Instrumented UAP Surveys

Speculative The condition comes first, because without it there is no prize: the value of an instrumented sky survey rests on somebody measuring how often it calls a known object unknown, and no programme in this field has published that measurement. Everything below is what becomes available once that number exists, and 7 describes the campaign that would produce it.

Established The prize is an upper limit with an error bar where the subject currently has a headline. The Galileo Project’s eight-camera infrared array reconstructed roughly 500,000 aerial trajectories over five months of commissioning and came out with 144 that reviewers could not resolve. The same paper reports that the array records 41% of the aircraft crossing its effective field of view and detects them in 36% of individual frames. Those two facts belong in one sentence and are almost never put there. Put there, they say something exact: the residual is a property of the pipeline at least as much as of the sky, and the paper’s own 95%-confidence upper limit of 18,271 outliers for the five-month interval — 127 times the observed count — is what honest accounting of that uncertainty costs.

Frontier The magnitude, in the units of the field that already solved this problem: the Dark Energy Survey injected 146 million synthetic sources into its own images across 5,000 square degrees to measure precisely this quantity. That is the scale at which a survey buys its selection function. An instrumented aerial array does not need 146 million injections. It needs a few hundred passes of known objects at times the analysis team does not know, and the objects are commodity drones and weather balloons.

Frontier Who collects: the office holding the data, the civilian observatories, and everyone downstream of an argument that has run since 1947 without a denominator. AARO’s fiscal year 2025 report received 319 new cases and moved 191 of them into an active archive for want of usable data, so the unresolved pile is dominated by a data-quality residual whose size nobody has quantified. The gain is symmetric, which is unusual: a measured misclassification rate constrains the credulous reading and the dismissive one in the same stroke. Handwave The step this brief refuses throughout is any inference from a residual to a cause. A calibrated residual bounds how much room is left; it says nothing whatever about what occupies it, and 5 lists the reasons that room is larger than the instruments suggest.

An instrumented unidentified-anomalous-phenomena survey is a ground station that watches a fixed patch of sky continuously, records what moves across it, tries to match each track to a catalogued source, and reports what is left. The subject of this brief is the last clause. The hard engineering is not sensitivity — uncooled longwave infrared cameras and consumer CMOS sensors already see more than the atmosphere usually contains — but calibration and provenance: knowing what the instrument does to an object before the analyst sees it, and knowing that the bytes the analyst sees are the bytes the sensor produced.

Established The architecture is converging across programmes. The Galileo Project’s observatory at the Harvard College Observatory pairs an all-sky longwave infrared array with visible cameras, passive radio, acoustic microphones, a magnetometer and a weather station; its hardware series in the Journal of Astronomical Instrumentation treats the observatory concept, object localisation, passive multistatic radar and multi-band acoustics as separate papers. AARO’s prototype GREMLIN is the government analogue, a transportable multi-sensor package that in 2024 began a ninety-day pattern-of-life collection at a national-security site. Norway’s Project Hessdalen has run an automatic measurement station since the mid-1990s at Ostfold University College. Philippe Ailleris catalogues fifty-two instrumented field studies over 1947–2023, about 46% in North America and 31% in Europe, peaking in the 1970s and 1980s, nearly vanishing in the 1990s and resurgent since 2000.

Established The pipeline has five stages and the ambiguity is manufactured in the last two. Detect a moving source; link detections into a track; associate the track with an external catalogue — ADS-B transponder returns for aircraft, orbital element sets for satellites, flight-plan and weather data for balloons; classify what is left; report the residual. Stage three is where most identifications happen and stage three is exactly where the catalogues are incomplete. Not every aircraft squawks ADS-B, and almost no drone, bird, insect or free balloon appears in any catalogue at all. A track that fails association is not anomalous. It is uncatalogued, which is a different and much larger class.

Frontier The quantity that makes a survey a survey is the selection function, and it is the quantity this field does not publish. In survey astronomy a selection function states, for an object of given properties, the probability the pipeline detects it, measures it correctly and assigns it to the right class. It is measured by injection and recovery: put objects of known properties into the real data stream, run the unmodified pipeline, count what comes back. Without it a source count is uninterpretable, because the difference between “the sky contains N objects” and “the pipeline finds N objects” is exactly the selection function. Speculative Every other observational field treats this as a precondition for publication; instrumented UAP surveys have so far treated it as a refinement.

Established “Unidentified” is a label produced by a pipeline, not a property observed in the sky. It is the conjunction of five things: the object was above detection threshold; the tracker linked it; no catalogue matched it; a human reviewer could not name it; and nobody later found the mundane explanation. Each has a failure rate, and multiplied they set a floor on the residual that has nothing to do with the phenomenon. The height of that floor is unmeasured in every published instrumented survey this brief could obtain.

Established The boundary with neighbouring briefs is drawn on the instrument, not the hypothesis. Interstellar Archaeology owns the technosignature search as a programme — what has been looked for, with what instruments, over what fraction of the search space. This brief owns a ground sensor pointed at the troposphere, watching a population dominated by birds, balloons, aircraft, drones and satellites, where the whole problem is telling those apart from each other. Biosignature Standards owns the analogous problem for remote-sensing chemistry, and the structural similarity — a detection claim turning on a false-positive budget nobody has measured — is why both briefs exist.

Established The anchor result exists, is peer-reviewed, and says less than the coverage of it suggested. Dominé and sixteen co-authors, Commissioning an All-Sky Infrared Camera Array for Detection of Airborne Objects, Sensors 25(3):783 (2025), report the first system-performance baseline for the Galileo Project’s infrared array: eight uncooled FLIR Boson 640 longwave cameras at 640 × 512 pixels, band 7.5–13.5 µm, seven hemispherical units at f/1 with 50° × 40° fields of view and optical axes 30° above the horizon, plus an f/1.1 zenith camera at 95° × 72°. Extrinsic calibration is done against aircraft positions from ADS-B broadcasts, a genuinely good trick: the sky provides its own surveyed targets.

Established The headline numbers, stated together. Over five months the array reconstructed about 500,000 trajectories. A deliberately crude outlier search on the sinuosity of the reconstructed two-dimensional tracks flagged about 16% of them, roughly 80,000. Manual review reduced that to 144. The paper’s words on the 144 are worth quoting exactly, because they are routinely dropped: the trajectories “remain ambiguous, which are likely mundane objects but cannot be further elucidated at this stage of development without information about distance and kinematics or other sensor modalities.” Folding in systematic uncertainty, the authors quote an upper limit of 18,271 outliers over the interval at 95% confidence.

