You found the source.
But what exactly are you looking at?
A provenance breakpoint is a point in an information lineage where the receiver can no longer reconstruct enough of the acquisition, transformation, selection or context history to know whether the current representation still supports the claim being made.
This is the third pillar beneath How Information Works. The master owns provenance broadly. This page owns the harder question: when does “it came from somewhere reputable” stop being enough because too much changed along the route?
Quick Read
Traceability is necessary but not sufficient. A statistic may originate from an authoritative report, then be rounded, subsetted, reframed, graphed, screenshotted and reposted until the final claim no longer matches what the original source established. Provenance therefore needs transformation history, not just a source name. Strong information systems preserve who observed the original state, how it was measured, what transformations occurred, what filters or selections were applied, which version is current and where interpretation entered. A provenance breakpoint occurs when one of these steps becomes opaque enough that the receiver cannot evaluate whether the final representation remains faithful to the original evidence.
world state → acquisition → source record → transformation → summary → selection → presentation → receiver → claim
Source Name Is Not Full Provenance
“According to the government.”
Which agency?
Which dataset?
Which date?
Which variable?
Which transformation?
Provenance begins with origin but must continue through the route to the current claim.
Copied Claims Can Create False Corroboration
Ten websites repeat the same number.
That looks like ten sources.
All ten may trace back to one article that misread one table.
surface diversity is not source independence.
Every Transformation Can Change the Claim
- rounding;
- normalising;
- filtering;
- averaging;
- translation;
- summarising;
- cropping;
- relabeling;
- changing denominator;
- merging categories.
These may be legitimate.
But they need to remain inspectable when the transformation can change interpretation.
Screenshots Are Provenance-Poor Objects
A screenshot preserves pixels.
It may remove:
- URL;
- date;
- surrounding context;
- interactive filters;
- revision history;
- footnotes;
- denominator;
- source link.
Useful for showing what appeared on screen.
Weak as a complete provenance chain.
A Summary Can Be Accurate and Still Lose the Evidence Boundary
Original:
In this sample, under these conditions, we observed an association.
Summary:
X causes Y.
The source remains traceable.
The claim no longer respects it.
Version Drift Creates Hidden Breakpoints
A page changes after publication.
A quote was taken from version 1.
The link now shows version 3.
Without archival/version information, the current source may no longer demonstrate what the earlier citation claimed.
Lineage Needs Both Data and Interpretation Layers
Data lineage:
where values came from and how they changed.
Interpretation lineage:
where conclusions, labels and framing entered.
Both matter.
AI Can Create Long, Hidden Transformation Chains
An AI answer may combine retrieval, model priors, summaries and transformations.
For consequential claims, move outward toward the primary source or authoritative record and verify that the final wording matches what that source actually supports.
Provenance and Evidence Are Different Jobs
How Evidence Works owns whether information supports a claim relative to alternatives.
Provenance Breakpoints owns whether the receiver can still reconstruct the information route well enough to perform that evidence evaluation honestly.
A Practical Provenance Packet
CLAIM: Average score rose 8 points ORIGIN: School assessment export OBSERVED: 28 Aug 2026 TRANSFORMATIONS: - absent students excluded - marks converted to percentage - class average calculated - rounded to nearest whole number CURRENT PRESENTATION: dashboard tile BREAKPOINT CHECK: denominator and excluded cases visible SOURCE LINK: original export retained
A 20-Lens Provenance Breakpoint Audit
- What world state was originally observed?
- Who observed it?
- How was it measured?
- What source record was created?
- What date/version?
- What transformations occurred?
- What was rounded?
- What was filtered?
- What was excluded?
- What denominator changed?
- What interpretation was added?
- What translation occurred?
- Was the source copied by apparently independent outlets?
- Can the original still be inspected?
- Did a screenshot remove context?
- Did a later version replace the cited one?
- Can the current claim be reconstructed step by step?
- Where does uncertainty increase?
- Where does the lineage become opaque?
- Does the final representation still support the same claim as the origin?
For Primary Readers
If five friends repeat the same story, it may still come from one person. Ask where the story began and whether it changed as it travelled.
For Secondary Readers
Trace one statistic from a social post back to the original table. Record every transformation and identify where the meaning could have changed.
For Advanced Readers
Model provenance as an auditable transformation graph. Reliability degrades when lineage edges are missing, transformations are non-invertible without metadata, source independence is misidentified, or interpretive operations are presented as if they were direct observations.
Final Thought: Traceable Is Only the Beginning
Good provenance does not merely answer “Where did this come from?” It lets us see what happened to the information on the way here.
INFORMATION · FOUR PILLAR LEGS
Return to How Information Works, or continue through Information Half-Life, Missing Information and Decision-Relevance Threshold. Return to the How X Works Hub.