Two investigations can gather identical material and produce opposite reports. The difference is rarely access. It is what each investigator decided to believe, and whether that decision was ever written down.
Three buckets, not one
Everything an investigation touches belongs in one of three places:
Accepted. Supported well enough to carry a conclusion. It entered the case with a reason.
Uncertain. Plausible, unresolved. Still a lead. It must not quietly become a fact because it was mentioned three times in a meeting.
Rejected. Ruled out — wrong person, unreliable source, contradicted, undatable. Rejected is a result, not a deletion.
Most failures come from the boundary between the first two. Under deadline pressure, "probably" is written as "is", and by the time the report is read the hedge has vanished.
Why rejected material must stay visible
An investigation that shows only what it accepted is impossible to audit. A reader cannot tell whether contradicting evidence was weighed and dismissed for good reason, or never encountered at all.
Keeping what was ruled out — and why — does two things. It proves the work was adversarial rather than confirmatory, and it saves the next analyst from re-running a dead end.
The questions that decide admission
- Provenance. Where exactly did this come from? Not the aggregator — the origin.
- Independence. Do my two sources actually derive from the same one?
- Identity. Is this about my subject, or about someone with the same name?
- Date. When was this true, and is it still?
- Falsifiability. What would have to exist for this to be wrong, and did I look for it?
Write uncertainty as uncertainty
The strongest reports are explicit about what they do not know. "No evidence was found that X" is a different statement from "X did not happen", and a reader making a decision needs to know which one you are making.
Confidence that is not supported is the fastest way to lose a case in review — and the easiest thing to avoid.
How software should help
Not by deciding. By making the distinction structural: accepted findings kept apart from uncertain and rejected material, each carrying its source and the decision behind it, so the report is assembled from what passed review rather than from memory.
That is how the Archive and Engramite Writer are built.
Frequently asked questions
Is this not just source grading? Source grading rates the source. Admission decides whether a specific claim enters this case, which is a different question.
Who decides? A person. Software can propose, organise and remember; the judgement belongs to the investigator.
What if new material changes an earlier decision? Revisit it and record the change. A case file that shows its revisions is stronger, not weaker.
Read next: the homonym problem or what is OSINT.