Established The two efficiency numbers in the same paper are the most important result in the instrumented-UAP literature and the least cited. The array has an acceptance rate — the fraction of in-range aircraft passing through the effective field of view that get recorded at all — of 41% for ADS-B-equipped aircraft, and a mean frame-by-frame detection efficiency of 36%. A system missing three in five aircraft it should see, and two in three frames of those it does see, is not a system whose null result can be read off its residual count. The team reports both figures openly, which is why this brief can make the argument at all.

Frontier The Galileo team did run an injection-recovery study, and it does not answer the question this brief asks. The paper evaluates tracking against synthetic three-dimensional trajectories — straight, curved, and piecewise with inflection points — and reports purity and efficiency for the tracker. That measures whether the association stage reconstructs a path it was given. It does not measure the probability that a real, catalogued, thoroughly mundane object arrives at the end of the pipeline labelled “ambiguous”, because synthetic trajectories are not real objects passing through real optics, weather, compression and human review. That second quantity is what converts 144 into a bound, and it is the gap 7 is built around.

Established What the array actually finds, by category, is a catalogue of the mundane. The manual review classes named in the paper are flocks of birds, single birds, the Moon, airplanes, clouds and helicopters. That list is the base rate made visible, and it is the correct prior for any single unexplained track. Established The programme is also extending the sensor suite rather than the claim: Vervelidou and colleagues describe a geomagnetic variometer station with a noise floor at or below 10 pT rms per root hertz at 1 Hz, sampling at 1,612.9 Hz over a 100 µT range, targeting the roughly 10 Hz signals reported anecdotally during sightings. Speculative That such an association exists at all is a premise resting on testimony rather than measurement, and building an instrument to test it is the right response.

Established The government record is quantified and its shape is consistent across two years. AARO’s fiscal year 2024 report covers 1 May 2023 to 1 June 2024: 757 reports received, 485 inside the window and 272 from 2021–2022, against total holdings of 1,652 as of 24 October 2024. Forty-nine cases were resolved with a further 243 recommended for closure; 444 “lacked sufficient data to facilitate analysis and were placed in the Active Archive”; 21 merited further analysis by intelligence and science-and-technology partners. The fiscal year 2025 report covers 2 June 2024 to 30 May 2025: 319 reports received, 114 of them resolved alongside 256 older cases for 370 resolutions, 9 referred onward, and 191 moved to the active archive.

Established Every resolved case, in both years, resolved to something ordinary. The fiscal year 2025 report states that “all resolved cases are attributable to prosaic objects such as balloons, satellites, birds, aircraft, and unmanned aerial systems (UAS)”, with one commercial rocket launch and one manned jet pack among them; the fiscal year 2024 report states that “to date, AARO has discovered no evidence of extraterrestrial beings, activity, or technology.” Frontier Hold onto the asymmetry: the resolved set is uniformly prosaic, and the resolved set is not a random sample of the reports. It is the subset with enough data to work on, which is precisely the subset most likely to resolve.

Established AARO says plainly that its limiting factor is data quality, not phenomena. Fiscal year 2025: “A lack of timely and actionable sensor data continues to constrain AARO’s ability to resolve cases.” Fiscal year 2024: “The majority of AARO’s case holdings remain unresolved due to a lack of data needed to further analysis.” Director Jon Kosloski, at the November 2024 media roundtable, put the residual at “less than 3.5% of our cases” retaining characteristics that might indicate a breakthrough technology, and said “over 900 reports lack sufficient data for scientific data for analysis and are retained in our active archive.” In the Naval Postgraduate School interview he is blunter: military sensors “aren’t general purpose scientific instruments”, which leads to “incorrect initial assessments.”

Established AARO’s published case resolutions are the best worked examples of sensor-artefact analysis in the open literature. For the 2015 “GoFast” Navy video, AARO reconstructed the aircraft track across all possible headings with historical wind data, put the object at approximately 13,000 feet with high confidence, and derived a true speed between 2 m/s (5 mph) and 41.3 m/s (92 mph) depending on wind geometry, with pixel analysis indicating one metre or less — “comparable to a small drone or bird.” The stated mechanism: “The object’s apparent high speed is attributable to motion parallax.” The office “cannot definitively identify the object, but it displayed no anomalous performance characteristics.”

Established Two further resolutions show the same arithmetic. A December 2018 shortwave-infrared clip from a military uncrewed aircraft over Mount Etna showed an object reported at 345 mph; AARO assessed roughly 24 mph (39 kph) moving west to east with the wind, about 0.3 m across at about 30 km, and concluded with moderate confidence that it was a balloon and high confidence that it did not behave anomalously. A 23 October 2017 full-motion video from Al Taqaddum, seventeen and a half minutes long, resolved with high confidence to a cluster of fully and partially inflated balloons at 850–2,200 feet moving east to west at 4–14 mph, dangling strings visible in several frames. In all three the anomalous impression came from the observing geometry and the compression, not from the object.

Established AARO has published the artefact mechanisms itself. Its 2024 information paper on forced perspective and parallax states that “because of parallax, stationary objects can appear to have motion, and slow-moving objects can appear to move very fast”, and that forced perspective “can cause large, faraway objects to appear smaller and closer than their actual size and position — or vice versa.” The fiscal year 2024 report adds the sensor-side modes: birds are “commonly misidentified as UAP due to sensor artifacts resulting from compression and pixilation that often renders the object as an amorphous blob or orb”, and “electro-optical/infrared sensor glare can also cause distorted pixilation of the object’s true shape.” A separate paper argues that satellite reflection “can explain some of the UAP reports received by AARO”, working through a 9 October 2022 pilot report near Gallup, New Mexico attributed to Starlink flares. Frontier That paper quantifies nothing: “some” is never converted into a count, which is the same missing denominator in the opposite direction.

Established The civilian expedition record is thin, honest and small. Szydagis and colleagues report the first UAPx field expedition: one week in July 2021 in the Catalina Channel off Avalon, California, yielding about one hour of triggered visible and night-vision video, over 600 hours of untriggered far-infrared video and 55 hours of background radiation measurement. One event survived, “a dark spot in the visible/near-IR camera possibly coincident with ionizing radiation that has so far resisted prosaic explanation.” The paper offers 3-to-5-sigma conventions and warns against “confirmation bias toward mundane / speculative conclusions” in both directions. Ailleris records the same expedition from outside: anomalies detected, “none definitively classified as true anomalies”, and a recommendation to use “at least two sensors of each type.”

