Lurkers in the Dark

by Christopher Scott Carson, J.D., M.A.

Dr. Beatriz Villarroel, the Nordic University

Three independent researchers have confirmed Dr. Beatriz Villarroel’s findings across four distinct evidentiary axes: the physical detections, the statistical architecture, the geomagnetic environment, and the internal coherence of the whole. A fifth confirmation has now arrived, and this one is written into the optics of the telescope itself.

Christopher Scott Carson | Revised and expanded, 9 June 2026

The Telescope as Witness

In the first week of June 2026 a second preprint appeared on arXiv from Ivo Busko, the retired Space Telescope Science Institute software engineer whose March paper furnished the first independent replication of the VASCO transients. The new paper (arXiv:2606.08319) attempts something subtler than the first, and in one respect something more decisive: it enlists a defect of the telescope as a forensic instrument.

The persistent objection to every transient yet reported from the photographic archives, from Hambly and Blair in 2024 through Watters and colleagues this past winter, has been the plate artifact: the dust mote, the micro-hair, the fingerprint, the chemical stain leached over seventy years from a paper envelope into the emulsion it was meant to protect, any of which can counterfeit a star well enough to slip through automated filters. The objection has force precisely because a photographic plate is a chemical object with a long and eventful private life, and because a single-epoch detection cannot, by definition, be re-observed. What Busko has now demonstrated is that there exists a class of evidence which the artifact hypothesis cannot touch, however ingenious its advocates. A blemish knows nothing of the telescope’s optical prescription. A photon that traversed the optics cannot help confessing that it did.

Busko’s new sample comprises eleven transients on plates from the Hamburg Observatory’s Doppel-Reflektor, a 0.6-meter parabolic instrument whose images suffer, usefully for once, from significant coma. Every one of the eleven carries the comatic signature that off-axis light is compelled by the geometry of a parabolic mirror to leave behind: the asymmetric, comet-like figure, oriented toward the center of the plate, scaled in proportion to its distance from that center, and consistent in structure with the genuine stars in its immediate neighborhood. The brightest of them shows emulsion saturation and halation, the backscatter of real light within the emulsion layer, effects that no envelope chemistry can produce. The plate-artifact hypothesis, the skeptic’s first and most durable redoubt, does not survive contact with these images.

Since my essay of April, the field has moved, and in more directions than one. The most substantial skeptical critique yet published, by Watters and colleagues, appeared in late January and has received a detailed response from the Villarroel team, revised at the end of April. A machine-learning analysis of the transient catalog has appeared, with Brian Doherty, whose independent replication I described in April, now among its authors, a change of status that honesty obliges me to record. And now Busko has published a second paper resting on an evidentiary principle entirely distinct from his first. What follows, then, is the essay as it ought now to read: first a review of the literature as it stands in June 2026, with the skeptical case stated at full strength; then the new paper in detail; then the statistical and geomagnetic replications; and finally an honest accounting of where the argument rests, and where it remains genuinely open.

The State of the Question: A Literature Review, 2020–2026

The controversy has a definite architecture, and it helps to see it whole before descending into the particulars. The foundational claim, advanced by Villarroel and her collaborators beginning in 2020, is that the first-epoch Palomar Observatory Sky Survey plates, exposed before any human object reached orbit, contain point sources that appear in a single epoch and never again. The first-order dispute concerns whether these detections are images of light from the sky at all, or merely the detritus of seventy-year-old emulsion chemistry and modern scanners. The second-order dispute, which only arises if the first is resolved in the transients’ favor, concerns what the light was: an unrecognized astrophysical or atmospheric phenomenon, or specular reflection from objects in near-Earth space that predate Sputnik. The literature divides cleanly along these two questions.

Villarroel et al. (2020), The Astronomical Journal. The founding paper of the VASCO project (Vanishing and Appearing Sources during a Century of Observations) cross-matched roughly 600 million objects between the USNO-B1.0 catalog and Pan-STARRS DR1, and identified approximately one hundred point sources visible in only a single epoch in the red band. The project had been conceived to find stars that vanish; what it found instead was a population of lights that appeared once and never returned. The phenomenon was established; its interpretation was deliberately left open.

Villarroel et al. (2021), Scientific Reports. The team reported multiple transients appearing simultaneously within small regions of single POSS-I exposures, and examined whether micrometeorites striking the atmosphere face-on could account for them. The analysis found they could not, removing one of the more economical natural explanations from the table.

Solano, Villarroel and Rodrigo (2022), Monthly Notices of the Royal Astronomical Society. A Virtual Observatory search identified 5,399 unidentified transients in POSS-I data and worked systematically through the conventional astronomical candidates: flare stars, asteroids, variable sources, plate-to-catalog mismatches. A residue survived every conventional assignment. This 5,399-object sample would later become the battleground of the Watters critique.

Hambly and Blair (2024), arXiv:2402.00497. The artifact thesis received its most careful statement. Hambly and Blair observed that the transient image profiles were narrower and rounder than those of neighboring stars, and argued that this pointed toward emulsion defects rather than astronomical sources. The argument deserves to be stated at full strength: photographic archives are saturated with star-like blemishes; the burden of proof lies with the anomaly; and a population of detections that is systematically unlike the confirmed stars on the same plates invites the suspicion that it was never starlight at all. For a year this was the strongest card in the skeptical hand.

Villarroel, Solano and Marcy (2025), arXiv:2507.15896. The optical-physics rejoinder. A sub-second flash recorded on a long-exposure plate does not accumulate the atmospheric seeing, tracking error, and emulsion spread that broaden a stellar image over many minutes; the physics of image formation therefore predicts that a brief specular glint will stamp a tighter, narrower point on the plate than any star of comparable peak flux. The narrowness Hambly and Blair read as evidence of defect is precisely what the flash hypothesis requires. The objection had mistaken a consequence of the hypothesis for a counterexample to it.

