Sources disagree
Anyone who has worked a hard case has met this. One database calls a variant likely benign and another has a report of it in an affected family. One paper links a gene to a condition and a later one finds the evidence thin. A phenotype in the clinic note does not match the one in the referral.
A tool has to do something with that. The easy things are to average the sources, to keep the newest, or to keep the one it trusts most and drop the rest. All three make the case look cleaner, and all three throw away the part a reviewer would most want to see.
Both stay
In the case record, conflicting evidence is kept side by side. Each item says where it came from, which release or version it was read from and the record it refers to. When two items disagree, the record shows that they disagree. It does not resolve the conflict by overwriting one with the other.
Evidence is not edited after it is stored. If a source changes, the new version is added next to the old one, and the case shows that something changed and when. That is what allows someone to open a case after a year and see what was known at the time of each decision.
Why it matters in review
A case review is usually an argument about weight. One person thinks the population frequency argues strongly against a variant, and another thinks the family segregation argues for it. A record that has already chosen a side takes the argument away from the people who should be having it.
A record that shows the conflict, with the sources, lets the argument happen on the evidence. It is also easier on the person who has to sign at the end, because they can see what they are signing over.
Nothing without a source
The other half of this is that nothing enters the record without saying where it came from. A statement with no source, release and record behind it is not stored as evidence. It is much less satisfying than a smooth summary, and it is the reason the summary can be trusted. Where an item was read automatically from notes, it is marked as such until a person has curated it.
An example from a review meeting
A variant in a candidate gene is reported in a public database as seen in several unrelated unaffected adults. A paper from a different group describes it in one affected child, de novo. The two statements cannot both be read at face value.
A record that keeps both lets the meeting do what meetings are for. Someone points out that the adults in the database were not examined for the feature in question. Someone else notes that the de novo claim rests on one family. The group decides how much each is worth and writes that down. The two items remain, with the reasoning next to them, for whoever reopens the case.