Detecting Unknowns

The Science Behind the Confronting Unknowns Framework

1. Why do modern sensor systems fail to reliably detect events that fall outside the categories they were built for?

2. How do we tell a genuine anomaly from a sensor failure or a processing artefact?

3. What does it mean to identify an event probabilistically, rather than as resolved or unresolved?

4. What can artificial intelligence add to this work, and what should it not be asked to do?

5. How does an analyst move from raw sensor data to a ranked list of explanatory hypotheses?

The Confronting Unknowns Framework (CUF) requires all its constituent elements of Detect, Identify, Respond, and Explain to perform effectively and their handoffs to carry an event from first observation through to public communication. Here we cover the key challenges of the Detection and Identification elements of CUF, including discussing gaps and opportunities for improvement in the current practice of dealing with anomalous events, and some of the limitations of our own framework and direction for future research.

Detecting unknowns starts with observing possible anomalies through technological sensors, human sensory observation (visual sighting), or both. A credibility dilemma has plagued observations of anomalies such as UAP. Keep or discard observations that go against commonly perceived beliefs of what is possible or real? We dive deep into the gaps and opportunities of human observations and sensors in this section.

What Makes Something Anomalous: Six Observables

The Confronting Unknowns Framework endorses the Unidentified Anomalous Phenomena (UAP) definition from U.S. policy discussions, including the UAP Disclosure Act, as events exhibiting one or more of six observables:

• Instantaneous acceleration absent apparent inertia

• Hypersonic velocity absent a thermal signature and sonic shockwave

• Transmedium travel (e.g., space-to-ground, air-to-undersea)

• Positive lift contrary to known aerodynamic principles

• Multispectral signature control demonstrating intelligent stealth capabilities

• Physical or invasive biological effects on close observers or the environment

We discuss examples based on “Aerial” UAP, which is perhaps the most common type of anomalous incidents, while recognizing that “A” stands for “Anomalous,” a broader category that spans across multiple domains (space, submerged, transmedium, etc.).

Inset A: Observables associated with Unidentified Anomalous Phenomena.

The Observation Credibility Dilemma

A core challenge of UAP detection has been evaluating the credibility of any given observation. The high volume of observations that have been outright dismissed due to “low credibility” of either the technological sensor or the human observer has hindered operational decision making and slowed sensemaking. This dismissal is part of a long-standing social stigma and has led to the broader field of UAP investigation being dismissed as lacking rigor and unworthy of serious investment.

A rigorous analysis of anomalous observations should not only preserve all relevant data, regardless of perceived credibility, but also rate detection events by their level of credibility. Two elements correlate with the credibility of a UAP observation: the training of human observers and the amount, diversity, and grade of sensor observations (collectively: technological capability). Both are typically the lowest for the public, the highest for the military, and between these extremes for law enforcement and civil aviation personnel. Operational credibility scales, including a composite detection-scenario scale and a human-observer rating system, are provided in Appendix J: Observation credibility rating scale for frontline operations. The scales are designed as shared vocabulary for downstream analysis. The credibility of an observer, whether human or sensor, has direct consequences for how much evidence is needed to reach a given confidence threshold. Figure 5 illustrates this ‘sensor shortcut’: to exceed ninety-nine percent certainty on the leading hypothesis from a skeptical prior, dozens of independent civilian eyewitness reports are required, whereas just several fused multi-sensor detections suffice. The numbers in this figure use heuristics from the forensic science literature (e.g. [16]), where we rate a civilian witness as providing weak evidence, a trained observer as moderate evidence, and a multi-sensor event as strong, starting from a highly skeptical prior of 1×10⁻⁶ for the hypothesis. This is an illustrative example¹Footnote 11 Analysts assigning different likelihoods will reach different thresholds. but the qualitative conclusion is robust: Bayesian updating rewards evidence quality exponentially rather than linearly. As such, investing in calibrated detection infrastructure is not merely preferable; it is the highest-leverage action available.

Figure 5. The ‘sensor shortcut’ is to improve evidence quality over quantity in Bayesian UAP identification.

