Critical Evaluation of Studies Alleging Evidence for Technosignatures in the POSS1-E Photographic Plates

Cameron Pratt; Kevin H. Knuth; Laura Domin\'e; Matthew Szenher; Sarah Little; Wesley Andr\'es Watters

REVIEW 3 major objections 6 minor 1 cited by

Critical Evaluation of Studies Alleging Evidence for Technosignatures in the POSS1-E Photographic Plates

T0 review · 3 major / 6 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper argues that the reported deficit of features inside Earth's shadow and the correlation with nuclear tests vanish when the most vetted public dataset is used and the survey's true observation days are counted.

desk verdict Solid critique of the Villarroel et al. shadow/technosignature claims, with a few inferential steps resting on the unpublished V dataset. read the letter →

arxiv 2601.21946 v3 pith:UWI3OMLM submitted 2026-01-29 astro-ph.IM

Wesley Andrés Watters , Laura Dominé , Sarah Little , Cameron Pratt , Kevin H. Knuth , Matthew Szenher This is my paper · ORCID

classification astro-ph.IM

keywords technosignaturesPOSS1-EplatesopticaltransientcandidatesEarthshadowdeficitplateartifactsnuclear-testcorrelationreplicationanalysisdatavalidation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper re-examines the evidence behind claims that unidentified features on pre-Sputnik photographic plates are glinting artificial objects near Earth. Using the most aggressively filtered public subset of the feature catalog, the authors find no deficit of features inside Earth's shadow: 66 of 4,866 features fall in the shadow (1.36%), versus 0.82% expected. They also find that the reported correlation between feature observation days and nuclear tests disappears (p≈0.1) once the analysis is normalized by the 312 nights the survey actually observed the northern sky, and that 96% of the reported overlap with test windows is just the observing schedule. The paper documents that the unpublished dataset used in the original studies is inconsistently defined, largely unseparated from catalog stars and plate artifacts, and that the features' spatial patterns point to plate defects rather than sky sources. A sympathetic reader would care because the technosignature conclusion rests on these vanishing signals, and the paper shows how to test such archival claims properly.

What carries the argument

The load-bearing tool is the 'remainder' set R, the most aggressively filtered public subset of the larger feature catalog, built by removing features near catalog stars, asteroids, variables, objects seen in only one scan, and other artifacts. Because R and the unpublished dataset share the same ~1.4% southern-hemisphere fraction, the authors use R as a stand-in for the unpublished data. Against R they run three quantitative tests: a cell-based shadow simulation comparing in-shadow fractions, a Clark-Evans nearest-neighbor ratio to measure clustering against complete spatial randomness, and radial density profiles to show edge and corner excesses. The temporal test uses the plate metadata t

What would settle it

A direct re-release of the exact unpublished dataset V with plate IDs and timestamps would settle the question: if, after matching to plates, feature days appear on substantially fewer than ~290 of the 312 northern observation days, or if the in-shadow fraction in V falls below the 1.36% seen in R, the artifact explanation would need revision. Conversely, microscopic inspection of a random sample of V features showing star-like 3-D silver-grain structure and no catalog counterpart could revive the transient interpretation.

Watch

Extended reading notes

Core claim

Working from the most heavily filtered public subset of the POSS1-E feature catalog (set R, N=5,399), the authors find 66 of the 4,866 unambiguously plate-assigned features inside the simulated geosynchronous shadow — 1.36%, versus 0.82% expected from the shadowed fraction of the survey area — so the reported 30-75% deficit is absent. After matching features to plates and normalizing by the 312 nights on which the survey actually exposed northern-sky plates (not the 2,718 calendar days of the study window), the feature/nuclear-test correlation drops to p≈0.1 with a relative risk of 1.07; 96% of the overlap between feature days and test windows is just the observing schedule. The paper also s

Load-bearing premise

The analysis assumes the unpublished dataset V, on which the original claims were based, has the same sampling properties as the public R and W sets — particularly that southern-hemisphere plates were skipped and that the 312 northern observation days define the relevant exposure window; if the original authors sampled V differently, the normalization that erases the nuclear-test correlation would be weakened.

