The Investigation

What 375 declassified
records actually say

An investigation does not begin with an answer. It begins with a pattern, a hypothesis, and the discipline to attack your own best idea. This is that process, start to verdict.

Scroll to begin
The collection

A pile of documents is not evidence. Not yet.

Five government releases, 375 declassified files. The first PURSUE tranche delivered 158 files: FBI memos from 1947, State Department cables, NASA crew debriefings, hundreds of Department of Defense mission reports. Eight agencies, eight decades. A second tranche followed in May 2026, 64 additional records and videos, including a classified Sandia Laboratories correspondence and a first-person senior intelligence officer account with a multi-sensor chain of evidence that no prior document in the collection had matched. A third, in June 2026, added 72 more: AARO's own analysis of the western U.S. “orbs launching orbs” case, five first-person witness narratives, and a deep historical corpus reaching back to the 1948 military flying-disc studies and the 1953 CIA Robertson Panel. A fourth, in July 2026, brought the verbatim 1949 Los Alamos green-fireball conference and the full Pantex incident report, turning two summarised bookends into primary sources. And a fifth, in August 2026, reached further back than any of them: 245 pages of Project Sign correspondence, a 1947 War Department assessment of a Scandinavian wave that predates Kenneth Arnold, a 1953 Navy photogrammetric analysis, and four FBI interviews dated 2026 that turned the western U.S. case from a two-evening event into a multi-year one.

Of every document read and rated, only 38 held up as strong signal, a noise-to-signal ratio of about 6 to 1. Most of the record is inconclusive by our own triage, and many routine mission reports describe objects the crews could not positively identify.

375
files released, per war.gov
8
agencies
1946–2026
years

A collection that size invites a temptation: to read it, feel the weight of it, and conclude that the sheer volume means something. It doesn't. Volume is not proof. The only honest way forward was to form a specific hypothesis and then test it without mercy.

Act One: the pattern

The machine looked familiar

Reading the documents in order, a structure emerged. The secrecy around UAP, with its compartmentalization, its reliance on private contractors, its classification categories, even its vocabulary, did not look improvised. It looked inherited.

It looked like the security system built for one specific purpose in the 1940s: the atomic bomb. The same weapons labs. The same contractor model. The same instinct to wall information off so that no single person ever sees the whole picture. The first hypothesis wrote itself: the Manhattan Project's security apparatus never ended; it was repurposed.

In 1947, J. Edgar Hoover, head of the FBI, had to negotiate for access to recovered material. The country's top law-enforcement officer was being walled out. Someone, already, had a system.
Redacted sensor still from FBI UAP files
Imagery from the FBI's own UAP files. Even here, the data around the tracked object is blacked out, the first sign of a redaction pattern that recurs across eighty years of records.

It was a compelling pattern. And a compelling pattern is exactly when an investigation is in the most danger.

Act Two: the stress test

Attack your own best idea

The discipline that makes an investigation trustworthy is not finding evidence for your hypothesis. It is deliberately building the strongest possible case against it. So the same documents were run through four rival explanations, and a single distinction did most of the work.

The machine, provable

The secrecy apparatus itself: compartmentalization, contractor opacity, the classification system. The documents establish this clearly.

The cargo, unprovable here

What the apparatus is actually protecting. The documents describe the box in detail. They never let you see inside it.

Almost every dramatic claim in this field collapses the two. The evidence proves a machine exists. It does not reveal the cargo. Of ten major categories of evidence, only five turned out to discriminate between explanations at all. The rest were consistent with everything, and evidence consistent with everything proves nothing.

Act Three: one thing, or many?

"UAP" is a leftover category, not a phenomenon.

The next test asked whether "UAP" is even one thing. A strong idea proposed three: natural plasma; an automated layer of metallic spheres; and the rare, hard cases that seem to defy physics. To test it, every file was scored on six signature axes.

Plasmoids

Luminous, drifting, natural plasma physics

Surveillance-orbs

Metallic spheres, automated, loitering

Anomalous craft

Structured, physics-defying, controlled

UAP report sensor still, Greece 2023
A typical modern report frame from Greece, 2023. The faint dark form near the crosshair is the whole of what was recorded. It is rarely enough to tell one kind of phenomenon from another.

The model held up as a way of thinking. But the test revealed something sharper about the documents themselves: the data needed to sort a real case into one of these categories, its motion and its behaviour, is present in only a fraction of the files.

Mean signature completeness across the modern reports: about 30%. The reports log "possible UAP observed," then stop. The description is gone.

