One decision, two portals
CNIL also publishes decisions on Légifrance. Match number, date and text; an identical decision counts as one record with two access paths.
In-depth practical guides
Count original records rather than links: identify republication, shared data and distinct observations before drawing a conclusion.
Updated :

AI-generated illustration
Domains, logos and formats do not establish independence. Several outlets can reuse a press release, and a registry can feed both a portal and an API. Evaluate one precise claim: documents may have independent commentary while relying on the same figure. The output is a map of origins with remaining uncertainties.
Separate identity, date, amount, attribution and consequence. Retain the supporting passage for each element. One page can be primary for a statement and secondary for an estimate.
Follow links, references, document numbers and producer details. Distinguish author, distributor and index. Stop at an identified document, not a search engine.
Match identifiers, versions, figures, wording and files. Unusual agreement can suggest republication but does not alone prove copying. Seek explicit attribution.
Distinguish the same record, the same collection, new analysis of shared data and a distinct observation. Mark undocumented lineage as unknown rather than inventing a connection.
Try another collection method, event or producer. Differences may come from scope or time; compare definitions before selecting a version.
Describe original records supporting each claim and what agreement establishes. Matching portals can check transcription without providing independent observations.
Typical situations for preparing a check. They do not describe completed assignments or actual observations.
CNIL also publishes decisions on Légifrance. Match number, date and text; an identical decision counts as one record with two access paths.
OpenAlex documents inputs from Crossref, DataCite and PubMed. Match DOI and version; another record may improve discovery without adding a scientific experiment.
Find the original statement. A reporter may contribute an investigation: attribute declared and independently checked facts separately.
Fictional example: D repeats E; E repeats D and A. Reaching A allows inspection of an upstream document, but does not support assertions exchanged by D and E without a matching passage in A. Separate what A establishes from the circular surplus and request support for the latter.
Fictional example: two teams independently calculate a mean from the same file. Agreement may check the calculation; it does not confirm the collection of observations twice. For a collection error, seek another measurement; for a calculation error, retain the file, rules and both processing runs.
Worked exercise
Fictional example: the values and documents teach the method and do not describe a real organisation.
The five publications cover the same subject. E’s stated collection still needs checking against the specific fact sought.
If the table exceeds the screen, scroll horizontally; arrow keys work when the region is focused.
| Publication | Stated origin | Record or role |
|---|---|---|
| A | A | Initial press release |
| B | A | Reproduction without a new record |
| C | A | Reproduction without a new record |
| D | B | Summary |
| E | Another stated collection | Data table |
A, B, C and D form one chain originating in the statement. E is a second declared origin to examine. Five links are not five independent confirmations; E may cover a different scope and fail to confirm the claim.
If E ultimately uses A's data, revise the origin map. A different domain alone does not establish independence.
Keep the reuse graph, original evidence and passage supporting each fact.
Documentation reference : W3C — PROV Model Primer, derivation and revision · Consulted on
Not necessarily for the claim. Check whether they share a press release, witness or dataset.
It is another language version of the same original document. Check the translation separately.
A number alone is insufficient. Use records competent for the question and disclose a single origin’s limits.
Proposed exercise: this protocol is not a performed service test.
Open three references from an answer and identify the producer of the fact used in each.
Keep repetitions of the same data in one origin group, with remaining uncertainties.
For one query, record the answer block and two final pages separately, then find their producers.
Compare origins by result type; coexisting blocks do not prove several independent confirmations.
Repeat a query without a Goggle and with an identified one; compare useful URLs, not only rank.
Retain filter name and rules; a ranking difference does not prove a new document origin.
DOCUMENTED REFERENCE
Six pages citing one another may describe just one observation. For a specific claim, map reuse before counting confirmations. The examples and reading rules below are this site’s teaching method; PROV-O documents provenance relations without supplying an independence score.
| Reference | Value or form | What it identifies |
|---|---|---|
| A: - | A | The − sign marks a declared stopping point. Identify the document and justify that choice: it certifies neither direct observation nor an independent origin. |
| A: ? | ? | The ? sign explicitly preserves an unknown origin. That document and every document leading to it retain an unresolved chain. |
| B: A,C | B → A · B → C | Links run from reuse to its source. A circular group may also reach a known origin; that exit does not remove the loop. |
Work on one claim and specific document versions. An arrow from B to A means here that B repeats A for this claim. Enter only links you have identified; retain the supporting passages in your case file.
Two declared origins do not establish independence. A missing link in the model does not establish an absence of dependency. These counts measure neither truth, reliability nor the number of independent pieces of evidence.
| Document | Reachable declared origins | Unknown origin (?) | Circular groups |
|---|---|---|---|
| A | A | — | — |
| B | A | — | — |
| C | A | — | — |
| D | A | — | — |
| E | A | — | — |
| F | A | — | — |
A circular group contains documents that lead back to each other. The table also flags documents leading into the group. An exit to a known origin does not remove the loop; the signal calls for checking, not a verdict of invalidity. An em dash means no item of that category was found in the model.
Six documents lead to A. F cites two repetitions, D and E, but both paths return to A: more links add no declared origin.
Six documents lead to A or D. The model counts two origins; authors, observations, methods and any unrecorded common source still need checking.
C leads to A and to B, whose origin is unknown. D and E form a loop with an exit to A. Five documents have an unresolved chain even though a known origin can be reached.
A, B and C refer back to each other. None is declared as an origin: by itself, the loop supplies no external observation for the claim being examined.
Calculation covers only the links entered for one claim. It does not detect hidden reuse, check passages or validate PROV graph conformance. Revisit the model when a version, attribution or the claim under examination changes.
Documentation consulted on .
DEEPER READING / PRACTICE
Several articles tell the same fact and seem to confirm it. It is necessary to look at whether they have separate elements or reproduce a common publication.
Updated on .
Compare quotes, formulations, links and timeliness. Returning to the original documents and distinguishing autonomous collection, comment and recall of the same declaration.
A large diffusion does not constitute a large number of confirmations. Different presentation texts may, however, depend on the same statement.
The map of origin connects publications and common origins. The number of families of truly distinct sources is described separately from the total number of links.