Bruehl and Villarroel (2025), Scientific Reports. The first of the two papers that carried the controversy into the general press. Across 2,718 observation days from late 1949 to early 1957, transient detections proved 45 percent more likely on days falling within one day of an atmospheric nuclear test (relative risk 1.45, p = .008), with a weaker association to contemporaneous UAP reports. The paper passed peer review at a Nature-portfolio journal, a fact its critics have been at some pains to discount and its defenders at equal pains to repeat.

Villarroel et al. (2025), Publications of the Astronomical Society of the Pacific. The second headline paper reported cases of multiple transients in geometric alignment on single plates, one alignment reaching 3.9σ, and mapped the full transient population against the theoretical boundary of Earth’s shadow at geosynchronous altitude. Objects in sunlight can glint; objects in shadow cannot. The observed deficit of transients inside the shadow zone registered at roughly 22σ against a hemispheric model and 7.6σ against a more conservative plate-coverage model. The shadow deficit remains, to my mind, the single most stubborn datum in the whole corpus, because it ties the population to the geometry of sunlight at a specific altitude, which no property of emulsion chemistry has any business knowing.

Watters, Dominé, Little, Pratt and Knuth (2026), arXiv:2601.21946. The most serious critique yet published, and one that cannot be waved away as reflexive debunkery, since several of its authors have spent years building instrumentation for the rigorous study of anomalous phenomena and have professional skin in the game of getting such claims right. Working from a heavily filtered subset of the Solano sample (4,866 of the 5,399 objects), they report no shadow deficit, and argue that the alignment and nuclear-timing claims rest on contaminated samples, uncertain plate-time reconstructions, and inadequate object-level validation. Their underlying methodological demand is fair: ensemble statistics computed over a catalog known to contain artifacts inherit the catalog’s impurities, and no individual transient in the sample had been validated as an image of light.

Villarroel, Streblyanska, Bruehl and Geier (2026), arXiv:2602.15171 (v2, 30 April 2026). The response argues that the critique conflates object-level validation with ensemble-level statistical inference, that the aggressively filtered subset is not demonstrably purer than the parent sample (a machine-learning classifier check suggests it carries a higher relative fraction of low-confidence objects), that the null results are presented without error bars or propagation of uncertainties, and that the plate-assignment geometry depends on a cos(Dec) factor whose omission would corrupt the inferred observation times. The response also makes a concession that any honest review must record: the team acknowledges that up to two-thirds of the full 107,875-object catalog may be false positives, while maintaining that the ensemble statistics were designed precisely to extract signal from a contaminated population. Whether one finds that posture rigorous or convenient depends largely on whether one believes the signal exists; which is to say, the exchange ended where such exchanges usually do, with each side’s priors intact. What the impasse made plain was the need for evidence of a different kind: object-level, physical, and independent of the Palomar pipeline altogether.

Bruehl, Doherty, Streblyanska and Villarroel (2026), arXiv:2604.18799. A machine-learning classifier trained on the historical imagery supports the existence of a previously unrecognized class of transient phenomena in the plates. The paper also marks Doherty’s passage from independent replicator to collaborator. His replication, described below, was performed and posted before this collaboration began, and its code remains public; but the reader should weigh his subsequent work as that of a team member rather than an outside auditor.

Busko (2026a), arXiv:2603.20407. The first independent physical replication, described at length in the April version of this essay. Working alone, without institutional backing, Busko searched plates from the Hamburg Observatory’s 1.2-meter Großer Schmidtspiegel using a method that consults no external catalogs at all: pairs of plates covering the same field, exposed in rapid succession, compared internally against each other. From 41 plates he reported 35 vetted transient candidates bearing the same narrow-profile signature that the flash hypothesis predicts. Different telescope, different hemisphere, different emulsions, different scanner, different method; the same signature.

Doherty (2026), arXiv:2604.00056. Independent statistical replication of the nuclear-timing correlation and the shadow deficit, with code written from scratch, environmental confounders controlled, and permutation tests addressing the temporal-autocorrelation objection. Detailed below.

Cann (2026), arXiv:2604.04950. Independent geomagnetic analysis cross-referencing the transient record against the GFZ Potsdam Kp index archive, reporting a monotonic dose-response between geomagnetic disturbance and transient detection rates, and showing that the nuclear-timing signal strengthens rather than dissolves under geomagnetic and lunar controls. Detailed below.

Villarroel et al. (2026), Monthly Notices of the Royal Astronomical Society, 546. A proof of concept for the forward-looking search: using Zwicky Transient Facility data, the team demonstrated that Earth’s shadow at geosynchronous altitude can serve as a natural filter, since any self-luminous object detected inside the shadow cannot be a sunlit satellite or debris. This is the methodological seed of the proposed ExoProbe network.

Busko (2026b), arXiv:2606.08319. The present paper, and the occasion for this revision. Where Paper I argued from image narrowness, a property the skeptics had contested as ambiguous, Paper II argues from coma, a property no artifact can possess except by an improbable conspiracy of accidents. It is object-level validation of exactly the species Watters and colleagues demanded, arrived at from a direction none of the parties to that dispute had explored.

The shape of the argument as of June 2026, then, is this. Nobody seriously disputes that the catalogs contain artifacts; the Villarroel team has conceded as much in print. The live first-order question is whether, beneath the contamination, a residual population of real, on-sky, sub-second optical events exists. The ensemble statistics say yes and the critics say the ensembles are impure; that exchange had reached a stalemate that more statistics seemed unlikely to break. Busko’s second paper bears precisely on this deadlocked question, and it does so without recourse to ensemble statistics at all.

Light Through the Telescope: The Two Busko Papers

Ivo Busko spent decades at the Space Telescope Science Institute writing data-reduction pipelines for Hubble and for the James Webb Space Telescope, instruments whose calibration demands are measured in fractions of a pixel. He understands, as well as anyone alive, how imaging data can mislead and how to ensure it does not. In retirement he turned to the VASCO results as an outside examiner, choosing his own data, his own telescope, and his own method, and publishing his entire pipeline as open source on GitHub, so that anyone with a laptop and Python can repeat the exercise.

