Ten Cluster Analyses Against PURSUE. The Phenomenon Has A Stable Phenotype And At Least Two Object Classes. Here's The Synthesis.

The U.S. Department of War's PURSUE Release 1 and Release 2 together comprise 222 declassified UAP records as of May 22, 2026. The press cycle covering each release focused on the items DoW pulled out in the press release. The clusters that actually carry the analytic weight have been sitting unsurfaced.

We ran ten cluster analyses against the full 222-record corpus over the last 24 hours. Each cluster cut the data along a different axis — geography, year, multi-object behavior, sensor modality, witness type, pairing relationships. Some clusters were obvious (the 2020 CENTCOM surge). Some were obvious-but-undercounted (the multi-object pattern, which we found ten events across five years, not the two we initially named). Some were buried so deeply that no major coverage of PURSUE has named them yet — the Late 2025 Western US orb swarm at a sensitive testing installation, the March 25, 2022 transmedium spherical UAP near a U.S. submarine.

This post is the synthesis. The interesting finding is that the ten cuts of the corpus describe an internally consistent phenomenon, with exactly one productive contradiction — the contradiction itself tells you something.

The stable phenotype

Across all ten clusters, the same properties recur. The phenomenon as documented by PURSUE has:

A modern object morphology that is dominated by spheres and orbs. Historical records from 1948 through the mid-1980s skew toward disc descriptions. From 2020 forward, the dominant shape is spherical. Tic Tac descriptions appear in only two records, both in 2024. The 2017-onward public narrative that fixated on the 2004 Nimitz Tic Tac is over-fitted to a single outlier shape that the larger corpus does not corroborate.

A sensor signature strongly weighted toward thermal infrared. Sixteen of eighteen CENTCOM 2020 records explicitly mention infrared. The Late 2025 Western US orbs were described as "super-hot," visible to night-vision goggles, and corroborated by ground-based NVG observers. The Apollo 12 records describe visible-spectrum observation in the 1969 lunar context. The shift from visual to thermal documentation tracks the deployment of FLIR-class sensor pods on military aircraft.

A behavior profile that includes multi-object swarming, formation flight, transmedium traversal, and instantaneous appearance and disappearance. The Iran August 26, 2022 video shows four objects in coordinated formation. The April 12, 2021 sphere set shows four sequential captures from one callsign on one day. The Late 2025 Western US case describes 4-5 orbs flaring up and down for 30 minutes with an additional "swarm of lights moving in all directions" across the area. The March 25, 2022 USO event near a U.S. submarine shows objects crossing the water surface in and out. The September 1, 2023 case describes an ellipsoid that materialized out of a bright light and disappeared instantaneously.

A geographic affinity for military chokepoints, restricted waters, and nuclear-adjacent infrastructure. Of 222 records, 32 are at maritime locations; of those 32, the heavy concentration is at narrow water passages — the Strait of Hormuz, the Persian Gulf, the Arabian Sea, the Yellow Sea, the Aegean, Lake Huron. The DOE-released records include PANTEX (the only U.S. nuclear weapons assembly facility), Sandia Base 1948-1950, and James Tuck's 1970s Manhattan Project correspondence. The "sensitive government testing installation in the Western US" framing for the Late 2025 case reinforces the same pattern.

A temporal pattern in which surge windows correlate with geopolitical tension. The 2020 CENTCOM surge runs August through November of 2020 — immediately after the Soleimani killing and during heightened U.S.-Iran maritime activity. The March 2022 USO event was thirty days after Russia's invasion of Ukraine. The Late 2025 Western US case is unexplained as of this writing but the pattern would suggest a corresponding strategic context if cross-referenced against late-2025 geopolitical events.

A witness fingerprint that, when access permits, always includes multi-modal corroboration. The Late 2025 case has senior USIC eyewitness plus two pilots on NVG plus other pilots in separate aircraft plus ground NVG plus 32 sensor captures. The September 1, 2023 case has three FBI 302 interviews plus a composite sketch overlaid on an actual site photo. The Apollo 12 case has three crew members in medical debrief plus voice transcript plus five archival photographs annotated by NASA. The strongest cases all stack four or more independent observation modalities pointing at the same event.

That is the stable phenotype. Six properties. Internally consistent across ten cluster cuts. Reinforced rather than contradicted by every additional axis we examined.

The one productive contradiction

The phenomenon as a whole is internally consistent. Within it, exactly one pair of records contradicts each other in a way that is itself informative.

On February 12, 2023, a U.S. Air National Guard F-16C shot down a UAP over Lake Huron with a redacted weapon system. The PURSUE record DOW-UAP-PR071 contains a 46-second IR video showing the kinetic interaction. At the 20-second mark, the object fragments in a radial displacement pattern consistent with a high-energy event. The UAP, in this case, was destroyed by a conventional U.S. missile.

On March 25, 2022, ten months earlier, multiple spherical UAP were observed by an IR sensor moving in and out of the water near a U.S. submarine. The PURSUE record DOW-UAP-PR067 contains a 4-minute 50-second video showing the spheres crossing the air-water interface multiple times. No engagement was attempted. The objects were operating in a medium — water — where no known foreign or domestic platform on the current military inventory can functionally operate.

These two behaviors are physically incompatible for the same object class. A platform that fragments on missile impact is not the same class of platform as one that transits through water freely. The phenomenon, therefore, is not unitary. The corpus describes at least two distinct object classes with different vulnerability profiles and different capability envelopes.

That distinction matters. Public discourse on UAP frequently treats the phenomenon as monolithic — as though every observation is necessarily the same thing. The cluster analysis says no. PURSUE documents at least Class A (kinetically destructible, observed and engaged in atmospheric flight) and Class B (transmedium-capable, observed but not engaged). The classes may share an origin or they may not. The data is silent on that. What the data is not silent on is that they exist as separate evidentiary tracks.

What the synthesis does not say

This is not an argument for any particular origin. It is consistent with the orbs being foreign-adversary advanced platforms operating against U.S. naval and strategic assets, in which case the morphology evolution from 2020 sphere clusters to 2025 swarm behavior maps onto a foreign-program drone-development curve. It is also consistent with the orbs being a classified U.S. test program where even cleared senior officials witnessing the activity did not have full briefing access. It is also consistent with a genuinely novel phenomenon that the U.S. government cannot identify and is releasing precisely because it cannot identify it. The data does not discriminate between the three. The data discriminates between "the phenomenon is unitary" and "the phenomenon includes at least two classes." Beyond that, the official record is silent.

The methodological lesson that travels

The interesting finding, structurally, is not any particular case. The finding is that running ten independent clustering passes against a 222-record corpus produced an internally consistent six-property phenotype plus exactly one productive contradiction. That kind of consistency does not emerge from a noisy dataset of unrelated observations. It emerges from a curated dataset where the curating party — AARO and the agencies that submitted to AARO — has already done significant analytic work in deciding which records to release together.

The synthesis you get from cross-cluster analysis is, by definition, the synthesis the agencies left visible to anyone willing to read the records as a corpus rather than as individual press-release items. The press did not do that. The analyst community did not do that. The cluster work that surfaces the synthesis is the same shape as the threat intelligence work that surfaces a supply chain campaign before public naming — same primitive, same discipline, different domain.

The data was always there. The detector finds the case. The case is timestamped. The receipt is on the public record. The methodology is repeatable. The next release will yield more.

---

How do AI models see YOUR brand?

AIPM has audited 250+ domains. 15 seconds. Free while still in beta.

Audit your site now → aipmsec.com