That three-part split was a working guess when it was made. The August 2026 release found that the Air Force had reached for the same move first. Its 1948 taxonomy sorted roughly 180 investigated incidents into four classes by “shape or physical configuration,” and the fourth is not a shape at all: balls of light with no apparent form attached. A category drawn from unaided human eyes, before radar tracking and long before infrared, describing exactly the formless luminosity this investigation labelled plasmoids seventy-eight years later.

That is the strongest single argument in the collection against the cheapest explanation of the modern record. If orbs were an artifact of the sensors, the category should not predate the sensors. It does. The heterogeneity guess has since been formalised into four competing readings, each with stated predictions and falsifiers, on The Hypotheses — where the descendant of “plasmoids” now ranks first, and the descendant of “anomalous craft” has to explain why eighty years have produced no artifact.

Act Four: the wall

Every road led to the same wall. So the wall became the subject.

Three analyses, three dead ends, and every one of them ended at redaction. So instead of treating redaction as the obstacle, the investigation turned it into the evidence. What is blacked out, under which legal authority, and how much, are all measurable.

The first finding was a relief. Nearly half of all the redaction in the collection is not hiding secrets at all. It hides the names of aircrew and witnesses. Personal privacy. Administrative. Noise.

45%names & privacy, ignore it
55%national-security redaction, the real signal
Heavily redacted sensor still, dated 1999
Redaction, made literal. Every black rectangle on this released sensor still is a removed passage. Counting them, and reading the legal codes behind each one, is what turned the wall into measurable data.

The second finding was sharper. The substantive redaction is justified, overwhelmingly, as the protection of military operations and weapons systems, and almost never as the protection of intelligence sources and methods. The wall is built like operational security rather than the concealment of a deep intelligence secret. A genuine clue, and a citable one.

The later releases sharpened that reading. The Tranche 2 documents arrive almost clean, averaging fewer than ten substantive removals per file against sixty-two in the Tranche 1 mission reports. Tranche 3 is mostly decades-old material released in full, and Tranche 4's founding studies and historical bookends the same — while its recent cases, the Pantex report and the Range Fouler tapes, keep their operational redactions. The wall, in other words, sits on the recent operational record, not on the history. Whatever is still being withheld is current, and it is withheld as military operational security.

The full version of this analysis, the triage in numbers, the classification-category fingerprint, the self-disclosed declassification schedule, and the ranked target list with a procedural path for each, now lives on Key Documents, where the seven highest-value targets are profiled individually.

The verdict

A precise answer, not a dramatic one

Here is where the evidence honestly lands. It does not crown a winner, and pretending otherwise would be the one genuine failure available.

Tilts mundane

The redaction fingerprint, operational rather than sources-and-methods, reads like the protection of ordinary military secrets.

Tilts the other way

On one night in January 1949, some thirty observers across five cities of the nuclear complex reported the same thing. Coincidence does not do that.

The Manhattan-descended secrecy machine is real and well-evidenced. Its cargo is undetermined. Two solid pieces of evidence point in opposite directions, and nothing in the declassified record breaks the tie. A second tranche of 64 records arrived in May 2026. It added a classified Sandia Laboratories file naming Edward Teller and closing the geographic triangle of the nuclear complex, and a first-person intelligence officer account with the most complete sensor chain in the collection. Neither document broke the tie. Both sharpened it.

A third tranche followed in June 2026, and it did not break the tie either. What it changed was the visibility of the machine itself. It surfaced the 1953 CIA Robertson Panel that made public debunking official policy, the U-2 and OXCART overflight history that recast real sightings as misidentified spy planes, and their modern echoes: an intelligence assessment calling a 2022 Cheyenne Mountain sighting possible sunlight backscatter, and AARO assigning roughly sixty percent of the western U.S. case to military flares while conceding that about forty percent stays unresolved. The cargo remained undetermined. The apparatus came further into focus.

A fourth tranche in July 2026 and a fifth in August did the same thing, harder. Neither broke the tie. What the fifth did was move the apparatus back four years and out of one country. In November 1948, having studied roughly 180 incidents and found a residue it could not explain, the Air Materiel Command advised headquarters that it was “not considered advisable to present to the press information on those objects which we cannot yet identify,” and that if pressed, the public should be told many objects had been identified as weather balloons or astral bodies. That is the posture, in writing, five years before the Robertson Panel. Released alongside it was a January 1947 War Department review recording the Swedish Defence Staff doing the same thing eighteen months earlier, for stated diplomatic reasons.