His first paper had an instructive weakness, which he reports himself with the candor that distinguishes the whole project. The Großer Schmidtspiegel is a Schmidt camera, and the great virtue of the Schmidt design, its delivery of sharp, round, aberration-free stellar images across a wide field, became a liability for this particular purpose: a significant population of plate artifacts is also sharp and round, sometimes rounder than the stars, and at the level of radial profiles and circularity metrics the two populations cannot always be told apart. The narrowness argument, however sound its optics, operates on contested terrain.

The second paper turns the problem inside out. Rather than seeking a telescope whose images are clean, Busko sought one whose images are informatively flawed. The Doppel-Reflektor’s 0.6-meter parabolic mirror imposes significant coma on off-axis sources: light entering at an angle to the optical axis loses its symmetry with respect to the optical surfaces, and the focal image of a point source is smeared into the small comet-like figure that gives the aberration its name, a bright apex flanked by two symmetric wings, trailing into a tail bounded by a semicircular contour that is in fact a defocused image of the pupil. The figure is lawful in every particular. It points toward the center of the field. Its size grows with distance from that center in a fixed relation. Its structure and its photometry stand in a determinate relationship to the genuine stars around it. An emulsion blemish, to counterfeit it, would need to fake the orientation, the radial scaling, the wing structure, and the consistency with neighbors, all at once, and a population of blemishes would need to fake all of this coherently across the plate.

From the APPLAUSE archive (Archives of Photographic PLates for Astronomical USE, the European consortium digitizing the continent’s plate heritage) Busko assembled 532 Doppel-Reflektor plates forming 407 usable pairs, spanning 1934 to 1957, representing some 145 hours of accumulated exposure over 1,872 square degrees of sky, with October 1957 as cutoff so that no glint in the sample can be blamed on Sputnik or anything launched after it. Plate pairs share field, exposure time, and emulsion; an object present on one plate of a pair and absent from its twin, taken minutes or tens of minutes later, is a transient candidate. The candidates were filtered against SExtractor parameters, cross-checked against the USNO and Gaia catalogs and against POSS-II blue plates, screened for asteroids through the Minor Planet Center (which returns empty for every event), and then subjected to visual vetting against four criteria derived from the physics of coma: alignment toward plate center; visible wings, or at minimum an asymmetry in the correct direction; size and visibility of the coma consistent with field position and brightness; and due allowance for saturation, halation, and reciprocity failure. Each candidate was judged against a control set of neighboring stars within a tenth of a magnitude of its own peak flux.

Eleven transients survived. Their star-equivalent magnitudes run from 11.9 up to a remarkable 7.1, and that qualifier matters: the magnitudes assume the source shone for the full exposure, as a star does. If the events were brief, the true brightness was far greater. Busko offers the arithmetic: a tenth-magnitude transient lasting one second on a fifteen-minute plate corresponds to an actual brightness near magnitude 2.6, which is to say a light that, for its instant, rivaled the brighter stars of the night sky. Three such events, the brightest in the sample, appear on a single plate exposed on 4 March 1951. The brightest of the three shows saturation of the emulsion’s response and halation, the halo of light backscattered within the emulsion layer itself. These are signatures of photochemistry, of real light acting on real silver halide, and they are not within the repertoire of envelope stains.

One analysis in the paper deserves particular attention, because it shows the comatic method doing work that no profile statistic could do. For every star on the plate in question, the USNO catalog position falls at the apex of the comatic wing structure, the point of peak intensity, exactly where the optics put it. Busko overlays the catalog positions on the brightest transient and finds the two faint stars nearest its location sitting off-center, displaced from the apex where a true stellar counterpart would sit. The transient is therefore not a misbehaving image of either star. The aberration pattern functions here as a coordinate system internal to the plate, and the transient fails to occupy the seat reserved for any cataloged occupant.

Then there is the clustering, which is the strangest feature of the sample and the one Busko handles most carefully. All eleven transients fall in just two small regions of sky: seven in a patch near RA 2h17m, Dec +57°, and four near RA 13h14m, Dec +18°. All eleven fall between 1949 and 1953, though roughly half the plate pairs in the sample lie outside that window, some fifty pairs before 1949 and seventy after 1953, all of them empty of detections. Within the clusters the temporal structure tightens further. On 9 December 1950, a vanishing transient on one plate is followed thirty minutes later by another vanishing transient a degree and a half away. On 8 February 1951, in the same sky field, a vanishing transient is followed twenty minutes later by an appearing one, 35 arcminutes away along a nearly east-west line, the two events similar in shape and brightness. And the field near RA 13h14m produces a solitary transient on 20 April 1949, then, four years later on 19 April 1953, three transients packed within roughly 45 arcseconds of one another, in that same field.

The 1953 triple invites a specific objection, which Busko raises against himself: the three images are appreciably narrower than stars of comparable peak flux, and a critic could on that ground refuse them the status of transients. His reply draws on the same optics that answered Hambly and Blair. The temporal power spectrum of atmospheric seeing falls steeply with frequency, as f to the minus eight-thirds, so most of what a long photographic exposure records as seeing is slow image wander; a sub-second flash largely escapes it and prints sharper than any star can, while reciprocity failure at such short effective exposures mutes the faint comatic wings. The three images nonetheless carry an asymmetry of common magnitude and, decisively, of common orientation, the orientation of the coma exhibited by the neighboring stars. Three independent blemishes conspiring in both respects strains credulity considerably more than three brief flashes does.