From Explanation to Better Detection: The Observer Education Feedback Loop

A systematic effort to explain resolved UAP cases to the public creates a positive feedback loop for detection quality. When observers, particularly untrained ones (ratings 1 and 2 on the frontline credibility scale, see Appendix J: Observation credibility rating scale for frontline operations), are educated on prosaic causes behind commonly reported aerial phenomena (satellite constellations, atmospheric optics, drone swarms, classified test platforms), they develop better mental models for distinguishing the mundane from the genuinely anomalous. This raises the signal-to-noise ratio of incoming reports and focuses analytical resources on cases that most warrant investigation. The same principle applies to trained observers at higher credibility ratings, who benefit from reference frames that account for rapidly evolving aerospace technologies, including those that are classified and therefore indistinguishable from truly anomalous phenomena to even well-trained eyes.

This educational effort must acknowledge that the boundary between "advanced technology" and "anomalous" is not static. The current UAP landscape is almost certainly a heterogeneous mix of misidentified prosaic events, advanced or classified human-made technologies, and a residual set of genuinely unexplained phenomena. Transparent explanation of resolved cases preserves the significance of the unresolved while sharpening the collective ability to isolate the truly anomalous signal. By closing the loop between investigation outcomes and observer education, the CUF can progressively elevate baseline observer competence across all categories.

Hotspots and Observation Bias

Spatial clustering of UAP reports is a familiar feature of the public record. Sightings concentrate in the American West, along the Atlantic seaboard, in the Pacific Northwest, and on coastlines bordering restricted military airspace. The temptation is to read these clusters as evidence of where the phenomenon occurs. A 2023 Bayesian regression of 98,724 public UAP reports across the conterminous United States from 2001 through 2020 [17], found instead that reports correlate with where the sky is easiest to see and where aerial traffic is densest.

The methodological implication is sharp: spatial clustering is, in part, a map of observation opportunity rather than of phenomenon density. Any geographic-pattern claim must be corrected for observer density, sky-view potential, and traffic before it can be treated as evidence about the phenomenon itself. Observer bias is a measurable confounding factor, present in every human report, that can be modeled and subtracted rather than treated as disqualifying. The same logic applies to temporal clusters around news events, military exercises, and astronomical phenomena.

The CUF accounts for this explicitly. Before any spatial or temporal clustering pattern is treated as evidence about the phenomenon itself, the framework requires that observer density, sky-view potential, and traffic exposure be characterized and corrected for. A hotspot that survives that correction carries genuine evidential weight. One that does not is a map of observation conditions, not of anomalous activity.

Sensor Capabilities and Challenges

Technological sensors extend detection and measurement capabilities beyond the limits of unaided human perception. These instruments enable detection, tracking, and characterization of objects and phenomena across a wide range of conditions, including at long range, in darkness, through obscurants, and across portions of the electromagnetic spectrum invisible to the human eye. Table 7 summarizes sensing capabilities that are relevant to characterize an airborne target; each is described in detail in Appendix D: Sensor Capabilities – Observables, Modalities, and Limitations, alongside the sensor modalities that detect them. (Established infrastructure is covered by Appendix C: Existing surveillance, sensing, and reporting infrastructure)

Table 7. Sensing capabilities to characterize airborne targets.

These observables are rarely available together. As such, the remainder of this section examines three structural consequences: spectral composition as the oft-missed modality, the fragmentation of detections captured, and the fusion necessary to compensate for these limitations.

The Case for Hyperspectral Sensors

Spectral imaging sensors are largely absent from the aerial platforms most likely to encounter UAP. Fighter aircraft such as the F-35 and F/A-18 carry advanced sensor suites that fuse radar, electro-optical, and infrared data, but these are optimized for threat detection and tracking, not for material identification. Similarly, c-UAS platforms rely on radar and EO/IR sensors sufficient for quick threat/no-threat decision making but yield no information about material composition. UAP identification has not historically generated the operational requirements that drive sensor integration on either fighter or c-UAS platforms.

