How to put a probability on a UAP case

Public conversation about anomalous phenomena almost always collapses into two boxes: either «they're extraterrestrials» or «there's nothing here.» That binary is comfortable for arguing and terrible for analysis, because it forces you to treat a planet mistaken for a craft and an incident with triple sensor recording exactly the same way. UAP Codex uses a different tool, borrowed from intelligence analysis: instead of a verdict, each case is assigned a probability distribution. Not «what it was,» but «how reasonable belief about what it was should be divided.»

The rule is that each incident case splits exactly 100% of its probability across six narratives that are, at once, mutually exclusive and exhaustive —what analysts call a MECE partition. The six: mundane/natural (misidentification, atmospheric phenomena), classified human (an unacknowledged domestic or allied program), adversarial (another state's technology), covert non-human (something non-human that a state knows of or hides), open non-human (something non-human that no one controls, in the vein of Vallée's hypotheses), and indeterminate. That they sum to 100% and do not overlap is what keeps the exercise honest: you cannot inflate one hypothesis without subtracting from another.

The key step is aggregation. If each case is a distribution, how do you talk about the whole corpus? By summing: the expected number of cases falling under each narrative is the sum, over all cases, of the probability each one assigns to that narrative. Because it is a mathematical expectation —a linear operation— the result is valid and comparable across narratives even if the cases are correlated. The total is divided into 200 pieces and each narrative takes its fraction. That answers questions the binary debate cannot: does the mundane outweigh the indeterminate? How large is the non-human band really, when you look at everything together rather than just the most famous case?

The shape of the data, as of June 2026 over the full corpus, deflates enthusiasts and hard skeptics alike. The aggregate weight leans clearly toward the prosaic: close to half the total falls under mundane or natural explanations. About a quarter remains, honestly, indeterminate —neither resolved nor forced into a conclusion. And the non-human fraction, far from dominating the debate as popular culture suggests, is barely a sixth of the total. The live, exact breakdown, which shifts as the corpus grows and is reclassified, is published transparently on the site's probabilities page.

It is worth stressing what this is not. These are not calibrated frequencies or measurements: they are structured analytical judgments, in the tradition of the intelligence community's ICD-203 standard, which separates the probability of an event from confidence in the evidence behind it. Comparability is not truth: that two cases are comparable on the same scale does not guarantee the scale is right. Document cases (memos, reports, laws) are excluded from the split, because the question «what was the object» does not apply to a piece of paper. And every number is revisable: if a new primary source appears, that case's split changes, and with it the aggregate.

The value of treating the phenomenon this way is not reaching a final answer, but imposing a discipline. It forces you to make uncertainty explicit instead of hiding it behind a headline, not to steal weight from one hypothesis to hand it to another, and to let anyone audit the reasoning case by case. In a field where almost everyone has already decided in advance what they will believe, putting a probability on it —and letting all 200 cases speak together— is a modest but demanding form of honesty.