Now the caveats, which I give in full because Busko gives them in full, and because an essay that softened them would forfeit the right to be believed about anything else. The identifications are fundamentally visual; the quantitative, automated analysis is promised for a forthcoming paper with data from additional telescopes. The pipeline was designed around vanishing transients and was only intermittently run in reverse, so appearing transients are undercounted. The sample of eleven is too small, and the plate coverage too patchy, for the spatial and temporal clustering to be assigned statistical significance; Busko concedes the groupings could be sampling effects, while noting that this reading carries its own awkward freight, since it implies the transients are commonplace and will be found more or less wherever one looks. He also floats, without pressing it, the possibility that single moving objects produced multiple grouped detections.

And the nuclear associations, in this sample, are genuinely equivocal, a fact I report with some relish precisely because Busko prints it in the body of his paper rather than burying it in a footnote. Three groupings are suggestive: the two weak transients of 8 February 1951 came two days after Operation Ranger fired the last of its five shots at the Nevada Proving Ground; the bright triple of 4 March 1951 followed Ranger’s close by about a month; the 45-arcsecond triple of 19 April 1953 came one day after Operation Upshot-Knothole fired the sixth shot of its series. But 1950 saw no nuclear tests anywhere on earth, and the sample contains two transients from that December; and the sole 1949 transient precedes the first Soviet test, the only detonation of that year, by some four months. A lesser author would have reported the three hits and omitted the two embarrassments. The embarrassments are in the paper, and they are why I trust the hits.

What, then, does the paper establish? Its claim is narrow and, within its narrowness, very strong. These eleven images were made by light that traversed the optical train of the telescope. Busko says plainly that the data do not by themselves establish the physical origin of that light, and the conscientious reader should hold the line exactly where he draws it. The confirmation this supplies to Villarroel’s program is therefore foundational rather than terminal: it secures the premise on which the entire edifice stands, the premise Hambly and Blair denied and Watters declined to grant, namely that at least some of the transients are images of the sky. Whether the sky contained machines is the second-order question, and it is to the evidence bearing on that question that we now turn.

The Doherty Paper: Statistical Replication

Busko’s replications answer the question: are the transients real? Brian Doherty’s answers a different one: are the patterns real? Doherty, a data analyst in financial services working from Dallas with no institutional affiliation in astronomy, obtained the original dataset from Bruehl and Villarroel, wrote his own analysis code from scratch, and tested whether the statistical claims survived independent scrutiny. (I repeat the disclosure made above: Doherty has since joined the Villarroel group as a co-author on the 2026 machine-learning paper. The replication described here predates that collaboration, and the code remains public on GitHub for anyone inclined to check the checker.)

His chi-square analysis reproduced the original result precisely: a relative risk of 1.45 (p = 0.011), meaning that observation days within one day of a nuclear detonation show a 45 percent higher transient detection rate than other days. He then extended the analysis with a negative binomial regression controlling for precipitation, lunar illumination, and cloud cover, assembled from historical NOAA station data and Astropy ephemerides. After adjustment the association strengthened: an incidence rate ratio of 1.80 (p < 0.0001). The most striking number emerges from a partition the hypothesis itself predicts. If the transients are solar reflections, the nuclear-timing signal should concentrate among transients in sunlit sky positions, where specular reflection is physically possible. Restricted to sunlit transients, the incidence rate ratio nearly quadruples, to 3.98 (95% CI 3.475–4.562, p < 0.0001), statistically compatible with the original paper’s 3.53. The nuclear timing signal lives where the sunlight lives.

The permutation test addresses the autocorrelation objection directly. A skeptic might argue that any cluster of roughly 350 dates drawn from a temporally structured record would show elevated rates. Doherty shuffled the nuclear-window labels across all 2,718 observation days ten thousand times; the observed relative risk of 1.447 exceeded 99.4 percent of the null distribution (p = 0.006), whose mean was 1.009 and whose 95 percent interval was [0.708, 1.319]. Block permutations at 30-day and 60-day scales, which preserve within-block autocorrelation, remained significant (p = 0.036 and 0.034). The specific test dates carry information; random dates do not reproduce the effect.

His shadow analysis confirmed the PASP result independently. Of 107,875 transients, 499 (0.46 percent) fall within the umbral cone at geosynchronous altitude; restricted to sources within two degrees of plate center, eliminating edge effects, the figure is 142 of 31,525 (0.45 percent). Both sit well below the geometric expectation of roughly 1.4 percent and below the conservative plate-coverage expectation of 0.78 percent, and their near-identity across the full and central samples tells against any plate-edge explanation.

The Cann Paper: Geomagnetic Replication

Kevin Cann, an independent researcher in California, continuing the pattern set by Busko in Baltimore and Doherty in Dallas, cross-referenced the transient record against the GFZ Potsdam geomagnetic archive, specifically the planetary Kp index, the standard measure of global geomagnetic disturbance (arXiv:2604.04950; reproduction script at osf.io/8ryhk). The motivation is direct. Objects at or near geosynchronous altitude inhabit the outer radiation belts, whose particle populations are dynamically coupled to the solar wind; if the transients are reflections from physical objects in that regime, their detectability might track the geomagnetic environment. Emulsion defects are indifferent to space weather. There is no pathway by which a Kp 9 storm should suppress the rate at which film grain produces spurious detections on plates exposed at Palomar.

The detection rate tracks Kp, and it does so with a dose-response gradient. Partitioning the observation days into five Kp intensity bins (2,039 days after lunar-phase correction), Cann finds the transient detection rate descending monotonically from 17.4 percent during geomagnetically quiet periods to 2.4 percent during severe and extreme storms; a Cochran-Armitage trend test returns Z = −3.391, p = 0.0007. The study period spans the decline of Solar Cycle 18 and the rise of Solar Cycle 19, the most active on record, so disturbed conditions characterize nearly a quarter of the dataset; this is no rare-event confound.

