09Data policy

Read protocol metrics correctly

Metrics describe different parts of a liquidity market. Clear definitions prevent deposited capital, trading activity, fee generation, and protocol revenue from being confused.

TVLThe value of all assets currently supplied to Silo contracts.

Protocol-owned liquidityThe portion of pool capital economically owned by Silo.

24-hour volumeThe value of executed trades during the previous 24 hours.

Trading feesFees paid by traders for access to pool liquidity.

LP feesThe share of trading fees attributable to liquidity providers.

Protocol revenueThe fee share retained by Silo, if protocol fees are enabled.

LP APRA fee-based estimate calculated with a stated time period and methodology.

Liquidity depthThe amount of capital available around specific execution prices.

These eight are separated because each answers a different question, and the questions are easy to conflate when every answer is denominated in dollars. TVL answers how much capital is present. Volume answers how much trading occurred. Fees answer what that trading paid. Revenue answers what the protocol kept. A figure combining any two of them answers nothing at all, while looking more impressive than either one alone.

Pool analytics

A pool page should show the market pair, current and reference prices, TVL, 24-hour volume, 24-hour fees, current fee, liquidity depth, bin distribution, user position, asset composition, and position earnings. Projected values must be labeled as estimates.

The list is long because a pool page serves two readers with different needs. A trader needs to know what a trade costs right now, which is current price, depth, and fee. An LP needs to know what a position is doing over time, which is composition, range status, accrued fees, and how far the market has moved relative to its reference. A page serving only the first reader is a quote screen. A page serving only the second is a portfolio view. A pool page has to be both.

How an APR figure is built

An LP APR is an estimate derived from fees over a period, not a rate anyone commits to. It must always be published with the inputs that produced it, because the same pool can show very different numbers under different windows.

Fee basisWhich fees are counted, over what window, and whether they are LP fees or total fees.

Capital basisWhich capital the fees are measured against: all TVL, or only the liquidity that was active.

Time windowThe stated period, shown explicitly rather than annualized silently.

LabelMarked as an estimate wherever it appears, including in summaries and cards.

An APR figure needs all four disclosures because it is an annualization of a period that may bear no relationship to a year. A pool that earned unusual fees during one volatile day produces an enormous annualized number from that day, and the number is arithmetically correct and completely uninformative. Publishing the window is what lets a reader discount it appropriately. Publishing the capital basis is what stops a pool from improving its own figure by measuring against only the liquidity that happened to be active.

The window changes the answer

Almost every derived metric depends on the period it is measured over, and that period is chosen by whoever presents the number. A figure without its window is not a measurement.

Short windowsResponsive to current conditions, and easily distorted by a single large trade.

Long windowsStable and comparable, but slow to reflect a market that has genuinely changed.

Windows chosen after the factA period selected once the data is known, which describes the choice rather than the market.

The same pool can honestly report very different APR figures over one day, seven days, and thirty days. Publishing the window alongside the figure is what makes those numbers legible rather than contradictory.

The defensible practice is to fix the windows in advance and show the same ones everywhere, including when they are unflattering. A surface displaying a seven-day figure during a good week and a thirty-day figure during a bad one is not reporting a metric, it is selecting an argument. That consistency is what makes comparison between markets possible at all.

Metrics that must not be combined

Most metric errors come from adding numbers that describe different things. These pairs measure different quantities and must never be presented as one figure.

TVL and volumeDeposited capital versus executed trading activity.
TVL and protocol-owned liquidityAll deposits versus the share Silo economically owns.
Trading fees and protocol revenueWhat traders pay versus the share Silo retains.
Fee income and underlying yieldMarket activity versus yield produced by the asset itself.

Each of those pairs has a plausible-sounding combination, which is what makes them worth naming individually. Adding trading fees to underlying yield produces a single headline return that no participant actually receives on that basis, because the two are earned differently, carry different risks, and may accrue to different parties. The combined figure is not a summary of its components. It is a new claim, and a weaker one.

Measuring depth

Depth is the metric most often replaced by a proxy. TVL is easy to compute and easy to compare, but it does not describe what a trader will actually receive.

Depth at a priceHow much can be traded before execution moves beyond a stated percentage.
Depth is directionalA market can be deep for buyers and thin for sellers, because bins above and below the market hold different assets.
TVL is not depthCapital parked far from the active bin contributes to TVL and contributes nothing to execution.
Depth is not durableAs price moves, positions leave range, so depth that existed an hour ago may not exist now.

What an analytics surface owes the reader

These rules apply wherever a number appears, including cards, summaries, tooltips, and marketing surfaces, not only on a dedicated analytics page.

State the windowEvery derived figure carries the period it was measured over.

Separate measured from projectedWhat happened and what is extrapolated are never presented as the same kind of value.

Show the age of the dataA figure computed from a stale input is labeled as such rather than shown as current.

Keep definitions constantA term means the same thing on every surface, as defined in the glossary.