Backfill
Filling in historical data for a period that was not collected at the time, by reconstructing it after the fact.
Backfilling happens whenever a provider adds a new chain or asset and rebuilds its history from archive nodes, introduces a new metric and computes it for past dates, or patches a gap left by an outage. It is normal and often necessary, but it changes what the data means. Reconstruction uses today's knowledge, including today's address labels, today's contract list, and today's venue inclusion rules, none of which existed at the historical date, which imports look-ahead bias into anything built on the series. Re-derivation from archives can also fail to reproduce what was actually published at the time. Good practice is to mark backfilled ranges explicitly so users can distinguish them from data captured live.
In practice
A metric introduced this year but shown stretching back several years was necessarily computed retrospectively, using the contract and label sets known now.
The common misunderstanding
A long, smooth history does not mean the data was collected throughout that period; much of it may have been recomputed later under today's rules.