The analysis also disposes of a confound any careful reviewer would raise: if nuclear test dates happened to cluster in geomagnetically quiet periods, the nuclear-timing correlation could be an artifact of space weather. They do not. Test days lean, if anything, toward the disturbed regime (31.1 percent at Kp ≥ 5, against 21.9 percent of baseline days), the regime in which detections are suppressed. In multivariate logistic regression with Kp and lunar phase as covariates, the nuclear association strengthens, from 2.6σ (odds ratio 1.53) to 3.1σ (p = 0.002, odds ratio 1.70). A correlation produced by confounding should attenuate when the confounder is controlled. This one sharpens. Each successive analytical layer applied to the nuclear signal, Doherty’s confounder adjustments, his permutations, Cann’s geomagnetic controls, has returned a more significant result than the layer before, and that pattern is itself diagnostic: artifacts degrade under added scrutiny, and real effects do not.

What the Nuclear Correlation Does and Does Not Assert

A word is owed here on the logical structure of the nuclear-timing result, because it has been misread in opposite directions by critics and enthusiasts alike. Critics have treated it as a causal-origination claim, as though the thesis were that detonations somehow summoned the transients into being, a thesis easy to ridicule and easier to refute. Enthusiasts have occasionally written as if the tests were a beacon drawing visitors from afar. Neither reading is the claim. The claim is conditional and modest: when tests occurred, detections rose. A relative risk of 1.45 over a nonzero base rate was never an origination claim; the majority of transients fall outside every test window, and the correlation asserts only that something about the windows elevated the rate. It is a claim about response.

And the claim about response carries a consequence that, so far as I can find, no party to the dispute has stated plainly. A detection-rate response within a day of a Nevada detonation is causally available only to an observer already in place. Nothing outside the solar system can register a test and respond, or even retask a sensor, on a twenty-four-hour clock; special relativity forecloses it absolutely, and even an asset parked at the heliopause sits some four months distant by light alone. If the correlation is real, then whatever produced the elevated detection rates of February 1951 and April 1953 was resident in near-Earth space before the shots that occasioned them. The tests cannot have attracted anything from outside the system on the timescales the data record. They can only have been noticed by something already here, whose attention sharpened afterward. The very lag structure that establishes the correlation establishes the residency; attraction in the interstellar sense is excluded by the same evidence that the correlation rests on.

Nor is a monitoring system keyed to nuclear detonation an exotic postulate. From geosynchronous altitude an atmospheric burst is almost insultingly easy to detect: the characteristic optical double flash, visible across an entire hemisphere, is the signature around which the Vela satellites’ bhangmeters were later designed. A surveillance architecture that watches a developing civilization and prioritizes the most unambiguous technosignature such a civilization can emit is what we ourselves built within fifteen years of Trinity. The hypothesis attributes to the observers nothing we did not shortly attribute to ourselves.

Read this way, Busko’s discordant cases change sides. Two transients in December 1950, a year without a single detonation anywhere on earth, are fatal to any model in which the tests are the sole occasion of activity, and are exactly what a standing surveillance presence looks like: attention that waxes with provocation and persists without it. The solitary transient of April 1949, predating the first Soviet test by four months, reads the same way. The counterexamples to “tests cause transients” are confirmations of the claim actually on the table, that the transients have a baseline and the tests modulate it.

One wrinkle belongs on the record for completeness. The Hamburg sample is the only dataset reaching back before the atomic era, and its fifty-odd plate pairs from 1934 to 1949 are empty of detections. The sampling is patchy and the totals are small, so this carries little independent weight; but a careful critic will notice that the one pre-Trinity window we possess shows nothing. The honest position is that the data cannot distinguish a presence that arrived during the atomic era from one that was on station earlier and quiescent, and the surveillance-intensification model accommodates either. What the data do not accommodate, on pain of violating relativity, is a presence that was elsewhere when the tests began.

The Skeptical Case at Full Strength, and What Survives It

It is the habit of advocates to present the opposition’s weakest argument and refute it at leisure. The opposite discipline is the only one worth practicing. Here, then, is the strongest skeptical case as I can construct it from the published record.

First: the catalogs are contaminated, and the contamination is conceded. The Villarroel team’s own response to Watters allows that up to two-thirds of the 107,875-object sample may be false positives. Ensemble statistics over so impure a sample demand unusual confidence in the impurity being random with respect to the variables under study, and the critics are entitled to doubt it. Second: single-epoch detections are individually unfalsifiable; nothing can ever be re-observed, and a research program built on the unrepeatable must clear a higher bar than one built on the repeatable. Third: critics have noted that the pipelines have not been demonstrated to recover known objects, modern satellites for instance, in comparable data, an absent positive control that weakens confidence in what the pipelines find. Fourth: portions of the corpus, including the independent replications of Doherty and Cann and both Busko papers, remain preprints, unrefereed; and one independence claim, Doherty’s, has since expired into collaboration. Fifth: even granting every detection, the artificial-origin hypothesis is underdetermined. Real sub-second flashes in pre-Sputnik skies could in principle be an unrecognized atmospheric or astrophysical phenomenon; Busko’s own clustering in two small fields is, until shown otherwise, as suggestive of some localized natural process as of visitation.

That is the case, and parts of it stand. What does not stand, any longer, is the first and load-bearing claim, the claim on which Hambly and Blair rested everything and on which the Watters critique tacitly depends: that the transient phenomenon as such may be nothing more than plate artifacts. The coma result is object-level evidence, exactly the species the critics demanded, and it is immune to every quarrel about ensemble purity, plate-time reconstruction, and filtered subsets, because it makes no use of ensembles, reconstructions, or subsets. Eleven events, individually inspected, individually bearing the optical signature that only the telescope can impose. The skeptic who wishes to maintain the artifact thesis against these must now explain how emulsion chemistry learned the focal geometry of a parabolic mirror, oriented itself toward plate center, scaled itself with field radius, saturated and haloed like real light, and did all of this while declining to occupy the catalog positions of the adjacent stars. I do not say it cannot be attempted. I say the attempt has not been made, and that the burden has visibly changed shoulders.

