How strong is the 2021 Gulf of Oman AC-130 "cold orbs" case?

Filed: 2026-08-07

Primary report: ../raw/reports/pursue-release-05/dow-uap-d101-iir-unresolved-uap-report-gulf-of-oman-2021

Released videos: ../raw/media/pursue-release-05/dow-uap-pr117-pr122-gulf-of-oman-2021

Gemini video inspection: ../raw/extracts/dow-uap-pr117-pr122-gemini-3-6-flash-cross-comparison-2026-08-07

PURSUE record description: ../raw/extracts/war-gov-d101-gulf-of-oman-record-description-2026-08-07

Official PDF: https://www.war.gov/medialink/ufo/release_05/Aug_07/documents/DOW-UAP-D101_IIR_Unresolved-UAP-Report-Gulf-of-Oman_2021.pdf

Short answer

DOW-UAP-D101 is strong evidence that an AC-130J crew made a sincere, detailed, sensor-aided report of targets they could not identify. It is not publicly reproducible evidence that four-foot objects flew at 250-1,300 mph just above the sea or reacted intelligently to gunfire.

The report is unusually specific and comes from a capable military sensor platform. Its central weakness is equally specific: the aircraft's native digital recording was reportedly corrupted. The six released clips are cellphone recordings of the sensor display, not primary sensor files, and they omit the telemetry and geometry needed to verify size, altitude, and speed. The right classification is a genuinely unresolved and operationally interesting case with extraordinary kinematic claims that remain unvalidated.

The PURSUE record description reinforces that distinction. It identifies IIRs as unevaluated inputs rather than finished intelligence, calls the speed figures aircrew visual estimates of apparent speed, and expressly says the report's descriptive and estimative language is subjective and not conclusive evidence of intrinsic object features or performance.

What the report says

The incident occurred during darkness on 8 September 2021 while a U.S. Special Operations Forces AC-130J was conducting live-fire training over the Gulf of Oman. The report was produced by Air Force Special Operations Command in April 2022, originally classified SECRET//NOFORN, and explicitly marked "not finally evaluated intelligence."

The reported sequence is:

The aircraft dropped a water-activated flare as a mock target and orbited it.
A Combat Systems Officer scanning with the aircraft's electro-optical/infrared sensor reportedly saw two approximately four-foot "cold orbs" stationary 0-20 feet above the water and over the flare for about 15 seconds.
When the officer pulled the trigger for the 105 mm cannon, the two targets reportedly departed rapidly before the cannon recoiled, without changing altitude.
Over the sortie, the crew reported approximately 25 UAP "instances" in white-hot and black-hot infrared modes. They described pairs, a formation of three, erratic movement, aggressive maneuvering, and coordinated wave-like movement compared to dolphins swimming together.
The Combat Systems Officer reportedly used TACTOOL to estimate maneuvering speeds between 250 and 1,300 mph. The report gives no calculation, input data, uncertainty, or validation.
The crew attempted to preserve the entire sensor recording, but the aircraft's digital video recorder reportedly failed and its data became corrupted. The report listed a PowerPoint attachment containing six embedded videos.

The source was described as having direct access through official duties and being available for recontact. The report was coordinated with U.S. Defense Attaché Office Muscat and SOCCENT J2X "without comment." That documents routing and source access, not independent confirmation of the targets or their performance.

What the released videos establish

DVIDS released six clips as DOW-UAP-PR117 through PR122. Together they run about 124 seconds, although PR121 and PR122 significantly overlap. Each is a cellphone recording of an infrared display. DVIDS expressly warns that these are not native sensor data and that filming a screen introduces lost resolution, blur, flicker, and other artifacts.

The clips visibly contain small areas of contrast moving through or remaining near the sensor field of view. PR120 includes sensor modality changes. PR121 and PR122 show relatively larger and smaller areas of contrast crossing the display, but DVIDS says those two clips substantially duplicate one another. Display symbology, sensor movement, mode changes, phone framing, and screen-recording artifacts are all mixed into the public imagery.

The clips do not publicly expose enough information to derive physical motion. They lack the original pixel data, full telemetry, sensor model and field of view, target range, aircraft position and motion, line-of-sight angles, tracking state, and an error budget. They also do not identify the flare-and-cannon sequence described in the report. The videos therefore support the narrow proposition that the crew recorded unexplained infrared contrasts. They do not independently verify the report's size, altitude, speed, or apparent response to the cannon trigger.

Gemini 3.6 Flash video inspection

An agy run through AOP gave all six files to gemini-3.6-flash-high as multimodal inputs under a closed-source constraint: inspect the videos directly and do not use the official descriptions, neighboring notes, filenames as evidence, web sources, or outside knowledge. The run reports complete decoding of all 3,710 frames and the audio streams. All seven requested artifacts passed AOP's output validation and are preserved verbatim in the linked raw review.