Frontier The Ukrainian military dataset is the largest recent operational sample and it demonstrates the identification problem rather than solving it. Bilyk and Nikolaiev of the Defence Intelligence Research Institute in Kyiv describe cameras, night-vision devices, telescopes, radio reconnaissance and radar feeding a two-stage identification process, with a crowdsourcing application carrying more than 800,000 users by late 2024; six documented 2023–2024 cases give objects 0.3–32.4 m across at 7.3–250 m/s and 20–300 m altitude. Speculative They classify by “anomaly factors” — high speed, instantaneous acceleration, no visible propulsion — the exact quantities a single passive sensor at unknown range cannot measure.

Established The field’s own survey names calibration as an open problem and then argues it should not block fieldwork. Ailleris’s ten limitations include inadequate familiarity with specialised instruments, uncertainty over whether projects paid sufficient attention to calibration, and the finding that results “were not consistently published, even when they were negative, limiting knowledge-sharing.” His own position is explicit: “While instrument calibration is generally important and should be diligently carried out, the absence of perfectly calibrated instruments should not deter field work.” Frontier That is defensible about getting started and indefensible about publishing a residual count, and the distinction is this brief’s argument.

Established The comparison case is not aspirational; it is routine and enormous. Anbajagane and colleagues, reporting Dark Energy Survey Year 6 results with the Balrog tool, injected 146 million synthetic sources into pixel-level on-sky images across the full 5,000-square-degree footprint, processing 10,169 coadd tiles. The recovered biases would be invisible without the exercise: 3–6 millimagnitudes in riz brighter than magnitude 20, 12 in g, about 30 for 23 < m < 24 from sky-background errors, rising to roughly 500 at magnitude 25, with stellar contamination below 1% across 18 < m < 25. The point is not the numbers. It is that a survey measured its own selection function, transfer function and contamination fraction before it reported a result.

Established Gravitational-wave astronomy went further and blinded the exercise. During the 2009–2010 joint LIGO–Virgo run a few selected collaborators were permitted to secretly insert simulated signals. One was injected at 06:42:22.955 UTC on 16 September 2010, designated GW100916 and nicknamed the Big Dog. The collaborations analysed it for roughly six months, wrote three scientific papers, prepared public data and outreach material, and passed an independent Detection Committee review. The envelope was opened on 14 March 2011 at a joint meeting in Arcadia, California; the signal was an injection and the draft was never published. The purpose was to “stress-test the full procedure and uncover problems that could not be found in other ways.” Advanced LIGO later built hardware injection into the instrument itself, physically displacing test masses for end-to-end validation, documented in Physical Review D 95, 062002 (2017).

Established The people closest to the government data say the same thing in professional venues. The Naval Postgraduate School’s 2025 CTX special issue on UAP opens with Vice Provost Michael Hesse writing that “scientific inquiry — falsifiable hypotheses, calibrated measurement, uncertainty quantification, reproducibility — is the essential tool for further progress”, and naming the gaps as “inconsistent metadata standards, limited sensor fidelity, uneven analytic frameworks, and cultural hesitancy in reporting.” His formulation is the cleanest in print: “Complex phenomena demand measurement. Measurement demands instrumentation. Instrumentation demands calibration. And analysis demands rigor.”

Established The provenance chain is broken in a specific, documented way. Randy Bostick, former science advisor to AARO, writes that determinations such as “it had a plasma field” or “it exhibited an advanced form of propulsion” “are not data — they are interpretations”; that many videos submitted to AARO “have undergone image processing”, where “contrast enhancement or edge detection applied to benefit the sensor’s primary mission ... makes it nearly impossible to estimate target brightness or even size”; that single passive optical sensors cannot recover range, giving “poor calculations of UAP speed, acceleration, and size”; and that insects near the lens produce “pinpoints of light often mistaken for distant, erratically moving craft.” He notes that historical projects could not save raw instrument data, which forecloses independent verification permanently, and counts approximately 12,618 US Air Force UFO reports over 1947–1969. Frontier The commonly quoted count of Blue Book “unidentified” cases is not carried here: this brief could not verify it against a primary source, and an uncalibrated residual is the exact quantity under examination. Established The archive itself is now statutory, accessioned into National Archives Record Group 615 under the fiscal year 2024 National Defense Authorization Act alongside the Blue Book files in Record Groups 64 and 341.

Frontier The central open question is numerical and nobody has attempted it: what is the probability that a catalogued, mundane object traverses an instrumented array and emerges labelled ambiguous? Call it the false-ambiguity rate. If it is 10⁻⁴ per track, then 144 ambiguous tracks out of 500,000 is roughly 50 expected from misclassification alone and the residual is mostly instrumental. If it is 10⁻⁶, the residual is mostly objects the system genuinely could not characterise — still not a claim about their nature, but a different research programme. The two hypotheses differ by two orders of magnitude and the published record cannot distinguish them.

Frontier Whether better instruments shrink the unidentified fraction or grow it is genuinely contested. The optimistic reading is that range, kinematics and multi-band spectra resolve everything the current arrays cannot. The pessimistic reading is structural: a more sensitive system detects fainter, smaller and more distant objects, and identification difficulty rises fastest exactly there, so the residual count grows even as the residual fraction falls. Speculative The test that separates them is cheap: run the same pipeline at two detection thresholds on the same archive and see which way the ambiguous count per thousand tracks moves.

Frontier Range is the physics bottleneck dressed as a frontier question. A single passive sensor measures angles and irradiance. Size, speed and acceleration all require range, and range from a monocular sensor requires a known object size, a known trajectory model, or a second station. Every dramatic kinematic claim in this subject — instantaneous acceleration, hypersonic transit, motion against the wind — is a range claim in disguise. AARO’s Mount Etna resolution converts a reported 345 mph into an assessed 24 mph purely by fixing range and wind: a factor of fourteen from one geometric constraint.

Frontier Is 3.5% a phenomenon or a data-quality floor? AARO’s director gives “less than 3.5% of our cases” as the fraction retaining potentially interesting characteristics after analysis, while reporting that over 900 reports lack sufficient data for any analysis at all. Those are different populations and the relationship between them is unpublished. A residual computed over cases that had enough data to analyse is a different number from one computed over all reports, and neither is quoted with the other. Speculative The data-quality hypothesis makes a specific prediction — the anomalous fraction should fall with sensor quality inside AARO’s own holdings — and GREMLIN’s controlled collection is the first dataset that could test it.