The honest residue of the skeptical case is therefore the fifth point: the transients are real, and we do not know what they are. That is a far smaller territory than the skeptics held in January, but it is genuine territory, and the instruments that could decide it are discussed below.

Five Axes of Convergence

In experimental science a single group’s finding, however rigorous, remains provisional until independently confirmed, because independent replication is the mechanism by which science distinguishes signal from systematic error. Every instrument has idiosyncrasies; every pipeline has assumptions; every researcher has blind spots. When multiple groups, using different instruments, methods, and data, converge on the same result, the probability that all are fooled by the same artifact collapses toward zero.

Consider the objections serially. Emulsion defects? Different emulsions, different manufacturers, different decades of production, two different Hamburg telescopes and the Palomar Schmidt besides; the signatures persist. Scanner artifacts? Different scanners, different digitization facilities, different reduction pipelines. A fluke of catalog cross-matching? Busko consults no external catalogs at all. A peculiarity of one telescope’s optics? Hamburg’s two instruments differ from Palomar’s and from each other, and in the Doppel-Reflektor’s case the optics themselves now testify for the prosecution. And neither defects nor inert debris has any mechanism by which to produce a monotonic dose-response across five geomagnetic activity bins at p = 0.0007. An artifact is, by definition, specific to the system that produces it. A signal that survives transplantation across systems has forfeited the name.

The evidentiary structure now has five axes. Busko supplies physical replication twice over: the narrow-profile signature on the Schmidt, and the comatic signature on the Doppel-Reflektor, the second invulnerable to the counter-reading that narrowness itself betrays a defect. Doherty supplies statistical replication: independent code, independent regression models, independent permutation methods, the same nuclear-timing correlation and the same shadow deficit. Cann supplies geomagnetic replication: a dose-response that physically situates the detectable population in the radiation-belt environment, and a confound control that sharpened the very signal it was built to test. And the internal physical coherence supplies the fifth axis: the near-quadrupling of the nuclear-test incidence ratio when restricted to sunlit positions is predicted by the solar-reflection hypothesis and predicted by nothing else on offer. Each axis addresses a different class of objection. All point the same way.

What has been detected, on the evidence: point sources consistent with sub-second optical flashes; bearing, on one telescope, the narrow profiles such flashes must produce, and on another, the comatic signature that only light through the optics can produce; distributed non-randomly with respect to Earth’s shadow at geosynchronous altitude; temporally correlated with atmospheric nuclear weapons tests at a level that survives confounder adjustment, permutation, and geomagnetic control; physically coupled to the geomagnetic environment in the manner of objects inhabiting the outer radiation belts; and appearing in skies that predate every human satellite. The evidence now lends serious probative support to the hypothesis that reflective objects of non-human origin occupied near-Earth orbital space before the Space Age began. This is an extraordinary claim, and it has been subjected to exactly the scrutiny extraordinary claims require. It has, so far, survived.

Aggregating the Evidence: A Bayesian Aside, Revised

Together the corpus now furnishes six lines of evidence bearing on a single hypothesis: that reflective objects of artificial origin occupied near-Earth orbital space before October 1957. Bayesian inference offers a disciplined framework for aggregating such evidence, and even a toy calculation is instructive, because it reveals the structure of evidential convergence when independent lines combine multiplicatively. For each line we estimate P(E | artificial) and P(E | natural), where “natural” is the composite of every conventional explanation: emulsion defects, scanner artifacts, astrophysical transients, atmospheric phenomena. The ratio is the Bayes factor; for independent lines, the factors multiply.

E1: The narrow-profile signature (Villarroel, Solano and Marcy; Busko Paper I). Sub-second specular reflection predicts it; defects and conventional astrophysical transients do not. Conservative Bayes factor 8.5; moderate 15.3.

E2: The Earth-shadow deficit at geosynchronous altitude (Villarroel et al. 2025; Doherty). Sunlit reflectors at that altitude predict the deficit; no proposed natural or artifactual mechanism does. Conservative 18.0; moderate 31.7.

E3: The nuclear-test timing correlation (Bruehl and Villarroel; Doherty’s permutation p = 0.006; Cann’s controlled regression at p = 0.002). Conservative 25.0; moderate 70.0.

E4: The sunlit-only amplification (Doherty). The timing signal concentrates where specular reflection is physically possible, as the reflection hypothesis predicts and nothing else does. Conservative 16.0; moderate 30.0.

E5: The geomagnetic dose-response (Cann). Detectability declines monotonically with Kp, and the nuclear signal strengthens under geomagnetic control; neither defects nor inert debris predicts either property. Conservative 12.0; moderate 22.0.

E6: The comatic signature (Busko Paper II). Here a word of method is owed, because honesty requires assigning this line the most modest factor of the six despite its rhetorical force. The coma evidence discriminates real light from artifact; it does not, by itself, discriminate artificial reflectors from an unknown astronomical or atmospheric source, since light from any of these would traverse the optics and acquire the signature. Its Bayes factor against the composite natural hypothesis therefore depends on how much of that composite’s prior mass sat on the artifact branch; and since nearly all published skepticism has rested precisely there, the factor is real but bounded. Conservative 5.0; moderate 10.0. Its deeper effect is structural: it amputates the limb of the natural hypothesis on which the skeptical literature has chiefly stood.

