate must not be read as a result.

swapsvolumefee bpnet bp 95% CIverdictreference
ETH/USDC — ARMSys614$11,0810.79+3.89[+3.0, +4.7]beats holdingBinance, 1s
v4 static 0.30%284$232,53034.99+32.22[+30.2, +34.1]beats holdingBinance, 1s
v3 0.05%9,838$5,390,8545.00+5.35[+4.9, +5.8]beats holdingBinance, 1s
NVDA17$44312.48+0.35[-54.8, +57.1]inconclusiveneighbour pool
INTC12$17613.13-18.21[-42.6, +10.7]inconclusiveneighbour pool
NVDA21$51211.27+7.04[-43.0, +59.3]inconclusiveHL perp, 1m
INTC27$37613.13+39.47[+25.5, +52.6]beats holdingHL perp, 1m
SPCX19$34987.60+19.49[+14.4, +25.0]beats holdingneighbour pool

Weekly verdict

One square per pool per UTC day, week anchored on Friday. A day with fewer than 10 priced swaps is treated as inconclusive whatever the interval says.

Fri07Sat08Sun09Mon10Tue11Wed12Thu13reference
ETH/USDCBinance 1m
NVDAHL perp 1m
INTCHL perp 1m
SPCXneighbour pool

Token basis vs exchange

Spot at generation time. A separate risk for a liquidity provider, unrelated to LVR: if the premium of the token over the underlying collapses, the holder loses regardless of how the fee ladder performed.

our poolHL perpMassive basis to HL
NVDA$219.58$217.93+0.76%
INTC$97.06$97.33-0.28%
SPCX$132.88no perp

How this is measured

For every external swap, LVR = qty × (external price − executed price), with the external price taken synchronously with the trade. There is no horizon parameter, so a thin tape does not break the estimate, and a persistent price gap between venues does not create a phantom signal — it enters only to the extent someone actually traded on it. Fees come from the fee field the pool itself writes into each swap, so the figure is the rate actually charged, not the nominal tier. Our own keeper traffic is excluded everywhere: it is warming, not demand.

LVR is non-negative by construction (σ²/8 per unit time); what is computed here is its realised estimate, which is unbiased but noisy. A negative reading means the window was too short for the signal to clear the noise — never that arbitrage paid the liquidity provider.

References. ETH/USDC is measured against Binance; NVDA and INTC against both a neighbouring pool and the Hyperliquid perpetual, deliberately — two independent measurements of one quantity, and a disagreement in sign shows that one of them is lying. Where they disagree, trust the row with the larger swap count. SPCX has neither a listed price nor a perpetual, because SpaceX is private, so only the neighbouring pool is available. A neighbouring pool understates LVR in quiet hours and is biased.

What this is not. It is the pool's LVR, not an individual position's profit and loss — that depends on the provider's range, their time in range and their gamma. It is also normalised by volume for comparability across rows, which flatters low-turnover pools: LVR accrues on capital over time while fees accrue on volume.