Deploying spectral imaging in these operational scenarios presents non-trivial technical challenges, including detection range trade-offs with instrument aperture and integration time, illumination dependency, complexity of real-time data processing, and platform integration constraints (sensor size, cooling, weight, and power). However, recent military R&D efforts [18] targeting miniaturized hyperspectral sensors for small UAS, as well as early flight demonstrations [19] of spectral payloads on tactical UAS and manned aircraft, suggest these challenges are tractable and comparable in complexity to those already solved for existing FLIR and IRST systems. The recommended first step is to fund the development of a Concept of Operations (CONOPS) and engineering requirements specifically addressing spectral sensing for UAP identification, and enabling conceptual designs and costing for integration across relevant aerial platforms.

Detection by Accident, Evidence in Fragments

Because detections originate from systems built for other missions, each data point captures only the fraction of an event its mission requires: a defense radar fixes position and velocity but reveals nothing about composition; an infrared targeting system produces striking imagery but cannot bound range. The most compelling cases involve several simultaneous modalities, yet these are precisely the cases whose data is least often assembled into a standardized package, with classification limiting validation in military contexts and stigma suppressing it in civilian ones. The result is an evidentiary record built from mismatched fragments.

Anomaly or Artifact? Ruling Out Non-Anomalous Explanations

Most sensor records of UAP events are serendipitous, unexpectedly captured by instruments calibrated for other missions, in environments not chosen for the observation. The NASA UAP Independent Study Team identified poor calibration, missing metadata, and absent baselines as the central data deficit. Before an event is treated as anomalous, it must survive a first-pass assessment against the ordinary explanations that can mimic an anomaly: sensor faults (calibration errors, processing artifacts, and environmental interference), deliberate adversarial effects (jamming and spoofing), and the canonical signal-processing pipeline itself, which routinely discards the very returns that matter. Common ‘hygiene’ of a signal-processing pipeline includes Moving Target Indicator (MTI) and Doppler filters to suppress near-zero radial velocities, Constant False Alarm Rate (CFAR) algorithms to censor sub-threshold targets, and tracker gates to reject kinematically improbable plots, producing a synthetic blindness in which UAP-like signatures exist in raw data but never reach the display, as illustrated in Figure 6, which contrasts what the sensor actually captured with the cleaned-up picture the operator is shown. Machine-learning detectors do not inherently escape this: a model trained on conventionally labeled returns learns the same rejection rules, reproducing the blindness in less transparent form unless built to flag anomalies rather than match familiar traces. A verifiable anomaly is what survives all of these checks and still shows signatures across different sensor types. Appendix E: Diagnostic Triage for Sensor Faults, Interference, and Deception catalogs the principal classes of explainable effect, the signatures by which each is recognized, the diagnostic questions that distinguish it, and the corresponding mitigations.

Figure 6. Algorithmic blind spots can manifest when legacy radar-processing pipelines filter out low-radial-velocity or kinematically improbable returns, producing synthetic blindness in which UAP-like signatures exist in raw sensor data but never reach the tactical display.

Determining whether any of these underlies a given detection requires ancillary sensor data that typically exists but is rarely released for independent analysis: calibration status, maintenance logs, firmware version, and raw telemetry, along with records of how the same sensor behaved on known, prosaic objects. These are generally recorded but often not retained or not made accessible²Footnote 22 This is owing to classification, retention policies, or the simple absence of any requirement to preserve them. and may be irretrievably lost as a result. Establishing detection-record standards that bundle this ancillary information with the primary observation would substantially increase the credibility and analytical utility of sensor-based UAP reports.

Multi-Sensor Fusion

No single modality resolves a UAP detection, as each leaves gaps the others can fill. Multi-sensor fusion combines them to constrain the hypothesis space: an event registered across radar, optical, and infrared simultaneously rejects most artifact and interference modes that would mimic a single channel. Fusion also enables triangulation across spatially separated stations, which converts apparent angular motion into three-dimensional kinematics and corrects for forced-perspective effects.

The NASA UAP Independent Study Team identifies multi-sensor, well-calibrated data as paramount for advancing the field. Operationally, the Galileo Project commissions all-sky infrared arrays paired with visible cameras, RF, acoustic, and magnetometer sensors at fixed sites [20]. This multi-modal architecture is what makes it possible to discriminate between explanatory hypotheses, thus building confidence in genuinely anomalous explanations only after artifact and interference explanations have been independently ruled out across sensors.