These are not aspirations. They are enforced in code and in tests, and they are
the reason some pages show less than a competitor’s would.
Measured, not estimated
Every figure is derived from chain data we ingested. Where a derivation isn’t
implemented: CoW Swap in the pair finder, Curve’s CryptoSwap and metapools,
LST-ratio oracle adapters, GHO’s discount machinery, E-mode parameters: the
surface says so and shows nothing. It does not approximate with a formula
that doesn’t match the protocol.
A missing value is published as missing
Nulls survive aggregation. A group with no priced rows reports “not priced”, not
$0, because summing nulls would merge “we cannot price this” with “nothing
happened”. Coverage counts sit next to totals so you can see what the total
covers.
Validation against the chain, not against ourselves
Derived pipelines are checked against archive node calls and required to agree
exactly before shipping, not “close enough”. The Aave pipeline’s record is on
its guide.
Partial work says it is partial
A snapshot still being built reports how much is in. A dataset with a growing
backfill shows dashes where nothing has been computed, never a zero standing in
for unknown.
Corrections are visible
When a re-derivation disagrees with a value we published: a decoding fix, a
corrected decimal, a reorg repair: the value is flagged with both figures
rather than quietly rewritten. A number that changed is itself a finding.
Hashes are never truncated
Not in tables, not in exports, not in webhook payloads. Researchers match them
by eye.