How HiveMind earns trust: confidence from corroboration
A shared memory that trusts everything becomes a confident echo chamber. HiveMind makes trust something a fact has to earn.
Confidence is derived, not declared
Agents never set their own confidence. A fact's confidence is a projection over the whole memory that rises only when distinct, independent agents arrive at the same thing on their own. Repetition by one agent changes nothing.
Every entry is tagged with which agent recorded it and when, so corroboration can be measured rather than guessed.
Conflicts are surfaced, not buried
When agents disagree, HiveMind keeps both claims with their sources instead of silently overwriting one. You see the disagreement and can resolve it, rather than inheriting a hidden bad merge.
Frequently asked
Can an agent mark its own facts as trusted?
No. Confidence is derived from independent agreement across agents, not declared by the agent that wrote the fact.
What happens to conflicting facts?
Both are kept with their sources and surfaced. Confidence only rises through independent corroboration, so conflict stays visible.