That changes the character of the finding. The machine is not only real and Manhattan-descended; the instinct to manage rather than resolve is not even distinctively American. It looks like what states do with unresolved information that carries risk. And the same tranche supplied the sharpest evidence yet that the failure to resolve was not always suppression: the Navy asked for a spectrum of the Utah film in 1953 and the released record never shows it being taken. The cargo stayed undetermined. The reason it stayed undetermined came into much better focus, and it is as much under-resourcing as concealment.

That is not the analysis failing. That is the analysis being honest about a question the available evidence cannot close, and naming, precisely, what it would take to close it.

What comes next

Knowing exactly what you don't have

An investigation that ends in honest uncertainty is not an investigation that failed. It is one that now knows precisely what it lacks, and the redaction analysis converted that into a concrete list of documents to pursue, in three tiers.

1

Tractable and high-value

Seven specific mission reports whose operational redactions, if lifted, would restore the missing descriptions; already-public agency reports; the National Archives UAP records channel.

2

The historical backbone

Largely delivered. This tier asked for Project Blue Book, the early Project Sign files and the nuclear-weapons agency records. Tranche 4 supplied the Blue Book correspondence, the 1948 and 1949 Air Intelligence analyses, the Project Sign progress report and the verbatim Los Alamos conference; Tranche 5 supplied the founding Project Sign file itself, 245 pages of it. The continuity claim it was meant to test now has its archive, and the answer it gave was that the pattern reaches back further than the target list assumed — to a Scandinavian wave in 1946. What remains outstanding here is the Blue Book case-file series in full, and the Project Twinkle instrumented-watch records.

3

Hard, but decisive

Sealed congressional testimony; the full sensor data behind the famous cases. And one item this collection can now name precisely, because a released document asked for it and no released document shows it was ever done: the spectrophotometric analysis of the Tremonton film, requested by the Naval Photographic Interpretation Center in 1953. A spectrum would separate emission from reflection, which is the single cleanest discriminator between the readings set out on The Hypotheses. It is also, by a wide margin, the cheapest unrun experiment in this subject.

The collection took the investigation as far as a collection can. The next move is acquisition. For the first time, there is an exact map of where to aim.

Tranches 2–4 · 2026

What the newer releases changed

When the first phase ended, the honest conclusion was this: the machine is well-evidenced, the cargo is undetermined, and the collection cannot break that tie. Two more releases arrived through 2026 with a precise advantage the first phase never had, a map of what was missing and a second and third look at the same subject.

The second government release delivered 64 records across documents and Department of Defense videos. Two of the documents are among the most significant in the entire collection.

DOW-UAP-D017: Sandia Laboratories correspondence (classified SECRET)

The Sandia weapons-engineering lab generated this classified SECRET file. Edward Teller's name appears six times. The Atomic Energy Commission and Kirtland Air Force Base are named repeatedly. Embedded in the correspondence: UAP descriptions from the Southwest nuclear complex in the late 1940s, the same green fireball cluster the FBI was separately tracking. This is the first primary Sandia document in the collection, and it closes the geographic triangle of the Manhattan inheritor complex: Los Alamos, Sandia, and the production sites are now all directly represented in the UAP record.

ODNI-UAP-D001: First-person narrative, senior U.S. intelligence officer

A senior member of the U.S. intelligence community, flying in a helicopter on a dedicated UAP investigation mission, describes an encounter lasting over an hour. Ground teams tracked an object on FLIR before visual confirmation. Radar at the Joint Operations Center provided independent tracking. Fighter jets diverted from a training mission were called in to assist. One object split into two and accelerated beyond the ability of pursuing aircraft to match. The author writes it down, in full, in a document now declassified. This is the most complete sensor chain (FLIR, naked eye, radar, fighter intercept) in the Tranche 2 PDF collection.

The redaction wall that defined the first phase is almost entirely absent from these new documents. Tranche 2 PDFs average under 10 substantive redactions per file versus 62 in the Tranche 1 mission reports. Either what arrived in the second tranche is genuinely less sensitive, or there is no classified version. For a release driven by congressional mandate, that ambiguity is itself a data point.

Then, in June 2026, a third tranche added 72 more records, and its character was different again. Most of it is history: the 1953 CIA Robertson Panel, the U-2 and OXCART overflight programs, NASA's Gemini crew debriefings, and decades of FBI files, almost all released in full. Where Tranche 1 showed the wall, Tranche 3 showed the machinery of explanation behind it, and one modern case that the apparatus could not fully close.

