Behavior
150 classified events · observed conduct, not appearanceAn axis orthogonal to shape and sensor: what the object did, classified from the report text on a scale from passive (transit, hover) through maneuvering to observer-relative (approaching/pacing the observer, or changing behavior when detected). This is the dimension on which a hypothesis of organized activity would have to register. Constraint: these are passive observations, not controlled tests — they can show whether behavior is conditioned on the observer, but cannot establish intent. Classification is judgment-based from the text; treat it as hypothesis-generating.
Most classifiable events are passive transits or maneuvers. 15 of 150 are observer-relative — the object approaches/paces the observer, or its behavior changes on detection.
Observer-relative events
15 cases · the interaction-suggestive subsetEvents where the object’s conduct was tied to the observer. Note the composition: these are largely the well-known close-encounter cases, which raises a confound — dramatic, interactive accounts are more likely to be retained and declassified, so this subset may reflect retention bias as much as a real concentration of interaction. Red = behavior changed in response to detection/illumination/approach.
Recurring combinations vs. chance
do attribute motifs repeat above random?Combinations meeting pairs with support ≥ 6 and lift ≥ 3: 4 observed. Shuffling each attribute independently (300 runs) produces a mean of 0.09 (95th pct 1, max 2).
The corpus is therefore not random — attribute combinations recur well above chance. What recurs, though, is mostly the reporting/era structure (see right): the test detects association, not intent.
Most are era/reporting-linked (silver-disc, NASA-spacecraft) or known shape–colour pairings. The two non-era families worth noting: multi-colour events (objects reported in multiple colours together — e.g. orange + red, the orb-emitting-orbs signature) and green + orb (the nuclear-site green-fireball cluster). These are the candidates for structure beyond era — flagged for scrutiny, not asserted as meaningful.
Method note: behavior is classified by reading each record’s summary, verbatim claim, and movement descriptors. It is independent of object shape and sensor. The corpus is small and biased; this view is for hypothesis-generation, not inference of cause or intent.
Negative space — what the record does not contain
constraining intent by absenceWe do not know why these objects are present, what they are doing, or where they are going. One way to narrow the question is to ask what consistently fails to happen across 212 records. The answers are strikingly uniform.
Where a term appears at all (e.g. offensive language in 8 records) it is almost always a near-miss, an equipment effect, or an explicit non-event — not the object acting.
- Transient passage. Median encounter 60s; 70% under five minutes; 1 over an hour. They move through.
- Evasion / departure. 33 records describe the object accelerating, climbing, or leaving — it breaks contact rather than escalating.
- Mostly non-interactive. Only 15 of the classified events are observer-relative at all.
The recorded profile is brief, evasive, non-acquisitive and non-hostile — consistent with a transit or reconnaissance function, and inconsistent with invasion, contact, occupation, or resource extraction. “Probe-like” is the hypothesis this negative space best fits.
Why this is weak evidence, deliberately stated: Negative evidence is the weakest evidence: this is the RECORDED, DECLASSIFIED set, filtered by what was observed, survived, kept, and released. Absence here is not absence in reality (e.g. crash-retrieval claims exist publicly but not in this corpus). Keyword presence ≠ the event occurred. 'Function consistent with' is not 'intent'. A consistent function is not intent — a brief, non-hostile transit profile is what we would record whether the cause were craft, instrument artifact, or something mundane seen briefly. The value here is in narrowing which hypotheses survive, not in selecting one.