Editorial extensions

If this is right

The reported 30-75% deficit of features inside Earth's shadow, cited as evidence for geosynchronous glinting objects, is not reproduced with the vetted remainder set; the in-shadow fraction (1.36%) actually exceeds the expected 0.82%.
The claimed feature/nuclear-test correlation (χ²=6.94, p=0.008 in the original study) becomes p≈0.1 with relative risk 1.07 when normalized by the 312 true northern-sky observation days.
Because features appear on 93-99% of actual observation days, their occurrence is almost completely determined by when the survey observed the sky, not by nuclear testing.
A third of the candidate aligned-cluster features match catalog stars within 2 arcseconds, so they were not confidently distinguished from known objects.
The spatial patterns — edge/corner excess, right-ascension stripes, and plate-boundary-related clusters — indicate plate and digitization artifacts rather than sky sources.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper's own claims, the results imply that any future archival-plate search for artificial satellites should require independent per-object validation (e.g., microscopic inspection of the emulsion) rather than relying on statistical correlations over unvalidated catalogs.
The conspicuous deficit band in right ascension (roughly 87°-107°) that survives aggressive filtering suggests a systematic pipeline effect; identifying its cause would provide a corrected background model for any reanalysis of these plates.
A testable extension would compare feature detection days against other weather-dependent activities, such as aerial surveys or artillery tests, to see whether the apparent seasonal coupling with nuclear tests is generic rather than specific.
Since even the vetted R set still contains 4-5% clear stars and artifacts, the cleanest available catalog needs per-object morphology screening before any single candidate can be used as evidence of a real optical transient.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here's my read. The paper is a systematic re-analysis of the POSS1 plate features that Villarroel and colleagues have been pushing as evidence for pre-Sputnik artificial satellites. Taking the public datasets (R and W) seriously, the authors show three things worth knowing. First, the alleged deficit of features in Earth's shadow does not appear in the most vetted dataset R: 66 of 4,866 unambiguously assigned features land in shadow, 1.36% versus the expected 0.82%. Second, the much-publicized correlation between feature dates and nuclear tests is essentially a product of the Palomar observing schedule: 56 of the 368 actual observation days overlap nuclear test windows, and 54 of those are claimed as coincidence, so 96% of the overlap is geometry and weather, not physics. Third, the target papers rely on a dataset V that is unpublished, defined inconsistently across the papers, and built by re-filtering a set that had already been cleaned to R; the original papers also use circular reasoning, treating the very correlations under test as validation of the data. These are serious points, and they land.

What's genuinely new: the observation-schedule normalization, the R-based shadow test, the documentation of RA-stripe voids and edge/corner excesses, and a careful inventory of dataset inconsistencies. The manual inspection of 540 R features (4–5% residual contamination) is honest empirical work. The GRB plate-history review is context, not a result, but it is well chosen.

The soft spots are real but not fatal. The shadow comparison on R comes without uncertainty quantification; 66 versus an expectation of 40-ish sounds suggestive, but we are not given a p-value or confidence interval, and only ten plates contribute. More importantly, the quantitative refutation of the nuclear correlation—the p=0.1 Fisher exact result—is computed on R with a 312-day denominator inferred from R and W, not verified on V. V is unpublished, and if its plate selection differs in declination or in which northern plates were included, the denominator changes and the p-value could shift. The authors claim access to V is unnecessary; that is true for exposing inconsistencies and circularity, but it is false for the precise quantitative normalization claim. They should soften that. There is also a small independence issue: the acknowledgments state that Watters received a copy of V while collaborating on an earlier version of one of the target papers, then withdrew. He may well have prior knowledge of V's contents even if it was not used analytically. That does not invalidate the public-data analysis, but it undercuts the clean-room claim.

Central argument holds. The target papers' evidence does not survive contact with the public data, and the burden is on them to show V behaves differently. I'd send this to a serious referee, asking specifically for error bars on the shadow statistic and a more careful statement about what the R-based normalization can and cannot establish. For a reading group on archival-plate methods and null-model design, it is a good case study.

Referee Report

3 major / 6 minor

Summary. This paper re-examines claims by Villarroel et al. (2025c) and Bruehl & Villarroel (2025) that unidentified features in POSS1-E plates show an Earth-shadow deficit, linear clusters, and correlations with nuclear tests. Using two public datasets, R (N=5,399, the most aggressively filtered set) and W (N=171,753), plus the MAPS celestial-object set M, the authors argue that the assumed uniform-random background is false: SPF density rises toward plate edges, and large-scale stripes and voids appear in W and R. They report no shadow deficit in R (f_obs=1.36% vs f_exp=0.82%), and find that the nuclear-test correlation becomes insignificant (p≈0.1) when normalized by 312 actual POSS1 observation days rather than 2,718 calendar days; they further show that 54 of the 56 days overlapping nuclear-test windows are pure schedule overlap. They also document inconsistent definitions of the unpublished dataset V and identify a circular argument in the target papers.

Significance. If the quantitative claims hold, the paper materially weakens the technosignature interpretation of the POSS1-E feature sets and provides a useful cautionary case study for archival-plate searches. Its strengths are the use of public datasets, the careful documentation of dataset inconsistencies in Appendix A, and the transparent schedule-overlap argument, which is simple and robust. The historical GRB-plate review is also well placed and supports the demand for independent validation. However, the strongest quantitative refutations—the p=0.1 nuclear-test result and the shadow comparison—are computed on the public dataset R, not on the unpublished target dataset V, and rely on inferred sampling properties of V. The paper would be more convincing with a direct test on V or a thorough sensitivity analysis; as written, the normalization claims are conditional on an unverified assumption.

major comments (3)

