A page says that three other pages have reported the story. One of those pages links back to the first. Another uses the same image with the corner cut off. The third says that the allegation is all over the internet.
There may be three pages. There may be only one source.
We often use the word source for things that do different jobs. A witness is a source. So is a recording. So is a newspaper reporting what a witness said about a recording. A link can take us to any of these. If we count links without asking what sits behind them, repetition can look like confirmation.
This matters most when a person’s name is already attached to the story. Ten search results can feel harder to challenge than one. The task looks enormous: answer ten accounts, explain ten statements, overcome ten apparent witnesses. A careful source map may reveal a smaller evidential problem. Several reports may stand or fall together because they depend on the same mistaken identification.
The number of people repeating an accusation tells us something about its distribution. It does not tell us, on its own, how many people observed the event.
Follow the arrows
Consider an invented dispute after a neighborhood meeting. We have two underlying records. One is a short excerpt from a recording, which we call O1. The other is an attendee’s note, O2. Ten reports circulate. For the moment we call them R1 to R10, so that names and logos cannot lend them authority they have not earned.
R1 uses the recording excerpt. R2 copies R1, and R3 copies R2. R4 also relies on R1. A second line begins with R5, which uses the attendee’s note. R6 follows R5; R7 follows R6. R8 draws on both R4 and R6. R9 follows R8, and R10 follows R9.
That is the complete map. The model is hypothetical. We have stipulated these routes rather than discovered them through fieldwork. Its advantage is that nothing is hidden: we can examine every connection and see what our counts mean.
The ten reports depend on two underlying records. R8 combines the two lines, but combining them does not produce a third observation. R10 is several steps from both records. Its distance does not automatically make it wrong. It does make the basis of its confidence worth checking.
Suppose R10 says that multiple accounts confirm the event. The phrase might be literally true if accounts means web pages. A reader could reasonably hear something stronger: several independent people saw what happened. The wording should make clear which claim is supported. An honest report can say that two source records have been repeated across several outlets and that their relationship is still being checked.
Even our two records cannot be treated as independent by default. The attendee may have written the note after watching the excerpt. The recording may capture the attendee reading from an earlier account. Both may depend on a third thing we have not observed. The graph tells us where the known paths lead. It cannot certify what we have not checked.
One good source could establish more than ten weak ones. Two independent sources could disagree. A useful investigation therefore asks two questions: how many distinct bases of evidence do we have, and how good is each one for this particular claim? The map answers neither question completely, but it helps prevent a basic counting mistake.
What the source can actually show
A recording can show that certain sounds and images were captured in a file. It may still leave questions about the speaker’s identity, the date, a missing section, or the meaning of the surrounding exchange. An attendee’s note may supply context the excerpt lacks. It may also contain an honest error. We should examine the strengths and limits of each instead of arranging source types in a permanent ladder.
For example, the note says that a named person spoke during the meeting. The video excerpt contains a voice but no view of the speaker. These items might agree about the event without independently establishing the identity. If the name in the first report came only from the note, every later report that names the speaker may still depend on that one identification.
Now suppose the full recording shows the person was quoting an argument they opposed. The words were spoken, but the widely shared description of their view was wrong. Correcting the claim requires restoring the context. A forensic test of whether the excerpt was digitally altered would not be enough. The excerpt could be an accurate piece of an inaccurate account.
Or the full recording shows a different person speaking. Then the correction concerns identity. It need not announce that the entire meeting was invented. It should replace the wrong attachment with the supported one, while protecting unrelated people from needless exposure.
These distinctions reduce the size of the argument. We can stop fighting over whether the whole internet is trustworthy and ask whether a specific piece of evidence supports a specific sentence. This is work that ordinary readers can begin, even if some technical questions later require an expert.
There is a trap here for someone trying to defend a friend. Finding a weak source does not prove the friend is right. It tells us that this source cannot bear the weight placed on it. Another source may exist. Conversely, an accusation does not become well supported because its subject has behaved badly in an unrelated matter. Evidence has to meet the claim at hand.
Independence has to be earned
Two people can honestly believe they saw the same thing because both saw the same misleading clip. Their honesty does not make their accounts independent evidence of the event. It makes them sincere witnesses to what they were shown.