Speculative Event-based sensing is the most interesting hardware proposal on the table and its calibration problem is unsolved. Hagopian and Shank of Sandia National Laboratories describe sensors in which “each pixel acts as an independent, asynchronous optical change detector”, with 100-microsecond response, dynamic range above 120 dB, “1–5 percent of the data volume a traditional framing camera produces”, and no motion blur by construction. Frontier The difficulty is that the photometric meaning of an event stream is not that of a frame: there is no established radiometric calibration chain for these devices, so an event-based array would trade one uncalibrated residual for another unless injection is designed in from the start.

Frontier Does human review help or hurt, and by how much? The Galileo pipeline reduced roughly 80,000 machine-flagged outliers to 144 by manual examination, a cut of about 99.8%. Human reviewers carry priors, fatigue and, in this subject, strong expectations in both directions; Ailleris notes that participants in instrumented campaigns have “shown high levels of suggestibility.” Speculative The review stage is itself a detector with an efficiency and a false-positive rate, and no published survey has measured either. Frontier Nor is there a shared rubric for what counts as a resolution: AARO attaches confidence language to its case resolutions, Galileo applies descriptive reviewer labels, and Hessdalen, UAPx and the Ukrainian programme each use their own, so residual fractions from different programmes are not comparable quantities.

Established Range determination from a single passive sensor is the hard constraint and everything else is downstream of it. Bostick states it directly: passive optical sensors on phones or cameras cannot reliably calculate distance, which yields “poor calculations of UAP speed, acceleration, and size”. The fix is geometric and well understood — two or more stations with a known baseline and synchronised clocks — and it is the recommendation UAPx reached from its own data: at least two sensors of each type.

Established Uncooled microbolometer arrays are cheap, deployable and radiometrically awkward. The FLIR Boson devices in the Galileo array need periodic non-uniformity correction and drift with ambient temperature; the commissioning paper reports a temperature dependence that is uneven through the day. An instrument whose photometric zero point wanders with the weather cannot report radiant intensity, and radiant intensity is one of the few quantities that could separate a bird from a drone at unknown range.

Established The duty cycle is the least glamorous bottleneck and the largest. Acceptance of 41% and frame-level detection efficiency of 36%, falling to roughly 15% in poor visibility, mean most of what crosses the field of view never enters the dataset. Every statement about what the sky contains is therefore a statement about what got through a narrow, weather-dependent gate.

Established The association catalogues have holes exactly where the interesting tracks are. ADS-B covers most transponder-equipped civil aircraft and not military aircraft operating dark, gliders, most small drones, or any free balloon. Orbital element sets go stale between updates and constellation satellites manoeuvre. Birds and insects appear in no catalogue anywhere. A pipeline that calls everything unmatched “unidentified” is measuring catalogue coverage.

Established The optical artefact modes are enumerated, and each produces a signature that reads as anomalous. The AARO field guide written by J. Kevin Ryan, the office’s chief of operations, names autofocus hunting, “rolling-shutter artifacts (a distortion that happens because many phones and mirrorless cameras don’t record the whole frame at once but read the sensor line by line from top to bottom)”, compression artefacts, digital zoom interpolation, coma and astigmatism. Add the agency’s own electro-optical and infrared glare, out-of-focus point sources taking the shape of the aperture, and insects near the lens. Frontier Each has a known physical cause and none has a published occurrence rate per hour of sky monitoring, which is what a survey needs.

Established Compression destroys the measurement before the analyst sees it. A codec optimised for perceptual quality discards exactly the low-contrast, high-frequency information photometry and morphology depend on. AARO reconstructed the GoFast geometry from “the publicly available compressed video file” and had to model all 360° of possible aircraft heading because heading was not in it. The most-analysed videos in the subject are the ones furthest from raw sensor output.

Established The provenance chain has no custody standard, and the one official document addressing it is a photography guide. Ryan asks observers to preserve EXIF timestamp, GPS coordinates, lens and exposure data, UTC time, elevation, equipment details, a narrative log and SHA-256 hashes. That is the right list, and it is advice to members of the public rather than a requirement on the programmes; no instrumented survey examined here publishes per-frame hashes of its raw data. Content Authenticity Infrastructure owns the general problem of signing sensor output at capture.

Frontier Classification of military sensor data blocks calibration, not merely disclosure. An outside analyst cannot check a military electro-optical resolution without the sensor’s point-spread function, gain tables and pointing accuracy, all of which are performance data. That is why AARO’s resolutions can be persuasive and still not reproducible, and it is a structural reason the government programme cannot supply the calibration the civilian programmes also lack.

Established Nulls are not published, so the denominator cannot be assembled from the literature. Ailleris records that results “were not consistently published, even when they were negative, limiting knowledge-sharing”, across fifty-two projects spanning 1947 to 2023, and lists insufficient funding, logistics and a shortage of consistently available technical personnel as the recurring causes. Belgium’s 1990 Operation Identification — infrared thermal cameras, night vision and optical cameras, four days, nothing detected — is the shape of the missing literature.

Established The first dependency is a base-rate catalogue that does not exist: what a bird, a bat, an insect, a small drone and a weather balloon look like in a longwave infrared all-sky camera, as a function of range, aspect and air temperature. The Galileo review categories — flocks of birds, single birds, the Moon, airplanes, clouds, helicopters — are the beginning of one. Turning them into a classifier prior requires labelled examples with independently known identity, which is the same infrastructure an injection campaign builds.

Frontier The second is a contemporaneous airspace traffic model. Resolving a track requires knowing what is up there: aircraft from ADS-B, satellites from orbital catalogues, and — the missing piece — drones, free balloons and birds, for which no operational feed exists at the resolution a sky survey needs. Orbital Debris and Space Traffic owns the orbital half, where catalogue currency is the central difficulty too.

Established The third is metrological: a traceable radiometric and geometric calibration chain for cheap uncooled infrared arrays and for event-based sensors. Metrology Infrastructure owns how a measurement becomes traceable; the specific missing artefact is a field-deployable blackbody and star-field procedure a volunteer-run station can run monthly without a laboratory.

Frontier The fourth is statistical: inference on a rare-event count when the efficiency is itself uncertain. The Galileo paper does this properly, folding systematic uncertainty into a likelihood-based 95% upper limit rather than quoting 144 as a measurement, and says the method is “applicable to all of our future outlier searches”. That is the correct machinery, and it needs an efficiency input that does not exist.

