Operating notes, including the bad weeks.

A price guard is a cost until the day it isn't. Ours fired repeatedly this week during a thin overnight window when the GME pool printed a spike that the 30-minute reference did not corroborate. Each time, the rebalance aborted rather than resetting the range around a price that reverted within minutes.

The forgone fees from those missed resets are real and small. The alternative — placing a fresh range around a spike and holding it as the pool snapped back — is how a managed position converts a temporary dislocation into a permanent loss. We would rather publish the refusals and explain them.

Every blocked action carries its deviation figure in the detail column, so the trade-off is auditable rather than asserted.

Realized volatility on GME sits near 91% against 48% for NVDA. Running the same width on both would put the GME position out of range on ordinary days and leave the NVDA position earning far less than it could on quiet ones.

Wider is not free: it earns less per dollar deployed while in range. The exchange we accept is a lower headline rate for materially more time spent earning it — currently 92.2% of the trailing month for GME, against 98.2% for NVDA.

This is the whole argument for per-asset vaults compressed into two numbers. A generic manager cannot make that trade because it does not know which asset it is holding.

The third market opened this week. The cap did not change, and will not change on the basis of demand — only on the basis of recorded behaviour through a volatility event the strategy has not yet seen.

Caps are unpopular because they turn away deposits. They are also the only honest mechanism a young strategy has for limiting how much capital is exposed to a model that has not been stress-tested. Raising one because the queue is long is how a strategy becomes a story about a strategy.

The widely cited result is that a majority of Uniswap v3 liquidity providers lost money once impermanent loss was netted against fees, with $-260,100,000 of IL against $199,300,000 of fee income across the studied cohort.

The finding that actually shapes product design is quieter: an academic analysis found no statistical evidence that providers who repositioned more frequently performed better. If effort were the missing input, the most active wallets would separate from the pack. They did not.

That is why this protocol is not a dashboard with better tick controls. Better controls hand the same losing decision back to the user with a nicer interface around it.