Trust
Confidence, not verdicts.
Witness does not tell you what is true. It shows how confident the system is, what evidence moved it, and what remains unresolved. That distinction is the whole design.
Witness does not tell you what is true. It shows how confident the system is, what evidence moved it, and what remains unresolved. That distinction is the whole design.
How much the evidence supports this specific claim.
How reliable a contributor has been over time, scored independently of any single report.
How much independent evidence supports the underlying event across every report attached to it.
Keeping these apart is what stops a trusted account from carrying a false claim, and what stops one contested report from destroying a good contributor.
Signals shift confidence according to how much they actually distinguish a real event from a fabricated one. A signal that is absent counts as nothing, not as evidence against.
Two people on opposite sides of a square who do not know each other are worth far more than a cluster sharing one source.
A report can be confidently placed while a specific assertion inside it stays unresolved, so a real event cannot be used as cover for a false detail.
A first report is treated as unknown rather than presumed bad.
The system stops promoting a contested claim rather than removing it. Nothing is silently deleted, and any removal carries a visible reason.
Falsely burying a legitimate report is the worse failure. Ambiguous cases resolve in that direction.
Moderation exists for harassment, illegal content, and abuse. It does not decide what happened. Nobody at Witness can mark a report true or false.
Exact coordinates are never published.
Location and network signals are never used to penalize a contributor.
No facial recognition, under any circumstances.
Buyers and platform partners receive events, never identities.
Reporters, newsrooms, and data teams can request access.