Established The fifth is sociological and is owned elsewhere on this map. Pathological Science carries the natural history of anomalous-claim episodes and how they resolve; Forecasting Replication and Science of Science carry the publication-bias literature that explains why fifty-two field studies left so little behind.

Frontier The decisive experiment is a blinded injection-recovery campaign: fly objects of known type, position and kinematics past an operating array at times the analysis team does not know, and measure what fraction of them the pipeline reports as unresolved. Nothing else on this list changes the brief’s assessment as much, because nothing else converts a residual count into a bound. The output is one number with a confidence interval — the false-ambiguity rate per track, by object class — and that number is the missing denominator under every “unidentified” figure the subject has produced since 1947.

Established The design is borrowed wholesale and the parts are standard. A flight team independent of the analysis team runs a schedule of passes: a 250 g consumer quadcopter at 0.5–3 km, a 2 kg fixed-wing drone at 3–10 km, a latex weather balloon to 20–30 km, a matt-black low-emissivity target, a near-ambient-temperature target, and a fast crosser. Each pass is logged with GNSS position at 10 Hz into a sealed registry held by a third party. The analysis team runs the unmodified pipeline and produces its normal output, including its ambiguous list. The registry is opened afterwards.

Frontier The statistics set the campaign size and they are undemanding. Measuring a recovery fraction near 50% to within 5 percentage points needs of order 100 injections. Measuring a false-ambiguity rate near 1% to within a factor of two needs a few hundred to a few thousand, which is why the campaign should run for a year rather than a week. Speculative If the Galileo array’s 144 ambiguous tracks out of 500,000 correspond to a false-ambiguity rate of order 3 × 10⁻⁴, a campaign of 3,000 injections would expect about one misclassification and would set a useful upper limit — a tractable year of flying.

Established The campaign has two arms and both are needed. The hardware arm flies real objects through real optics, weather, compression and human review, and is the only thing that measures the whole chain. The software arm injects synthetic sources into raw frames before the pipeline runs, in the manner of Balrog’s 146 million sources, and is the only thing that can run at the volume statistics require. The Galileo commissioning paper already contains the seed of the software arm in its synthetic three-dimensional trajectories; extending those from tracker input to pixel-level injection is an incremental change to an existing codebase.

Established Blinding is the part that cannot be skipped, and the reason is documented rather than theoretical. LIGO and Virgo spent roughly six months analysing GW100916, wrote three papers, prepared a public data release and passed an independent detection committee before the envelope was opened on 14 March 2011 and the candidate turned out to be injected. An unblinded test measures a pipeline; a blinded one measures the pipeline plus the people, and in this subject the people are the stage with the largest and least-characterised effect.

Frontier The second-ranked experiment is a two-station baseline, because it removes the dominant systematic rather than measuring it. Two arrays separated by 1–10 km with GNSS-disciplined clocks synchronised to better than a millisecond give range by triangulation, and range gives size, speed and acceleration. Every headline kinematic claim becomes checkable. UAPx reached this recommendation from one week of data, the Galileo commissioning paper from five months, AARO’s parallax paper from case work.

Frontier The third is a held-out blind challenge across teams. Publish a corpus of tracks with ground truth withheld by a neutral party, invite the Galileo, Hessdalen, Sky360, UFODATA and university pipelines to classify them, and publish every team’s confusion matrix. This is ordinary practice in machine-learning benchmarking and in evaluating computer-aided detection in radiology, and it produces comparability between programmes that currently have none.

Speculative The fourth is the natural experiment already running, if its design permits the inference. GREMLIN demonstrated functionality in a March 2024 test event and then began a ninety-day pattern-of-life collection at a site of national security interest, to establish what normal looks like. A pattern-of-life baseline is a base-rate measurement. Handwave Whether any of it can be published at the resolution a calibration argument requires is a classification question with no technical answer, and this brief does not assume it will settle anything.

Frontier The fifth costs nothing but compute: re-run an existing archive at several detection thresholds and outlier cuts and publish the ambiguous count as a function of both. If the ambiguous count per thousand tracks is flat in threshold, the residual is plausibly physical; if it scales with the number of marginal detections, the residual is the threshold. The Galileo five-month archive is sufficient for this today.

Speculative The sixth is an adversarial-artefact campaign: deliberately produce every documented artefact mode and check what the pipeline calls it. Fly an insect past the lens, defocus a bright source to produce aperture-shaped bokeh, record a fast crosser with a rolling-shutter camera, let autofocus hunt, run a clip through a lossy codec at several bitrates. Each is named in AARO’s own documents as a misidentification mechanism and none has a published detection rate.

Handwave What no experiment here can do is identify an unidentified object. A calibrated survey bounds the residual and characterises the instrument. If a residual survives calibration it is a well-measured ignorance, and the inference from well-measured ignorance to any specific cause requires evidence of a kind no ground array produces. Established This is the structure Anomaly Resolution in Physics documents for physical anomalies: most resolve into systematics, and those that do not are closed by a second, different instrument rather than by more data from the first.

Established The sensor hardware is a solved and cheap problem, which is why it is not the bottleneck. Eight uncooled 640 × 512 longwave cameras, a zenith unit, visible cameras, a passive radio front end, microphones, a magnetometer at 10 pT rms per root hertz and a weather station is a bill of materials a well-funded university group can assemble; the Galileo Project built one on an observatory roof, with published hardware papers covering object localisation, passive multistatic radar and multi-band acoustics separately.

Established Timing and geometry are where the engineering effort has to go. A two-station triangulation baseline needs GNSS-disciplined clocks agreeing to better than a millisecond, boresight orientation known to better than a pixel, and a shared time base in every frame header. At a 1 ms timing error a 100 m/s object smears 0.1 m along track, which is negligible; at 10 ms a 250 m/s object smears 2.5 m, which is the size of the object. The tolerance is set by the fastest thing you intend to measure, and the fastest things in this dataset are the contested ones.

Established Radiometric calibration needs a field procedure, not a laboratory. Uncooled microbolometers require periodic non-uniformity correction and drift with ambient temperature, with efficiency falling to roughly 15% in poor visibility. The answer is a temperature-stabilised reference surface in the field of view plus a nightly star-field or Moon transit for geometric and photometric closure — the Moon already appears in the array’s own detection categories, so the calibrator is free.