DOW-UAP-D077: AARO analysis of the western U.S. "orbs launching orbs" case

AARO's own unresolved-case analysis attributes roughly sixty percent of the western U.S. event to military flares and conventional sources, then states plainly that about forty percent remains unresolved. Five first-person witness narratives accompany it. It is the clearest example in the whole collection of the apparatus explaining what it can, and naming, in its own words, the part it cannot.

A fourth release followed in July 2026 — 36 records — and it did something the first three had not: it reached back to the origin of the official question and forward to its most recent nuclear-site case, and it documented both with primary sources rather than summaries.

DOE-UAP-D004: the verbatim Los Alamos conference transcript (1949)

On 16 February 1949 the institutional core of the American bomb program — among them Edward Teller and the future Nobel laureate Frederick Reines — met in secret at Los Alamos with the meteoriticist Lincoln LaPaz to explain the green fireballs then crossing the New Mexico nuclear complex. Tranche 4 releases the classified minutes in full. LaPaz argues the lights are not ordinary meteorites; the transcript records the Air Force folding the phenomena into its newly-renamed Project Grudge. It is the earliest primary source in the corpus, and it puts named physicists on the record, unable to explain the lights over their own labs.

DOE-UAP-D005: the Pantex Plant incident report (2015)

Sixty-six years later, ground-surveillance radar at the Pantex nuclear-weapons plant detects an unidentified object, and the Protective Force pursues it. Security officers watch it through camera and binoculars and record “no sound and no identifiable propulsion.” It is the primary source behind the Pantex image already in the collection, and the modern bookend to Los Alamos: the same class of unexplained activity, over the same kind of installation, sixty-six years apart.

Read together, the two bookends do one thing unambiguously: they turn the nuclear-proximity pattern into a seven-decade, primary-source record. The government's own files now place unexplained aerial activity over its most sensitive nuclear installations from Los Alamos and Sandia in 1949 to Pantex in 2015 — the best-attested throughline in the entire release. The honest caveat still holds: the clustering has not been corrected for observation density, since nuclear sites are also where the cleared observers and the best sensors are. But the pattern itself is no longer an inference drawn by civilian researchers. It is on the record, in the government's own hand.

Between the bookends, Tranche 4 supplied the machinery of official study itself as primary documents: Project Sign (1948), the Air Intelligence “Analysis of Flying Object Incidents” of 1948 and 1949, Blue Book correspondence, and the 1966 committee that spawned the Condon report. The System had assembled that arc from secondary accounts; now the founding paperwork is in hand. The government has opened — and closed — this question before, and the record of how it did so is finally on the table. Tranche 4 did not break the tie; it pushed both ends of the documented record further apart in time and closer to primary sources.

The honest verdict is unchanged across all four releases: the machine is real, the cargo is still undetermined, and the record cannot break the tie. What changed is the size of the corpus, the pedigree of the official explanations now visible, and the precision of the open questions. The documents that would settle it are identified. They are not yet public.

Undetermined does not mean unexaminable. The two working explanations that fit the strangest patterns in this record better than their rivals, each stated formally, checked prediction by prediction against named documents, and given its falsifier, are on The Hypotheses. That page is where this investigation's verdict goes next.

Explore the Evidence Database → Watch the 133 declassified videos
Go deeper

The analyses, each an angle on the same question

The investigation narrative is the story. The analyses are the data behind it. Each one is interactive and standalone.

The Evidence Database33 signal-rated incidents with skeptical cases, primary sources, and DVIDS video for each The Video RecordThe PURSUE videos and audio, embedded and hosted with AARO descriptions Continuity Timeline1942–2026: people, incidents, and places on one scrollable canvas Nuclear-Geography MapEvery incident plotted against US nuclear infrastructure MorphologyThe vanishing disc and the constant orb: how the reported shape drifts across eight decades The AtlasCraft shape cross-cut against site and era, across the corpus Key DocumentsThe 35 files whose redaction hides something substantive BehaviorsEight recurring craft behaviors across the corpus: Cold IR, kinematics, nuclear proximity, and more How the System WorksThe Bennewitz blueprint and three contemporary cases: the machinery of explanation, documented The HypothesesThe Mirror and the Survey: two explanations tested, prediction by prediction, against named documents The Official StudiesSign to Grudge to Robertson to Blue Book to Condon to AARO: the government has closed this question before Before & After PURSUEWhat the declassified record moved from suspicion to documented fact, claim by claim