SUMMARY: Data lineage is the traceable path from an original observation through every transformation, transfer, and reuse that turns it into a decision-ready record.
A record rarely arrives alone. A dashboard value may begin as a sensor reading, pass through a cleaning rule, inherit a category from a vendor, and enter a model that produces a score. By the time a person sees the result, the number appears settled while the choices that shaped it have disappeared behind the interface.
Lineage restores those choices. It identifies who collected the source, which fields were excluded, when formats changed, and where uncertainty became a default value. This matters most when several downstream systems repeat the same record and make one inherited assumption look like independent confirmation.
Four Questions for Every Record
Origin: what event, person, device, or institution first produced the information?
Transformation: which filters, joins, labels, calculations, or summaries changed its meaning?
Custody: which systems stored or transmitted it, and where could unauthorized edits or silent loss occur?
Reuse: which later decisions treat the record as evidence, and do those uses exceed its original purpose?
Lineage as a Limit
Good lineage does more than prove authenticity. It marks where authority should stop. A field collected for billing may be accurate yet inappropriate for ranking reliability. A classification may be valid for one model version and misleading after the model changes.
Algorithmic witnesses explains how machine records acquire evidentiary power. Lineage supplies the cross-examination path. Pair it with a contradiction ledger when two branches of the record tell incompatible stories.
Field assessment: information should not carry more authority than its lineage can support.