The model assessed that phone and display artifacts dominate the imagery and that it could not establish an independently moving physical target trajectory. It also found no visible flare or cannon-trigger sequence. These are machine interpretations, not dispositive findings. In particular, labeling every ring or point as display symbology or a recording artifact is stronger than the public pixels can establish. A tracked real target could remain near the reticle, and absent range and geometry, negligible motion relative to the reticle cannot be converted into a physical speed of zero.

The most reproducible cross-clip finding is that PR122 is wholly contained within PR121 beginning at about 8.07 seconds. An independent FFmpeg comparison found the same visual sequence with an average PSNR of about 29.10 dB, consistent with a truncated, re-encoded excerpt rather than pixel-identical duplication. The model also assessed PR119 as a short excerpt from the visual sequence represented in PR120.

The review therefore reinforces the evidentiary limit already visible in the release: these clips do not support a public kinematic reconstruction. It does not prove that no physical target existed.

What raises the case's evidentiary value

The reporting chain is official and contemporaneous enough to be meaningful. The incident was in September 2021, the source debrief occurred in April 2022, and the internal report was published later that month.
The platform was capable. The Air Force describes the AC-130J as carrying two electro-optical/infrared sensors, advanced fire-control equipment, and a 105 mm cannon.
The report is incident-specific rather than folkloric. It identifies the platform type, training context, sensor mode, reported size and altitude, timing relative to the trigger pull, estimated speed range, and the method named for the speed estimate.
The observation was reportedly repeated during one sortie and included more than one target and more than one infrared polarity.
Some contemporaneous imagery survived in secondary form, which is better than a testimony-only case.

What limits it

The report is an Intelligence Information Report and labels itself "not finally evaluated intelligence." The accompanying PURSUE text likewise says IIRs disseminate unevaluated information and are inputs to later all-source assessments, not finished intelligence. It records what a source reported. The classification marking and wide distribution list do not convert those statements into validated findings.
The native sensor recording is missing. The reported DVR failure removed the most important evidence for independent analysis.
No radar track, second platform, independent sensor, or physical trace is identified.
The size, altitude, and speed estimates cannot be reproduced. Apparent angular motion from a moving aircraft over water can translate into very different physical speeds depending on target range and geometry.
"Cold orb" is a display description. A dark or bright unresolved infrared contrast does not by itself establish an object's shape or physical temperature.
"Approximately 25 instances" does not necessarily mean 25 separate physical objects. The public record does not resolve whether it counts repeated reacquisitions.
The formation and agency language is partly interpretive. The source comment that the targets "knew they were flying together" and the comparison to playing dolphins go beyond what the released pixels alone establish.
The timing relative to the cannon trigger is potentially important, but the public clips do not let an outside analyst verify it.

The portal says its redactions protect eyewitness identities, government-facility locations, or sensitive information about military sites unrelated to UAP, and that no redactions concern the nature or existence of the reported encounter. That narrows the stated reason for visible redactions. It does not supply the missing native recording, telemetry, TACTOOL inputs, or independent corroboration.

Competing explanations

The public material is insufficient to select a conventional or extraordinary explanation. Possible categories include physical objects near the water, more distant objects whose range was misestimated, line-of-sight and parallax effects, sensor or image-processing artifacts, screen-recording artifacts, or some mixture of these. The report does not provide enough data to test them against one another.

If the reported 0-20 foot altitude and 1,300 mph maneuvering speed are accurate, ordinary birds, balloons, and small drones would be poor fits. The problem is that those are precisely the quantities the public release does not allow anyone to validate. Calling the case unresolved is justified. Treating the unresolved label as evidence of exotic technology is not.

Prosaic candidates ranked by fit

No single ordinary object matches all the reported characteristics simultaneously. The best prosaic explanation would probably combine an ordinary target with incorrect range geometry and sensor or display effects.

