Memory does not store an event and replay it. Each recollection is assembled from fragments, and the assembly incorporates whatever has been learned since, which is why a witness can describe with complete sincerity a detail that was supplied to them after the fact. Experiments demonstrating this are among the most replicated in psychology. The forensic consequence is awkward: confidence and accuracy correlate reasonably well at the moment of a first identification and diverge steadily afterwards, as a witness who has been told they picked the suspect, questioned repeatedly and prepared for trial becomes progressively more certain. The witness a jury sees is therefore the least reliable version of the witness the investigation ever had.