Two hypotheses fit the strangest patterns in the released record better than their rivals. Neither is a finding. This page states each one formally, lists what it predicts, checks every prediction against named documents, presents the evidence against, and says what would falsify it. The documents are one click away throughout; read them and disagree.
Any theory of this corpus has to account for two patterns that survive every filtering pass we have run on the 216 documents.
The 1946–59 files are overwhelmingly discs: the FBI headquarters saucer file alone spans ten sections of them, and the Air Force’s own 1955 statistical study categorized thousands of reports in shape classes that barely exist in the modern record. The 2020s record is dominated by spheres and orbs on infrared. Underneath the drift, the luminous-orb family appears in every era: the wartime night-phenomena file, the 1949–50 green fireballs over the nuclear labs, and the modern FLIR record. The full data is charted on Morphology.
Across independent witnesses, sensors, and decades, the same behaviors recur: parent objects releasing or absorbing smaller ones; objects that approach or orient on observers; objects that vanish from tracking at the moment of characterization; objects visible to one sensor band and invisible to others; and a clustering on nuclear and defense infrastructure that includes the Soviet side of the Cold War. The catalog is documented behavior by behavior on Behaviors.
The phenomenon co-varies with its observers: what is reported tracks the sensing technology, expectations, and conceptual categories of the era doing the reporting, in a way that outruns simple reporting bias.
Intellectual lineage: Jacques Vallée’s control-system conjecture, which predates every modern release. The corpus’s relationship to Vallée’s specific claims is examined on The Theorists; the books themselves are outside the release and are treated as context, not evidence.
Visual-report era: discs, in the thousands, formally categorized by the Air Force’s own statisticians. Naval radar-FLIR era: tic-tacs and cylinders. Drone-IR era: spheres and orbs. The drift is the cleanest large-scale pattern in the corpus, and it tracks the instrument as much as the decade.
The D25 diamond was detected on shortwave infrared only: invisible to the eye and to standard IR on the same aircraft. The cold-IR objects read cooler than ambient air while flying. The East China Sea object registered as a lens-flare-like signature across two different sensors. The record contains no object that was ever pinned down consistently across bands.
AARO’s own analysis of its best-documented modern case leaves roughly forty percent of the reported phenomena unresolved. The IC’s Cheyenne Mountain assessment reached only “possible backscatter.” But “mixed” is the honest grade: the record also contains cases the system did resolve toward the mundane, which a strong Mirror reading would not expect the phenomenon to permit.
Sensor rendering explains much of the drift for free. Any distant point-source on infrared renders as a round blob. An era whose primary sensor is FLIR will produce “orbs” from a heterogeneous population of objects without any co-variance at all. This rival explains the 2020s orb dominance more cheaply than the Mirror does.
But note what the rival cannot explain: the same decade’s civilian eyewitnesses, filming on phones in the visual band, also report orbs, with FBI-assessed credibility. Two independent channels converging on the same form weakens the pure-artifact reading; it does not rescue the Mirror specifically.
The unfalsifiability trap. A theory in which the phenomenon adapts to observation can absorb any observation. Stated carelessly, the Mirror is unkillable, which makes it useless. Hence the falsifier below.
A stable, repeatable, multi-band characterization of a single object (visual + IR + SWIR + radar agreeing on form, size, and emission), or a future tranche showing shape distributions that fail to track newly fielded sensor modalities. The D25 case is already uncomfortable for the Mirror: an object visible only to the newest band looks less like a reflection of the observer and more like a physical emission profile.
The core phenomenon behaves like distributed, autonomous, observation-oriented reconnaissance of strategic infrastructure: patient, repetitive, indifferent to being seen, and resistant to being measured.
The 1949–50 green fireballs over the Sandia–Los Alamos complex. A modern unidentified-object incident report from the Pantex nuclear weapons plant. The 2022 Cheyenne Mountain object. A September 2023 sighting at a restricted test facility. The Western U.S. event beside a sensitive national-security site. And the check that matters: CIA reporting places the same phenomenon over the Soviet Union’s own weapons-testing range, which no theory of purely American secret technology survives. Tranche 4 hardens both ends of this line: the verbatim 1949 Los Alamos conference transcript and the full 2015 Pantex incident report are now primary sources rather than summaries.
