Source Catcher
Every verification tool has failure modes. Most do not publish them. These are ours, with the actual cases.
Source Catcher watches a video and tells you what it was cut from. It is right often enough to be useful and wrong often enough that you should know how. Everything below is a real failure found while building it, not a hypothetical.
Given a football clip, it answered:
Real Madrid vs Fiorentina, Trofeo Santiago Bernabéu, 10 August 2022, played in Vienna — certain
The fixture is real. It was played in 2017, in Madrid. The date and the venue were invented, fluently, with total confidence.
This is the failure that matters most, and it is why nothing reaches you as established unless something outside our own reader agrees. That clip now shows "our best reading — nothing independent confirms it", which is the honest answer.
What to do: read the line under the title. If it does not say confirmed, treat the dates and names as a guess, however confident the wording sounds.
Well-known films, songs, programmes and broadcasters are corroborated easily. A local news story, a minor match, a niche podcast, somebody's own footage — usually not. Reference works only hold what somebody has already written down.
Unconfirmed does not mean wrong. It means we could not find independent support, which is a statement about the reference works as much as about the clip.
Nothing indexes an advert made for a post, or a video somebody shot themselves. There is nothing to check against, so the tool says so rather than reaching for whatever a search happens to return.
A BBC News clip is corroborated by four reference works. All four confirm that BBC News exists. None of them say anything about whether the report you are watching is real, or whether the footage in it shows what it claims.
The page says this explicitly for anything that runs continuously — a news programme, a series, a podcast. A film is different: confirming the film does confirm where the clip came from.
There is no watermark being read and no forensic analysis. It is a careful look at the footage — hands, faces, physics, lighting, text — and it says "not certain" when it is not certain. Treat it as a well-informed second opinion, not proof either way.
Platforms strip content credentials on upload, so the metadata route that would settle this is almost never available.
The archive can only report a clip somebody has saved before, so anything recent comes back with nothing. We fixed a bug where the tool archived a clip itself and then reported its own save as evidence of the footage's age — so a snapshot from the last fortnight is now ignored entirely.
Where the platform gives a posting date, that is shown. The picture is also read for clues — clothing, weather, signage, vehicles, phone models — which is a hint, not a date.
A large study found 84% of false claims return nothing from the main public fact-check database. Silence there is not a clean bill of health, and the tool does not present it as one.
YouTube stills can be traced properly. TikTok and Instagram serve scrambled addresses that image search can only read generically — it came back with "poster" and a list of poster-design videos on one occasion. Rather than show that as a finding, those platforms show nothing.
Two or more independent reference works hold a matching record. Solid.
Likely right, not established. Worth a second look if it matters.
Our reading, unsupported. Often correct — but the dates and names are the first thing to doubt.
The work is real and something about when does not fit. Look harder.
We could not work out what it came from. The reverse-image links are the next step.
Genuinely useful, and the fastest way this improves. Send the link and what it got wrong to hello@everythingthreads.com. Several of the failures on this page were found that way — by looking at an answer and asking what it was actually measuring.