Probabilities · comparable model
What the corpus explains
The corpus's 387 cases are each classified into one narrative: the 286 incidents by the nature of the object and the 101 document cases by their content's lean. The prosaic opens into three narratives of its own —misidentification, natural phenomenon and possible hoax— and 'non-human' groups covert + open. Whatever cannot be decided —inconclusive incidents, documents with no clear direction— falls into 'Indeterminate'. Summed, they partition the corpus comparably: one can say which narrative accounts for more cases.
The corpus hypotheses
The center marks the corpus total (387): 286 incidents by their object + 101 document cases by their lean, and 'Indeterminate' for what cannot be decided. Forced, navigable classification.
Sum to 100% · 387 corpus cases · mundane opened into 3 · non-human grouped · Indeterminate separate
Heterogeneity over time
The donut above is the aggregate snapshot. This series adds the time axis: how much of each decade resists a mundane explanation. The 2020s uptick coincides with the disclosure cycle.
Heterogeneity = 1 − the mean share of the mundane/natural narrative, by decade (incidents with a MECE posterior). It measures how much of the corpus resists a conventional explanation. The donut gives the aggregate snapshot; this adds the time axis. Hover each point for the decade's n.
The hypotheses, one by one
What each means and which prior-framework hypothesis it preserves. Each block links to the list of cases where it is the most probable explanation, pre-filtered.
Misidentification
218 casesMisidentification of a known object (aircraft, balloon, satellite, planet, drone) or perceptual error / illusion. Human error about something ordinary.
Mistaken for what? · 172 incidents
- No single object identified106 · 62%
- Aircraft32 · 19%
- Astronomical20 · 12%
- Space (satellite/reentry)14 · 8%
Incidents only; misid document-cases are not sub-classified. 'No single object' = the analysis leans prosaic but pins no specific object (diffuse light, lighthouse, or an undetermined mundane explanation).
Indeterminate
47 casesIndeterminate — the evidence leans toward no narrative: inconclusive incidents and document cases whose content points in no clear direction. It is not an explanation but the honest absence of one.
See the 47 casesNon-human
28 casesNon-human intelligence or technology — whether a state controls or hides it (reverse-engineering, treaty) or no one controls it (Vallée-style, interdimensional / ontological).
See the 28 casesClassified human technology
14 casesA secret own or allied program (cover-up is intrinsic). Former 'classified programs' hypothesis.
See the 14 casesNatural phenomenon
44 casesA genuine, poorly-understood natural phenomenon: atmospheric plasma, ball lightning, bolide / meteor, atmospheric optics. Real physics, not a craft or a hoax.
See the 44 casesPossible hoax
26 casesPossible deliberate deception: staging, fabrication or hoax. The classification flags the most plausible candidate, not a closed verdict.
See the 26 casesAdversary technology
10 casesAnother state's surveillance technology. Former 'adversary technology' hypothesis.
See the 10 casesModel honesty
Per-case posteriors are structured analytical judgments, not empirically calibrated frequencies: comparability is not the same as truth. The model says which explanation is most coherent with each case's analysis, not which is objectively correct. There are two ways to aggregate the corpus, and both cover the same cases —incidents by their object, document cases by their content's lean. The 'expected number of cases per explanation' (the sum of each narrative's probabilities) is linear and comparable, holding even if cases are correlated, and it spreads the 'indeterminable' mass fractionally: it is shown on /calidad. The modal-hypothesis count assigns each case —forced classification— to its single most probable narrative, keeping 'Indeterminate' as its own narrative for what cannot be decided; the chart above and the links to the cases use the latter: integer, navigable counts, consistent with the '/cases' filter. The two views give slightly different figures because they are different estimators (expected vs modal), not because they contradict each other. Each case's full distribution —with its uncertainty mass spread across the narratives it supports— lives in the case detail.
Two independent axes, easy to confuse: the «tier» (S/A/B) measures the strength of the evidence —how hard the case is to dismiss— while this partition of explanations measures what it was. They are not the same: a well-documented case (Tier S or A) can have a possible hoax as its most plausible explanation, and a case with limited evidence (Tier B) is not, for that reason, a hoax. In fact, the cases classified as possible hoax split evenly between Tier A and Tier B.