Federal agents watched orange orbs release smaller red orbs on two separate days. A gunship crew recorded a possible sub-object detachment over the East China Sea. Civilian witnesses filmed two orbs moving “as though tethered” that appeared to merge. Unrelated observers, unrelated sensors, same grammar.
A senior intelligence officer’s helicopter was approached within an estimated ten feet; the object then split in two. Objects hold co-altitude with aircraft for minutes, then leave. The same record shows disappearance at the moment of tracking, single-band visibility, and not one recovered artifact. Things that hide do not approach helicopters. Things that pose do not stay unmeasurable for eighty years.
The New Mexico nuclear complex appears in 1949–50 and again in the modern DOE record. The Northeastern orb cases recur across 2021–2026 within a 25-mile radius. But the corpus is a curated sample of what agencies chose to release; revisit-rates computed from it inherit that curation. Graded mixed until a fuller release allows real base rates.
The selection effect is real and large. Sensitive sites are exactly where trained observers, high-end sensors, and mandatory reporting live. A phenomenon distributed uniformly over the Earth would still appear to cluster on the nuclear complex in a corpus assembled from military and FBI channels. The corpus cannot compute a base rate, and honesty requires saying the strongest Survey evidence sits on top of the strongest sampling bias.
The system sometimes resolves things. D7’s weapons-quality track ended in “looks like a balloon.” The ICA tried “backscatter” for Cheyenne Mountain. A survey hypothesis built only on the residue must acknowledge how much of the record is not residue.
No artifact, ever. Eighty years of alleged reconnaissance and the release contains no material, no isotope ratio, no wreckage claim beyond testimony. Absence of evidence is weak evidence of absence, but it is not nothing.
A future tranche with honest geographic base rates showing no site affinity once reporting density is controlled; or demonstration that the parent-child cases reduce to sensor artifacts and misidentified munitions; or a resolved mundane explanation for the Western U.S. event, which is currently the hypothesis’s load-bearing case.
Neither hypothesis needs a conspiracy, and the documents do not show one. What they show is an apparatus repeatedly choosing perception-management over resolution: the 1953 Robertson Panel recommending debunking and monitoring of civilian groups; the FBI filing UFO organizations under internal security for fifteen years; a 1963 NASA-to-State memo titled “Thoughts on the Space Alien Race Question” confirming the policy question was live at the highest levels; a 1998 senator’s constituent inquiry bouncing off the White House; and a committee of French defense officials concluding in 1999 that the U.S. was running systematic disinformation on the subject. This chain is consistent with both hypotheses and proves neither. It is documented in full on The System.
The hypotheses are not exclusive, but they make different bets, and future releases can score them.
| Observation | Favors the Mirror | Favors the Survey |
|---|---|---|
| A stable multi-band characterization finally lands | Strongly against | Compatible; expected eventually |
| New sensor modality fields; “new” shapes appear on it immediately | Predicted | Unexplained |
| Honest base rates confirm site affinity after controlling for sensors | Unexplained | Predicted |
| Parent-child behavior found in pre-1947 records | Neutral | Predicted (continuity) |
| The residue shrinks to zero under better data | Both die | Both die |
The last row is the point. Both hypotheses are hostage to the residue, and both should be. The D25 SWIR case currently cuts against the Mirror; the missing base rates currently prop up the Survey’s biggest weakness. The next tranche can move either.
The record behaves like reconnaissance conducted by something patient, and the institutional story reads like eighty years of managing the embarrassment of not knowing what it is. That sentence is an interpretation, and it sits here, in the interpretation layer, because the corpus cannot close it. What the corpus can do is what it has done on this page: constrain the space of explanations, kill the cheap ones, and state what evidence would kill the rest.
For the documents behind every claim above, open the cited files in place, or work through the full Corpus. For the strongest cases these hypotheses lean on, see The Standout Cases; for the ones we set aside, The Contested Cases; for how established UAP thinkers fare against the same record, The Theorists.