Established Provenance is an engineering requirement with an existing standard shape. Hash every raw frame at capture, sign the hash with a device key, store the raw stream unmodified, and treat every downstream product as a derivative with a recorded transformation chain — the list AARO’s own field guide gives to amateurs, SHA-256 included. The machinery is described in Content Authenticity Infrastructure. Bostick’s observation that historical projects could not save raw instrument data is the counterfactual: those datasets are permanently unverifiable, whatever they contained.

Frontier The injection apparatus is the one genuinely new build, and it is small. A sealed-registry service holding GNSS logs under third-party control, a flight team with standard operating procedures and airspace authorisations, characterised targets with measured infrared emissivity, and a pixel-level synthetic-source injector wired into the existing pipeline. Against the observatory it validates, this is a rounding error. Speculative Siting then decides what the residual means: an urban observatory roof sees traffic-dominated sky with good ADS-B coverage, a remote site sees fewer aircraft and worse catalogue coverage, and the Galileo Project’s stated plan of up to ten locations across the United States for up to five years is the scale at which residual statistics become site-resolved and therefore falsifiable.

Established Counter-uncrewed-aircraft sensing is the same instrument with a customer. Persistent ground-based electro-optical, infrared, acoustic and passive radar towers for drone detection are a funded product line, and the detection, tracking and classification stack is common with a UAP array. The transfer runs both ways: their labelled datasets are the base-rate catalogue a sky survey lacks, and a calibrated false-ambiguity rate is exactly what a defence customer needs before trusting an automated alert.

Established All-sky meteor camera networks already do this correctly and are the closest working analogue. They run continuous multi-station optical monitoring, triangulate trajectories from known baselines, calibrate astrometrically against the star field nightly, and publish orbital solutions with uncertainties. The instrumented UAP programmes are solving a problem amateur meteor astronomy solved decades ago, in similar wavelengths, with the same geometry.

Established Transient astronomy contributes the statistical machinery directly. Difference imaging, real-bogus classification and synthetic-source injection to measure pipeline efficiency are the standard toolkit; a published test of the Rubin Observatory difference-image pipeline using synthetic source injection is the direct methodological template, and Balrog is the wide-field photometric one. Mega-Telescopes and Multi-Messenger Astronomy carry the instruments and the multi-instrument confirmation discipline respectively.

Established Radar ornithology supplies the base rate nobody else has. Weather-radar networks have measured bird and insect biomass aloft continuously for decades, with seasonal, diurnal and altitude structure. Given that birds and flocks head the Galileo array’s own category list and that AARO names birds as the archetypal sensor-artefact misidentification, this is the most directly useful adjacent field, and it appears in none of the instrumented UAP papers this brief read. Established Space-object cataloguing supplies the other half: satellite flaring is now a recurring resolution category, and attribution accuracy depends on catalogue currency and manoeuvre knowledge, the working problem of Orbital Debris and Space Traffic.

Established The government programme has a statutory basis, a reporting mandate and a named director. AARO was established by the Deputy Secretary of Defense in July 2022 and is codified at 50 U.S.C. 3373, with reporting procedures at 3373b; it publishes a consolidated annual report, case resolution reports, information papers and reporting-trend statistics, and its records are being accessioned into National Archives Record Group 615. Whatever one thinks of its conclusions, this is the most institutionally solid apparatus the subject has ever had.

Frontier AARO is an interested party and the interest runs toward resolution. (the office being assessed) An office whose mandate is anomaly resolution is rewarded for closing cases and exposed by leaving them open. Its case resolutions are detailed, quantitative and checkable in a way that supports them; its unquantified claims — satellite flaring explains “some” reports — are exactly where the interest shows. Take the arithmetic and decline the adjectives.

Frontier The Galileo Project is an interested party in the opposite direction and publishes against its own interest often enough to be useful. (programme’s own figures) It is privately funded, Harvard-affiliated, led by an astronomer whose public profile is bound to the search for extraterrestrial technology, and it exists because its founders think the question is worth instrumenting. Its commissioning paper nonetheless reports the 41% acceptance rate, the 36% detection efficiency and the sentence calling its 144 ambiguous tracks “likely mundane objects”. A programme publishing the three numbers most damaging to a sensational reading of its own result has earned the citation.

Established Debunking organisations are interested parties too, and the corpus should say so with the same bluntness. (debunking community) Metabunk and allied analyst communities do real, checkable optical-geometry work, and were reconstructing parallax explanations for Navy videos years before AARO published the same conclusions. They also have a reputational stake in every case resolving, which is the mirror image of a community whose product is anomalies. Both are usable; neither is a referee.

Speculative The advocacy ecosystem is large, partly funded by the expectation of a result, and mostly not instrumented. (advocacy) Civilian reporting networks, crowdsourced sighting applications, private foundations and conference circuits generate most of the public volume and almost none of the calibrated data. Ailleris’s fifty-two instrumented projects against a vastly larger literature of reports is the ratio, and his finding that participants have “shown high levels of suggestibility” is a comment on exactly this population.

Established Academic hosting is thin but real and growing. Ostfold University College hosts Project Hessdalen; the University of Wurzburg runs an interdisciplinary UAP effort developing an AI-based camera system; the Naval Postgraduate School published a 2025 CTX special issue with contributions from AARO staff and Sandia National Laboratories; the Galileo Project publishes in the Journal of Astronomical Instrumentation. Frontier The obstacle Ailleris records for the Wurzburg effort — lack of public funding — binds all of it.

Frontier The missing institution is a calibration authority, and no existing body is positioned to be one. What the subject lacks is a neutral party holding the sealed injection registry, defining the shared resolution rubric, running the blind challenge and publishing confusion matrices for every participating pipeline. National metrology institutes do this for measurement and benchmark consortia do it for machine learning; neither has a mandate here, and both government and advocacy bodies are disqualified by interest.

Established Stigma distorts the input distribution in a way no instrument fixes. Hesse names “cultural hesitancy in reporting”, Kosloski calls for “better-trained observers”, and the Naval Postgraduate School editors set out to “encourage stigma-free reporting and dialogue”. A report-driven dataset selects on willingness to report, which correlates with belief, profession and career risk. An instrumented survey is valuable because it does not select on the observer — which is an argument for instrumenting, and also an argument against merging instrumented counts with report counts.

Established Both errors are expensive and they are not symmetric in who pays. Credulity costs public attention, research funding and the credibility of the people who will eventually have to adjudicate a real anomaly. Dismissal costs airspace safety: AARO’s fiscal year 2024 reporting includes uncrewed-aircraft incursions near nuclear infrastructure, and a category of reports treated as folklore is a category nobody triages. The reason to calibrate is that one measurement reduces both.

