03How it works
A signal is added and lives alongside your own intelligence, marked as external so its provenance is never confused with what your organization knows first-hand. Open a signal's source and mark it verified, and it records who checked it and when. Nothing automatic ever marks one verified.
Getting into the work is a deliberate climb, in three rungs. Starring one is private to you, and starring is on purpose NOT how something becomes citable: if a private bookmark made a source foundational, everybody's bookmarks would end up grounding everybody's work. Putting one forward, and then sharing it across the organization, are separate steps a person takes.
A signal you reject is kept, never deleted, and that is the part worth understanding. Refusing to hand a discredited claim to the writer is only half the job, because a figure repeated across the internet for twenty years is already sitting in the model's training data whether you supply it or not. So the rejected ones become a register, and every composed Expression is checked against it before review. The check reports; it does not block. It cannot tell a claim being used from a claim being refused, and a piece that says out loud "this statistic has no traceable source and we do not use it" would be stopped by a check that could not read intent.