The conservative estimates deliberately favor the natural hypothesis at every step; the moderate estimates represent reasonable best judgments. The combined factors:

Likelihood set

Combined Bayes factor (six lines)

Conservative

3,672,000×

Moderate

224,075,000×

Applied to two priors, the first (1 percent) representing a reader who finds the hypothesis improbable yet not absurd, the second (0.01 percent) representing a reader who would file pre-Sputnik artificial objects alongside perpetual motion:

Prior P(artificial)

Conservative posterior

Moderate posterior

1.00%

99.997%

≈99.99996%

0.01%

99.73%

99.996%

The prior barely matters, and that is the hallmark of strong evidence: when six independent Bayes factors multiply into the millions, they overwhelm any defensible starting point. A critic must now argue that all six lines have been overestimated by orders of magnitude simultaneously. And because the lines are orthogonal (image optics on two different telescopes, spatial geometry, temporal statistics, the shadow-partitioned amplification, the geomagnetic coupling), no single alternative explanation can account for them jointly. An artifact hypothesis that once explained the narrow profiles is now contradicted outright by the coma; an atmospheric hypothesis that explains the timing correlation does not explain the sunlit amplification; a debris hypothesis that survives the spatial and temporal tests fails the geomagnetic dose-response. The natural hypothesis space fragments under the weight of convergence; the artificial-origin hypothesis is fed by it.

This remains, to be clear, a toy model. The likelihood ratios are informed judgments, the independence assumption is the model’s weakest joint, and reasonable people will assign somewhat different values. The qualitative conclusion is insensitive to those adjustments; the quantitative one should be carried lightly.

Where Are They Now?

The most natural follow-up question is also the most persistent objection: if reflective objects populated pre-Sputnik orbit, why do modern surveys fail to detect them? The objection sounds devastating, and it rests entirely on the assumption that we have actually looked. We have not.

We have never conducted the right search. Robert Freitas, in a 1985 analysis in the Journal of the British Interplanetary Society, calculated that near-Earth orbital space is 99.9999 percent unexplored for objects in the one-to-ten-meter range, and four decades later the calculation remains substantially correct. A mirror-finished ten-meter probe at a quarter of an astronomical unit registers at magnitude 21, invisible to most surveys; smaller, darker, or unfavorably oriented, it drops below every operational threshold. The Space Surveillance Network tracks some 45,000 objects larger than ten centimeters in low orbit, but completeness collapses at geosynchronous altitude, and the GSSAP satellites stationed there monitor known, cataloged objects. No systematic all-sky survey has ever been designed to detect small, non-transmitting, non-cooperative objects at geosynchronous altitude.

Modern surveys use the wrong detection paradigm. Asteroid surveys are optimized for objects that move against the sidereal background; a geosynchronous object appears stationary against the stars, and automated pipelines would classify it as a star and discard it. Transient surveys such as ZTF, and now the Rubin Observatory’s LSST, are built for supernovae and afterglows evolving over hours to weeks; a sub-second glint registers as a single-frame point source indistinguishable from a cosmic-ray strike, and is flagged as an artifact by the very pipelines designed to keep the data clean. Space surveillance looks for the streaks that low-orbit objects leave; geosynchronous glints leave none. The irony is exquisite: every major detection paradigm in contemporary astronomy is optimized to reject exactly the class of signal these transients represent.

The objects may have changed. Seven decades of space weathering darkens metallic surfaces by orders of magnitude; the same probes that glinted brilliantly in 1951 may now be too dark to register in reflected sunlight. They may have raised orbit, gone dormant, reoriented to minimize cross-section, or departed. The physicist Gregory Benford, proposing the “lurker” concept in 2019, observed that on millennial timescales “the ruins of Lurker installations may be visible” long after their animating intelligences are gone. The concept is a straightforward application of the logic of remote surveillance to interstellar scales.

Vast orbital niches remain unsurveyed. The Earth-Sun Lagrange points L4 and L5 are gravitationally stable niches barely characterized; Earth’s first known Trojan asteroid was discovered only in 2010, though the regions have been theoretically understood since the eighteenth century. Horseshoe orbits, quasi-satellite trajectories, and the graveyard belt 300 kilometers above geosynchronous altitude, where a probe would share quarters with five hundred decommissioned human satellites that nobody scrutinizes, all receive minimal attention. If one wished to park a monitoring instrument where no one would notice it, these are the addresses one would choose. No one has checked.

What Can Be Done

The observational gap is a matter of instrument design and allocation priority rather than fundamental physics, and the tools either exist or are within reach. The most targeted proposal is ExoProbe, developed by Villarroel and collaborators: a network of telescopes observing the same sky patch simultaneously, so that coincidence detection eliminates cosmic rays and single-instrument artifacts in real time, multiple baselines permit triangulation of a flash’s altitude, and spectroscopy distinguishes reflected sunlight from intrinsic emission. The design philosophy is elegant in its directness: a small number of fully classifiable events in place of millions of ambiguous ones. A complementary strategy, demonstrated in the 2026 MNRAS proof of concept with ZTF data, exploits Earth’s shadow as a natural filter: any self-luminous object detected inside the shadow at geosynchronous altitude cannot be a sunlit satellite or debris, and the contamination that plagues every other method is eliminated by geometry.

Existing infrastructure can be re-mined with purpose-built filters: ZTF’s nightly alerts, LSST’s coming billions, TESS’s four co-pointed cameras ideal for coincidence work, LaserSETI already running and capable of catching specular reflections as a byproduct. A search for the anomalous reflectance spectra of long-weathered metal, distinct from fresh satellite materials, is feasible within existing surveillance data. And the archival archaeology has barely begun. Busko has now demonstrated two independent discriminating techniques on two Hamburg telescopes, has promised a quantitative successor paper folding in further instruments, and has analyzed a small fraction of the available holdings; the APPLAUSE archive continues digitizing European collections, and millions of plates worldwide remain unsearched. The comatic method is, moreover, ripe for automation: the control samples Busko built by hand are precisely the training data an automated classifier requires. His pipeline is public. Anyone with a laptop can join.

The Decision to Look

The question is no longer whether the transients are real. Two physical replications from a second observatory, one resting on image narrowness and one on the optical signature of coma, together with an independent statistical replication and an independent geomagnetic one, have settled that, for any reader willing to weigh evidence rather than reputations. The narrow-profile signature is no Palomar artifact. The shadow deficit is no statistical mirage. The nuclear-timing correlation has now sharpened under three successive layers of adversarial scrutiny. And eleven images from a parabolic mirror in Hamburg carry the one credential no blemish can forge: the telescope’s own handwriting.