Established An all-sky array is a surveillance instrument and the paperwork rarely says so. A continuous multi-sensor station records aircraft movements, drone operations, the acoustic environment and, at visible wavelengths and low elevation, people; passive radio and acoustic monitoring raise interception questions in most jurisdictions. None of the instrument papers read here carries a privacy impact assessment, and the sensor suite is close enough to a counter-drone security product that the distinction is a matter of who owns it.

Established Researchers here carry an unusual communication duty, because the gap between what the instrument says and what the public hears is unusually wide. “144 ambiguous trajectories, likely mundane, out of 500,000, from a system that records 41% of the aircraft it should” and “144 unexplained objects” describe the same result; the first is what the paper says. Frontier Press-release discipline is the cheapest available intervention in this field and the one most reliably skipped.

Speculative There is an honesty obligation running the other way, and the corpus should state it. Treating a subject as beneath investigation is itself a claim about prior probability, usually unstated and rarely quantified. The position here is that the instrument question is answerable, cheap and unanswered, and that refusing to answer it is a choice with the same evidential status as overclaiming from it. Philosophy of Science and Pathological Science carry the general form of the argument.

Frontier The durable product of this work is not an answer about objects; it is a continuous instrumental record of the low atmosphere, which nobody currently holds. Weather radar covers precipitation and biomass, air traffic control covers cooperative aircraft, space surveillance covers orbit. The uncooperative, sub-kilometre, sub-metre population between them is unrecorded at survey quality, and a ten-site five-year array would be the first census of it. Frontier The externality scales with the science: that array is also an always-on multi-modal sensor network over populated areas, and if it works the ordinary outcome is that sky monitoring becomes a municipal utility with the governance questions traffic cameras got and less debate.

Established The methodological transfer is the larger claim and it is an old one. An institution that can state, with a confidence interval, how often its detection system calls a known thing unknown is an institution that can adjudicate any extraordinary claim in its domain. That is what survey astronomy bought with Balrog, what gravitational-wave astronomy bought with blind injections, and what radiology bought with reader studies on phantom lesions. A civilisation with many automated detection systems and no measured false-ambiguity rates cannot tell its own instruments from its own discoveries.

Speculative The subject is a live test of whether a scientific culture can investigate a stigmatised question without either capture or contempt. Fifty-two instrumented field studies over seventy-six years produced very little usable data, mostly for institutional rather than technical reasons: no funding continuity, no publication of nulls, no shared standards. The current attempt has statutory backing, private observatories, university hosting and a service-academy journal. Whether it converges on a measured selection function or another round of unreplicated residuals is a fact about institutions.

Handwave The contingent branch, stated as contingent. If a calibrated array produced a residual that survived a measured false-ambiguity rate by a wide margin and was confirmed independently at a second station with range, the consequences would depend entirely on what the object turned out to be, and the most probable answers remain adversary uncrewed systems or unrecognised atmospheric physics. Speculative The extraterrestrial hypothesis is the least probable branch of a branch that has not been reached, and Interstellar Archaeology shows why the search-space argument gives it no support from this direction.

These horizons track the instrument and calibration programme, not the phenomenon — what would be measured and by when, on the funding and institutional trajectory the record shows:

• 10 yr: Frontier Multi-station arrays with triangulated range become the norm rather than the recommendation, and at least one programme publishes a measured false-ambiguity rate with a confidence interval. Frontier AARO’s GREMLIN pattern-of-life collections yield a published base-rate statement for at least one site. Speculative A shared resolution rubric emerges, most plausibly from a service academy or a metrology institute rather than from any advocacy body; Metrology Infrastructure shows what that machinery looks like when it works.
• 25 yr: Speculative Capture-time cryptographic signing becomes standard in sensor hardware for unrelated commercial reasons, and the provenance problem that has defined this subject since 1947 stops applying to new data while remaining permanent for the archive. Speculative Continuous low-atmosphere census data becomes a routine input to airspace management, and the residual becomes a line item in an air-traffic dataset rather than a separate literature. Frontier The historical corpus at the National Archives is fully digitised and reanalysed, settling what the mid-century unidentified fractions actually measured.
• 50 yr: Speculative Either the calibrated residual is consistent with zero anomalous objects at the sensitivity achieved, in which case the subject becomes a chapter in the history of measurement alongside the cases in Pathological Science; or it is not, and the field’s centre of gravity has moved to whatever second instrument can characterise the survivors. Handwave Which branch is taken is not forecastable from present evidence and this brief declines to weight them.
• 100 / 250+ yr: Handwave On any horizon this long the sensing layer over inhabited regions is continuous, multi-modal and machine-adjudicated, and the interesting question is no longer whether something unidentified crossed the sky but whether the adjudication system’s own false-ambiguity rate is auditable by anyone outside it. Handwave The subject of this brief, by then, is a governance problem about instruments rather than an observational one about objects.

• Depends on Metrology Infrastructure for the traceable calibration chain a field-deployable infrared array would sit on, Content Authenticity Infrastructure for capture-time signing of raw sensor output, Orbital Debris and Space Traffic for the satellite catalogue currency a growing share of resolutions rests on, and Pathological Science for the base rates on how anomalous-claim episodes actually resolve. None of those produces the specific missing measurement this brief waits on, which is typed below.
• Requires (not on this map) A blinded injection-recovery measurement of the false-ambiguity rate for an instrumented sky survey — the number that converts an unidentified count into a bound, and which no programme has published. A longwave-infrared signature catalogue for birds, insects, drones and balloons as a function of range and aspect, without which a classifier has no prior and a reviewer has no reference. Multi-station all-sky arrays with sub-millisecond time synchronisation and surveyed baselines, because range is the quantity every contested kinematic claim secretly depends on. A neutral calibration authority holding sealed injection registries and maintaining a shared resolution rubric, since both the government office and the advocacy bodies are disqualified by interest. A standing requirement to publish null results from instrumented aerial surveys, absent across the fifty-two projects catalogued between 1947 and 2023. And sustained decade-scale funding for a continuous census rather than the week-long expeditions that produced most of the existing record.
• Enables a defensible upper limit on genuinely uncharacterisable aerial objects, which is the precondition for any serious treatment of the subject; a labelled, calibrated dataset useful to counter-uncrewed-aircraft sensing and to radar ornithology; and a worked template for calibrating any automated anomaly-detection system whose output is a residual, which is the transferable part.
• Adjacent Interstellar Archaeology for the technosignature search as a programme, Biosignature Standards for the same false-positive-budget problem in remote-sensing chemistry, Anomaly Resolution in Physics for how physical anomalies actually get closed, Multi-Messenger Astronomy for confirmation across independent instruments, and Forecasting Replication with Science of Science for the publication-bias machinery behind the missing nulls.

