What DeFier measures, and what it refuses to

This document exists because the product's main risk is not a crash. It is a number that looks right and is not.

The rule the whole codebase is built on: a figure that cannot be measured is excluded and said on screen, never estimated. A bad number is worse than no number, because somebody acts on it.

Measured, from the chain

• Every concentrated liquidity position a wallet has opened on Aerodrome Slipstream or Uniswap V3 on Base, including positions whose NFT was burned.
• Deposits, increases, decreases and withdrawals, each valued at the price of the day it happened.
• Fees collected, and gauge emissions claimed, kept as separate figures.
• Gas paid, per position.
• The counterfactual: the same tokens valued as if they had never left the wallet.
• Trades: transactions where one asset left the wallet and another arrived, valued at today's price on both sides.

Declared, not estimated

Unresolved historical prices. If the price of a token on the day of a deposit cannot be resolved, that position is excluded from the total and the scope stops saying "all time". It is not filled in with today's price or an adjacent day.

Positions that could not be read. A position the chain would not answer for is counted as unread, not as empty.

Trades in tokens with no trusted price. Both sides must be priced with a confidence of at least 0.8 or the trade is left out and counted. Base wallets collect airdropped tokens with fabricated prices, and one of those would produce the largest figure in the report.

A truncated history. When the transfer index pages out before the wallet's full history is read, the answer says "partial" rather than "all time".

Empty pools. A pool with no liquidity at the current price cannot pay a fee, whatever TVL is published for it. The screen says there is nothing at this price — a fact about the pool — rather than "we could not read the liquidity", which is an apology for our own limits. Those are different statements and the reader needs to know which one they are getting.

Health factors come from the protocol, never from us

A health factor decides whether someone is liquidated, so it is computed from that protocol's own oracle and its own collateral factors, not from our prices. Valuing the collateral one way and the debt another produces a number that belongs to nobody, and it would be wrong in the direction that matters: comfortingly high.

Every figure we derive is then checked against one the protocol states itself. Aave gives account totals beside the per asset rows. Moonwell gives account liquidity beside the balances. Compound v3 says outright whether an account is liquidatable. When our figure and theirs disagree, neither is printed and the screen says so. A health factor that is close is worse than none, because it is the number people decide by.

The same check caught a real error: Compound quotes prices in its own market's base asset, so in the WETH market a supplied balance read as dollars was wrong by the price of ETH. The ratio survived, being taken inside one denomination, which is exactly why every dollar figure had to be priced a second way before it could be shown.

Not measured, and not claimed

• Anything outside Base. No other chain is read, so no total is a portfolio total.
• Lending outside Aave v3, Moonwell and Compound v3. Morpho in particular is not read. Every screen that shows lending names what was checked and names what was not, because a wallet borrowing somewhere we do not look would otherwise read as a wallet with no debt, and that is a wrong answer rather than a missing one.
• v2-style AMM positions. Frozen for lack of a verifiable test case rather than shipped as an approximation.
• What the wallet did next. The trades section compares the two sides of one transaction. It is not a profit and loss: someone who sold AERO for USDC and bought ETH the next day is not described by that line at all, which is why the copy states amounts and never a verdict.
• Anything about the future. The simulator solves a range's economics under stated assumptions. It is not a prediction and the assumptions are printed beside the result.

Concentration travels with the headline

A result driven overwhelmingly by one position is a different fact from the same number spread across ten. Where the headline goes, the concentration goes with it — including onto the share image, which is where the number travels furthest from anyone who could explain it.

Why this is written down

Because the tempting failure is silent. Nothing crashes when a total quietly covers three of five positions; it simply reads as a lifetime. Writing the rule down makes the silence a bug rather than a style.