Method
The rules of the game
Any UAP evidence collection needs an answer to three questions: what counts, how it's weighted, and when it changes. Here are all three, made explicit — so anyone can audit the analysis instead of having to believe it.
Chapter 1
Why Roswell isn't equivalent to Meier
3 evidence tiers (S/A/B) — Tier S (military+sensor) weighs differently than Tier B (single-witness); and the tier is not a verdict
Not all UAP cases are equivalent in evidential value. Mixing tiers obscures the analysis and artificially inflates the appearance of evidence. The public and the press frequently conflate tiers — that's a principal source of confusion.
| Tier | Category | Examples |
|---|---|---|
| S | Strong evidence: instrumental sensor + multiple witnesses (typically military) | Tehran 1976, Nimitz, Belgian Wave, Lake Huron |
| A | Institutional evidence: multiple verifiable witnesses or official documentation | Ariel School, JAL 1628, Manises, Westall, Roswell |
| B | Limited evidence: single-witness, local or without primary verification | Hessdalen, Communion, Maury Island, Bonnybridge |
A Tier B "evidence" should not be used to support conclusions that require Tier S. The scale (descended from Hynek's Close Encounter Categories) measures only the strength of the evidence — not what the case was.
Three independent axes — don't conflate
Each case is described with three distinct measures, spread across the site's views. They are easy to conflate because all three sound like «quality»:
- Tier (S/A/B) — the strength of the evidence. Shown as a badge on each case and as the marker color on the map.
- Probability (0–100%) — how genuinely unexplained the case is. A natural phenomenon can remain unexplained, so it does NOT equal «non-prosaic». Shown as «%» on each case and as the marker size on the map.
- MECE partition — what it was: the distribution over six mutually exclusive narratives. see /probabilidades →
They are orthogonal: a well-documented case (Tier S/A) can have a possible hoax as its most plausible explanation, and a Tier B is not, for that reason, a hoax.
Chapter 2
Why case number 50 adds nothing new
A sensor case yields a sharp posterior that moves mass between narratives; case 50 of the same pattern only reinforces the one already dominant
Each case adds its full posterior to the partition —the aggregate is linear, with no hidden 'diminishing returns'—. What changes is how informative that posterior is: an institutional case with a sensor yields a sharp posterior that can move mass toward a previously near-empty narrative; case number fifty of the same pattern only reinforces the narrative that was already dominant, teaching us nothing new about the shape of the partition.
Moves mass between narratives
- ·Tier S/A with multi-sensor → sharp posterior (Tehran, Nimitz)
- ·New class of evidence (Hessdalen, Lake Huron)
- ·Case that contradicts the established pattern
- ·Official sensors + correlated video
Only reinforces the dominant
- ·Isolated Tier B contactee → mostly mundane/indet (Meier)
- ·Case #50 of the same pattern (scales the dominant narrative)
- ·Failed contactee prediction (loads mundano_natural)
Status: isolated "contactee" cases contribute posteriors dominated by mundano_natural and indeterminable — they do not redistribute the partition. Institutional sensor cases (Tier S/A) yield sharp posteriors that do move mass between narratives.
Chapter 3
Why probabilities sum to 100%
The six narratives are mutually exclusive per case; the distribution is comparable and sums to 100%
Per case, the narratives are mutually exclusive: each case had one real cause and the uncertainty is split among the candidates, summing to 100%. Summed across the corpus they give a comparable partition — one can say which explanation accounts for more cases. The site's presentation applies a forced classification: the prosaic opens into three hypotheses (misidentification, natural phenomenon, possible hoax), the two non-human ones are shown together, and the mass a case cannot assign is spread across the hypotheses it does support — so no case is left unclassified. The prior framework, where percentages neither summed to 100 nor were comparable, was replaced precisely to fix that.
The per-explanation detail (with numbers and modal cases) lives in /probabilidades → to avoid duplicating content here.
Chapter 4
What is settled vs the frontier
Most of the corpus is prosaic (mostly misidentification); the non-human hypotheses are a minority and concentrate in the best-documented cases.
The model was reformulated to yield COMPARABLE numbers. Before, each hypothesis was an independent existential claim (at least one case is X) and its percentages neither summed to 100 nor competed — one could not say which explanation was more probable than another. Now each case splits 100% over the same six mutually-exclusive narratives, and the corpus aggregates them into a partition that splits 100% comparably. Each narrative bundles object + institutional stance, so 'non-human + state cover-up' is its own class. The prior framework hypotheses are preserved as a mapping inside each narrative, and 'non-human entities' (the sum of the two non-human narratives) and 'heterogeneity' remain derived views.
The full partition, the six narratives and each one's modal cases live at /probabilidades →.
Chapter 5
What would move the analysis now
Lake Huron isotopic analysis, verifiable recovered material, new country acknowledging
- +++Independent isotopic analysis of physical residues published
- +++Recovered material with verifiable photos
- +++Lake Huron fragment analysis published
- +New country formally acknowledging
- +Another Tehran-equivalent with sensor data
- ~0New contactee with detailed cosmology
Per-case posterior, mechanical aggregate
How each new case moves the probabilities automatically
Corpus probabilities are derived at build-time in a public, comparable way. Each case splits 100% among six mutually-exclusive narratives: a posterior that sums to 1. The corpus aggregates them into the expected number of cases per narrative —Eⱼ = Σᵢ P(narrativeⱼ | caseᵢ)—, which splits 100% and is comparable across narratives; being an expectation (linear), it holds even if cases are correlated. These are structured analytical judgments, not empirically calibrated frequencies: the model says which explanation is most coherent with each case, not which is objectively correct. Every new case recomputes the partition without human intervention.
Each case's posterior is a declared judgment, not a computation from weights: the case analysis explicitly splits 100% by how well each explanation accounts for the evidence. There is no hidden formula — the split is the judgment, and it stays visible on the case page. The mass a case cannot assign (the former 'indeterminable') is redistributed, in the charts, across the hypotheses the case does support: forced classification, no case left unclassified.
The corpus aggregate is purely mechanical: it sums those posteriors into the expected number of cases per narrative and rounds them with the largest-remainder (Hamilton) method so the displayed percentages sum to exactly 100. Each new case enters with its posterior and recomputes the partition with no further intervention.
Each case shows its posterior on its own page, in the "Distribution of explanations" section. For incidents the posterior measures the nature of the object; for document cases (memos, hearings, leaks) it measures the evidential 'lean' —which explanation their content tilts toward—, since they have no object to classify.
You now understand the method
Now see the result: six narratives partitioning the corpus
The full methodology lives on this page — no external documents or technical dependencies. Auditable directly from here.