The difference is worth preserving. If someone says they saw a recording, do not promote that into having witnessed the event itself. If they say a colleague told them, keep the colleague in the account. Each missing step makes the story seem closer to direct knowledge than it is.
A simple question often helps: What did you personally observe, and what did you learn from someone else? Asked calmly, it gives a well-intentioned person room to narrow their claim without being humiliated. They can say, I saw the post, not the meeting. That is a correction in the direction of truth.
We ought to make such narrowing easier. If every change is treated as an admission of wickedness, people have a reason to defend claims beyond their evidence. We can hold them responsible for an error while welcoming the act that helps repair it. Responsibility and an exit from error belong together.
An institution can assist by retaining the source distinctions in its own records. A complaint form that records only “reported by several people” loses important information. Several direct accounts differ from several forwards of one screenshot. The people involved need not be named in a public notice, but someone reviewing the decision should know the difference.
Protection for confidential sources complicates this work. A source may have a sound reason not to be publicly identified. Confidentiality does not make an account false. Nor does it permit a publisher to pretend that they have checked more than they have. A responsible editor can explain the basis of verification without exposing every private detail. The public may have to judge partly through the editor’s accountable process.
There is no universal solution in a diagram. A diagram is a way to ask better questions, not a device for forcing vulnerable people to publish their identities.
The first available copy
Tess finds an older version of the image. It has a different crop and no business name. Is that the original?
It is the earliest copy she has found. Those words should remain in her notes. Search results, timestamps and accessible archives may not reveal every earlier version. A private group could have carried the image before a public page did. A post can be edited after its initial date. A screenshot can preserve a clock without preserving what that clock measures.
Calling the earliest found copy the origin can send a correction request to the wrong person. It can also create a fresh allegation: you made this. The evidence may support only: you shared this version at this time.
There is enough responsibility in the supported claim. Someone who spread an unsupported attribution can correct their own post even if they did not invent it. We do not have to prove authorship of the whole chain before asking each participant to repair their part.
This is a useful release from a common dead end. The origin may remain unknown. A correction can still reach a customer, an editor or a hiring panel. The missing first link should be recorded, but it should not stop action at the links we can see.
At the same time, the search should remain open to a changed conclusion. If a reliable original source later appears, the analysis may need to change. A correction that states its evidence clearly can be updated without pretending that earlier uncertainty never existed.
Use the map for repair
Return to our invented ten-report graph. Assume corrections have been delivered through R1, R5, R8 and R10. That is four of the ten report nodes, or 40 percent.
It is tempting to say that 40 percent of the damage has been repaired. We know no such thing. We have counted nodes, not people. R1 may have many readers and R5 only a few. Readers may overlap. A correction attached to one report may be visible only to people who revisit it. Some reports may have been downloaded or quoted elsewhere.
The number is still useful if we give it the right name: report nodes with correction delivery under the model’s assumptions. It tells the person managing the response which known reports have been addressed. It does not describe belief, income or fairness.
A map can also show where more work could have wider effects. If a report is a common source for many copies, its correction may help downstream publishers understand what changed. That is a possible route, not an automatic cascade. The people operating those later pages must receive the update and decide what to do with it. An arrow describing past copying is not a promise of future cooperation.
We should also notice where a single source has been turned into an apparent consensus. Those points deserve an explicit explanation. “The five reports we cited all used the same misidentified photograph” is more informative than “new information has come to light.” It tells a reader why the earlier confidence was misplaced.
The correction should not create a new fog of sources. If an outlet verified only the identity in a photograph, it should say so. If it also checked the date and the full recording, it can say that too. Each verified step helps the reader assess the revised account.
A source map may look like a technical aid. At its center is a plain social duty: do not make one person’s mistake look like ten people’s knowledge. And when you discover that this has happened, tell the people who trusted the larger number.
Tess may never identify every copy of the picture. She can still ask a clear question of each person she reaches. What did you use as your source? From that answer, the next step becomes smaller, more exact and more fair.
Source notes
The graph, node labels, routes and counts in this chapter are original hypothetical constructions. See the methods appendix and the supplied analysis files. No empirical dataset, independent-witness count or estimate of actual audience size is claimed.