Candidate	What fits	What does not fit
Small drones or a drone group	Roughly four-foot scale, hovering, low-altitude operation, pairs or groups, coordinated motion, and abrupt direction changes	Representative industrial multicopters top out near 50 mph, not 250-1,300 mph. A drone should also have some thermal signature, although its unresolved body could appear cooler than warm water. This becomes plausible only if the speed and range estimates are wrong.
Distant aircraft	Speeds of 250-1,300 mph and formations of two or three are conventional aircraft characteristics. Distant aircraft can reduce to point-like infrared contrasts.	They would not be four feet wide, hover 0-20 feet above the sea, or make tight dolphin-like maneuvers. This candidate requires a major range and altitude error.
Seabirds or flying fish	Multiple targets, coordinated wave-like motion, and operation immediately above the water are good behavioral matches. Flying fish would be close to sea temperature.	Neither can hover for 15 seconds or travel hundreds of miles per hour. Birds normally appear warm in thermal imagery. They fit only if apparent motion was strongly inflated and some reported details were interpretive.
Balloons, debris, or chaff-like material	Small, low-contrast or cooler targets can drift in pairs or groups and seem stationary relative to a tracked point.	They do not independently perform aggressive maneuvers or reach the reported speeds. Aircraft motion, sensor tracking, and incorrect range would have to create most of that appearance.
Sensor, optical, or display effects	Point-like or ring-like contrasts, rapid apparent motion, sudden disappearance, modality changes, and a seeming response coincident with a weapon-system event can arise from image processing, optical ghosts, tracking behavior, or fire-control coupling. The released phone recordings visibly mix targets, display symbology, sensor motion, and screen artifacts.	A simple fixed detector defect does not naturally explain repeated apparently coordinated tracks across a sortie. A more complex combination of real targets and sensor effects fits better than one isolated bad pixel.
Training ordnance or projectiles	High speed, the live-fire setting, and lack of sustained propulsion fit ballistic objects.	The first pair was reported before the cannon fired and allegedly hovered for 15 seconds. Projectiles do not form playful groups or repeatedly maneuver around one another.

The drone hypothesis is the best object-level behavioral fit if the speed is discarded. Distant aircraft are the best speed fit if the size, altitude, and hover descriptions are discarded. Sensor and range-geometry effects are the only prosaic mechanism that can reconcile either candidate with the full set of apparent kinematics.

Range is load-bearing. Transverse speed derived from angular motion scales with assumed distance. A target 1 km away interpreted as 25 km away can acquire a calculated transverse speed roughly 25 times too large. An orbiting AC-130 also generates strong parallax against the sea. AARO's GoFast reconstruction is a useful demonstration of the general problem: an object that looked extremely fast near the ocean was calculated at roughly 5-92 mph once the aircraft, range, wind, and viewing geometry were modeled. DOW-UAP-D101 does not release the corresponding inputs needed to audit its TACTOOL result.

The claimed response to the cannon is less clear than it sounds. The report places the departure after trigger pull but before recoil. If that timing is exact, the targets could not have reacted to the projectile or blast because neither had occurred. A coincidental departure, a fire-control or sensor-state change associated with the trigger, or imprecise retrospective timing are prosaic possibilities. The public clips do not identify this sequence, so none can presently be tested.

White-hot and black-hot modes are also not independent sensors. They reverse the display polarity of infrared data. Persistence in both modes makes a simple palette misunderstanding less likely, but it does not independently confirm an object's temperature, shape, range, or motion.

External checks

The Air Force AC-130J fact sheet confirms that the platform carries two EO/IR sensors, advanced fire-control equipment, and a 105 mm weapon: https://www.af.mil/About-Us/Fact-Sheets/Display/Article/467756/ac-130j-ghostrider/
A representative four-foot-class industrial multicopter, DJI's Matrice 300 RTK, has an 895 mm wheelbase, can hover, and has a stated maximum speed of 23 m/s or about 51 mph: https://www.dji.com/support/product/matrice-300
AARO's GoFast methodology shows how aircraft motion, range, wind, and parallax can turn modest target motion into an impression of extreme speed: https://www.aaro.mil/Portals/136/PDFs/case_resolution_reports/AARO_GoFast_Case_Resolution_Card_Methodology_Final.pdf
AARO's official imagery page includes a separate case in which distant commercial aircraft appeared only as small infrared dots, illustrating the loss of identity and scale at long range: https://www.aaro.mil/UAP-Cases/Official-UAP-Imagery/

Net

This is one of the better operational UAP reports in the public PURSUE material because it combines a capable platform, repeated observations, named measurement tooling, a direct source, and surviving contemporaneous clips. It is also a textbook example of why an impressive narrative can outrun the released evidence. The public record supports "trained crew encountered puzzling infrared targets." It does not yet support the stronger performance claims at measurement-grade confidence.

Followup items

Obtain the original PowerPoint attachment and any surviving native or less-degraded EO/IR segments.
Obtain the TACTOOL inputs, target-range assumptions, track files, aircraft telemetry, sensor mode and field-of-view data, and uncertainty calculations behind the 250-1,300 mph estimate.
Seek the source debrief, crew statements, and any radar, mission-system, or second-platform records.
Identify the exact public clip, if any, that shows the flare and 105 mm trigger sequence.
Determine whether AARO completed a later case analysis beyond publishing the report and descriptive video notes.