Speculative “The Galileo Project found 144 unexplained objects.” Established The paper says the 144 trajectories “remain ambiguous, which are likely mundane objects but cannot be further elucidated at this stage of development without information about distance and kinematics or other sensor modalities”. Ambiguous is not unexplained; it says the instrument lacked the range and kinematic information to finish the job. The authors also quote a 95%-confidence upper limit of 18,271 outliers for the same interval, which is the honest expression of how much they do not know.

Speculative “A residual of 3.5% at AARO means something anomalous is out there.” Established The same office reports that over 900 reports lack sufficient data for any analysis, and that 444 cases went to the active archive in one year and 191 in the next for that reason. A residual computed over the analysable subset is a statement about that subset. Frontier The honest form is that under 3.5% of analysable cases retain characteristics AARO has not explained — compatible with the anomaly being entirely instrumental and compatible with it not being, and the published data do not separate those.

Speculative “Modern sensors are so good that anything they cannot identify must be exotic.” Established The opposite is closer to true. A more sensitive instrument detects fainter, smaller and more distant objects, and identification is hardest at the detection limit, so improved sensitivity moves mass into the ambiguous bin. The Galileo array is a good instrument and records 41% of the aircraft it should see, with 36% frame-level detection efficiency falling to about 15% in poor visibility. A sensor good enough to find something is not automatically good enough to identify it; those are different performance curves.

Speculative “Military pilots are trained observers, so their reports carry special weight.” Established AARO’s own director says military sensors “aren’t general purpose scientific instruments” and that this leads to “incorrect initial assessments”, and the office publishes an information paper explaining that “because of parallax, stationary objects can appear to have motion, and slow-moving objects can appear to move very fast”. Training in flying an aircraft is not training in monocular range estimation against a featureless background from a moving platform, a task at which no human is reliable.

Speculative “The GoFast video shows an object moving at extraordinary speed just above the ocean.” Established AARO’s reconstruction puts it at approximately 13,000 feet with high confidence and derives a true speed between 2 m/s and 41.3 m/s depending on wind geometry, with pixel analysis indicating one metre or less, comparable to a small drone or bird. The apparent speed is motion parallax from the sensor platform. The office still cannot say what it was, and says so, which is the correct place to stop.

Speculative “Objects have been recorded moving at hundreds of miles per hour against the wind.” Established The Mount Etna case was reported at 345 mph and assessed at about 24 mph once range and wind were fixed; the Al Taqaddum object resolved to a cluster of balloons at 4–14 mph with dangling strings visible in the video. Reported speed from a single passive sensor at unknown range is not a measurement of speed. It is a measurement of angular rate, and the conversion needs a number nobody had.

Speculative “Orbs and amorphous glowing shapes are a distinct class of object.” Established AARO states that birds are “commonly misidentified as UAP due to sensor artifacts resulting from compression and pixilation that often renders the object as an amorphous blob or orb”, and that electro-optical and infrared glare distorts apparent shape. An out-of-focus point source takes the shape of the aperture, which is why triangular and hexagonal “craft” appear in night-vision footage. Shape at the detection limit is a property of the optics.

Frontier “Debunkers have shown it is all birds and balloons.” Established This is the mirror-image error and the official record does not support it either. AARO’s position is that all resolved cases are prosaic and that the majority of its holdings remain unresolved for want of data. “Everything we could check was ordinary” and “everything is ordinary” are different claims, and the gap between them is the same uncalibrated residual the credulous reading exploits from the other side.

Speculative “A crowdsourced civilian sensor network will settle this.” Established Ailleris catalogues fifty-two instrumented field studies from 1947 to 2023 and ten recurring reasons they did not settle it: funding, logistics, personnel, instrument familiarity, maintenance, suggestibility among participants, unprepared official teams, failure to publish even negative results, absence of collaboration, and uncertainty about calibration. Frontier Distributed low-cost stations such as the Sky360 network address coverage, which was never the binding constraint. The binding constraint is that nobody knows what the network would do with an object it already understood.

Handwave “Unidentified means extraterrestrial.” Established This is the step the whole brief exists to mark. Unidentified is a pipeline output with an unmeasured error rate; extraterrestrial is a causal hypothesis requiring evidence of a kind no ground-based sensor array produces. AARO states that “to date, AARO has discovered no evidence of extraterrestrial beings, activity, or technology”, and the Galileo Project — founded to look for exactly that — describes its own residual as likely mundane. Speculative The search-space argument in Interstellar Archaeology cuts the same way: a survey covering a vanishing fraction of the technosignature space cannot license an inference from a local residual either.

Speculative “Calibration is a counsel of perfection that would stop anyone doing fieldwork.” Established This is the field’s own position, quoted in full because it is honest: “While instrument calibration is generally important and should be diligently carried out, the absence of perfectly calibrated instruments should not deter field work.” Frontier True for starting, false for publishing a residual count. The distinction is not perfection; it is whether the paper reports a number that requires an efficiency the authors did not measure. Balrog did not make DES perfect and blind injection did not make LIGO’s interferometers perfect; both made a published number mean something.

Speculative “The government has the answer and is withholding it.” Frontier The classification barrier is real and has a boring consequence already named here: outside analysts cannot obtain the point-spread functions, gain tables and pointing accuracies needed to check a military sensor resolution. That is an obstacle to reproducibility, not evidence of a hidden answer. Established The observable behaviour — annual counts, case resolutions with confidence language, information papers explaining why the community’s favourite videos mislead — is what a programme finding mostly balloons looks like, and also what a programme wanting the subject to go quiet would do. The evidence does not distinguish those, and nothing here rests on the distinction.

Handwave “Enough data will eventually resolve it.” Established Volume is the one thing this subject already has: over 600 hours of untriggered infrared video from a single week-long expedition, roughly 500,000 trajectories in five months from one array, 1,652 reports in one government database, approximately 12,618 US Air Force reports over 1947–1969. What it has never had is a measured selection function, and adding data without one multiplies the residual and the uncertainty together.