The question that remains is what the light was, and that question is answerable. ExoProbe could be operational within years; the shadow technique requires only telescope time on existing facilities; the plates are digitized and waiting. What a retired NASA engineer, a financial-services analyst, and an independent researcher in California have demonstrated, working alone with laptops and public data, is that the barrier to progress is neither technological nor, any longer, evidentiary. It is institutional: the willingness to treat the evidence with the seriousness its weight demands and to direct resources toward the next logical step. After five convergent lines of confirmation, declining to look is itself a scientific position, and it is one that will need, increasingly, to be defended.

Appendix: Sigma Equivalents of the Bayesian Posteriors

Readers accustomed to frequentist thresholds may find it useful to express the posteriors in equivalent sigma units, mapping the residual probability assigned to the natural hypothesis, P(natural) = 1 − P(artificial), to the one-sided Gaussian tail probability yielding the same p-value. This is a translation convenience, since the two frameworks answer different questions: a frequentist sigma quantifies the rarity of data under a null, a Bayesian posterior quantifies confidence after updating. For reference, 3σ corresponds to p ≈ 0.0013, 4σ to p ≈ 3.2 × 10⁻⁵, and 5σ to p ≈ 2.9 × 10⁻⁷.

With six evidence lines, from the 1 percent prior the moderate likelihood set yields approximately 4.9σ, comfortably past the conventional discovery threshold in particle physics, and the conservative set approximately 4.0σ. From the maximally skeptical 0.01 percent prior, the moderate estimate yields roughly 3.9σ and the conservative roughly 2.8σ, the latter still past the level at which initial astrophysical observations typically prompt serious follow-up investment. The addition of the comatic line (conservative factor 5, moderate 10) multiplies the previous five-line combined factors of 734,400× and 22,407,000× to 3,672,000× and 224,075,000× respectively.

The individual sigma values reported in the underlying papers measure individual effects: 22σ for the hemispheric shadow deficit, 7.6σ for the plate-coverage model, 3.9σ for the geometric alignment, Cann’s Z = −3.391 trend and his 3.1σ controlled nuclear signal. The aggregated figures here represent something different: posterior confidence across all six lines, translated into a common unit. The individual sigmas say each pattern is statistically real. The aggregated sigma says the patterns, taken together, increasingly resist every explanation but one.

References

1. Villarroel, B., et al. 2020, “The Vanishing and Appearing Sources during a Century of Observations,” The Astronomical Journal, 159, 8.

2. Villarroel, B., Marcy, G. W., Geier, S., et al. 2021, “Exploring nine simultaneously occurring transients on April 12th 1950,” Scientific Reports, 11, 12794.

3. Solano, E., Villarroel, B., & Rodrigo, C. 2022, “Discovering vanishing objects in POSS-I red images,” Monthly Notices of the Royal Astronomical Society, 515, 1380.

4. Hambly, N., & Blair, A. 2024, arXiv:2402.00497.

5. Villarroel, B., Solano, E., & Marcy, G. W. 2025, “On the Image Profiles of Transients in the Palomar Sky Survey,” arXiv:2507.15896.

6. Bruehl, S., & Villarroel, B. 2025, “Transients in the Palomar Observatory Sky Survey (POSS-I) may be associated with nuclear testing and reports of unidentified anomalous phenomena,” Scientific Reports, 15, 34125.

7. Villarroel, B., et al. 2025, “Aligned, Multiple-transient Events in the First Palomar Sky Survey,” Publications of the Astronomical Society of the Pacific, 137, 104504.

8. Watters, W. A., Dominé, L., Little, S., Pratt, C., & Knuth, K. H. 2026, “Critical Evaluation of Studies Alleging Evidence for Technosignatures in the POSS1-E Photographic Plates,” arXiv:2601.21946.

9. Villarroel, B., Streblyanska, A., Bruehl, S., & Geier, S. 2026, “A Response to Watters et al. (2026),” arXiv:2602.15171 (v2, 30 April 2026).

10. Bruehl, S., Doherty, B., Streblyanska, A., & Villarroel, B. 2026, “Machine Learning Supports Existence of Previously Unrecognized Transient Astronomical Phenomena in Historical Observatory Images,” arXiv:2604.18799.

11. Busko, I. 2026a, “Searching for Fast Astronomical Transients in Archival Photographic Plates,” arXiv:2603.20407 (Paper I).

12. Busko, I. 2026b, “Fast Astronomical Transients in Archival Photographic Plates: Using optical aberrations as a tool for discerning real images from plate artifacts,” arXiv:2606.08319 (Paper II).

13. Doherty, B. 2026, “Independent Replication of Nuclear Test-Transient Correlations and Earth Shadow Deficit in POSS-I Photographic Plates,” arXiv:2604.00056; code at github.com/dca-doherty/VASCO-Replication.

14. Cann, K. 2026, “Geomagnetic storm suppression of photographic plate transient detections in the POSS-I archive,” arXiv:2604.04950; reproduction script at osf.io/8ryhk.

15. Villarroel, B., et al. 2026, “A Cost-Effective Search for Extraterrestrial Probes in the Solar System,” Monthly Notices of the Royal Astronomical Society, 546.

16. Busko, I., plateanalysis software repository: github.com/cuernodegazpacho/plateanalysis.

17. APPLAUSE Archive (Archives of Photographic PLates for Astronomical USE): plate-archive.org.

18. Freitas, R. A., Jr. 1985, “The Search for Extraterrestrial Artifacts (SETA),” Journal of the British Interplanetary Society.

19. Benford, G. 2019, “Further Thoughts on Lurkers,” Centauri Dreams.

20. GSSAP Fact Sheet, United States Space Force.