Backfilled and Revised History
Historical series are recomputed, backfilled, and corrected over time, so a chart today may not match the same chart pulled last month.
How it happens
Time series in this sector are derived rather than recorded. When an adapter is fixed, the correction is usually applied to the whole history rather than only going forward; when a price source is switched, every value computed from it changes; when a venue is removed from a volume aggregate, past volume falls retroactively. Chains themselves contribute, since indexers reprocess data after reorganizations and after schema changes. Survivorship compounds the effect, because protocols that failed are often dropped from coverage, so a backward-looking average of what remains describes only the survivors. The result is a record that appears precise, carries no revision marker, and can differ between two pulls of the same endpoint.
What you can actually observe
Check whether the provider publishes a changelog, versions its methodology, or timestamps revisions, since most do not. Snapshot any series you rely on, store it, and re-pull it later to measure whether history moved. Step changes that align with a methodology note or a code commit rather than an on-chain event are the clearest sign that a revision, not an activity change, produced the shape.
What makes it more or less material
Consider whether the provider versions and logs changes, how far back a revision reached, whether failed protocols remain in coverage, and whether independent providers agree on the same historical period.