[§6.2, Table 3] The 'loses significance' claim (p=0.1) is computed with the public dataset R, while the target correlation is for the unpublished V. The 312-day denominator is inferred from the 1.4% southern-hemisphere fraction in R and W (§3.2) and then used to remove 56 southern-only observation days. Appendix A shows that V's construction is inconsistently defined (five different statements), so its effective observation window is not known. If V sampled southern-only plates, or excluded some northern plates, the denominator and p-value could change. The statement in §3.1 that access to V is 'not necessary' is therefore too strong: it is sufficient for exposing dataset inconsistencies, but not for the quantitative normalization claim. Please obtain V (one author had a copy) or provide a sensitivity analysis over plausible V constructions and denominators.
[§5] The in-shadow test is reported as f_obs=66/4866=1.36% versus f_exp=0.82%, with no uncertainty, confidence interval, or significance test. The expected fraction depends on simulation choices (30×30 cell grid, 8.5° in-shadow radius, plate-assignment radius) that are not varied. Although f_obs exceeds f_exp, the paper's claim that the reported deficit is 'not present' needs a formal test (e.g., Poisson with plate-level overdispersion) to be comparable to the 2.5–22σ claims in the target study. Moreover, the calculation is on R, not V; because V is unpublished and ambiguously defined, it cannot exclude a V-specific deficit if V's plate selection differs from R's. Please add a significance/sensitivity analysis and limit the scope of the conclusion to R (or V once obtained).
[§4.2–4.4] The argument that the uniform-random null is false relies heavily on set W, which is a subset of S selected by proximity to NeoWISE objects. The paper calls W 'a dense, uniform-random sampling of S' without demonstrating that the NeoWISE positions used for matching are spatially uniform. Clustering in W could in principle reflect structure in the infrared catalog or in the matching procedure. The MAPS set M is a useful control for celestial sources but is not matched to NeoWISE. Please verify the uniformity of the sampling positions (e.g., compare W to a random subset of NeoWISE positions) or re-derive the nonuniformity conclusion directly from R's intra-plate density profiles (which are shown but not formally tested). This matters because the invalidation of the Poisson null is load-bearing for rejecting the target's shadow significance.

minor comments (6)

[§3.2, 'SetV' paragraph] V is given as N=107,185, whereas Table 1 and §6 use N=107,875. Please correct the typo.
[§5] 'solid angle (in degrees)' should read 'square degrees'.
[§4.2 and §5] Plate 090R has 2,149 SPFs with an average of 257.5 per plate (§4.1); it is therefore ~8.3 times the average, not 'nearly 20 times the average'.
[§8, item 5] The statement 'SPFs are found on 310 of the 312 observations days' conflicts with §6.2, where R has 289 unique days and W has 307; state that this refers to V and reconcile the numbers.
[§6.1, Table 3] Specify which statistical test produced p=0.1 (Fisher exact, chi-square with/without Yates) and report the test statistic; the text mentions Chi-Square=6.94 for the original but not for the recalculation.
[§4.1 and §6.1] Consider releasing the plate-assignment and shadow-simulation code to make the analysis fully reproducible; the current text lists parameters but not the algorithms.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the paper's re-analysis is self-contained and tests public datasets against external plate metadata; the V-proxy inference is an extrapolation, not a circular reduction.

full rationale

This paper is a critical re-analysis, not a derivation that assumes its conclusion. Its central quantitative claims are direct empirical tests on publicly available datasets R and W, combined with external POSS1 plate metadata and nuclear-test date sources. The shadow comparison (f_obs = 66/4,866 = 1.36% vs f_exp = 0.82%) is computed from R and a geometric simulation; it has no fitted parameters and does not presuppose the absence of a deficit. The nuclear-test correlation is recomputed on R with the denominator replaced by the actual 312 northern-hemisphere observation days derived from the POSS1 plate list; Table 3 transparently states 'Dataset used R' rather than V, so this is not a prediction disguised as a fit. The only mild concern is the inference that unpublished dataset V shares R/W sampling properties (e.g., the ~1.4% southern-hemisphere fraction and the exclusion of southern-only observation days). That is an extrapolation from reported V sizes and public dataset statistics, and the paper explicitly reports the conservative 368-day alternative as well; it is not circular because the inference is not defined in terms of the conclusion it supports. Self-citations in the introduction (e.g., Watters et al. 2023, Knuth et al. 2025) are contextual and not load-bearing for any derivation. No circular step can be exhibited, so the appropriate score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No parameters are fitted to the target hypotheses, and no new entities are introduced. The listed quantities are methodological thresholds inherited from the original studies or adopted for plate assignment and shadow geometry; the listed axioms are domain assumptions about what the public datasets represent and how the plate metadata should be interpreted.

free parameters (3)

plate_matching_radius = 3 deg in declination
SPFs in R and W lack plate IDs, so each is assigned to the nearest plate center within 3 degrees in declination; this choice determines which observation day is used in the nuclear-test correlation (Section 4.1).
in_shadow_radius = 8.5 deg at GSO altitude
Inherited from Villarroel et al. 2025c and used to define shadowed cells for the expected in-shadow fraction; the radius choice affects f_exp.
shadow_cell_grid = 30x30 cells per plate
Discretization used to estimate the expected in-shadow fraction; a different grid size could slightly change f_exp and the conclusion threshold.

assumptions (5)

domain assumption W is a dense, uniform-random sample of S
Stated in Section 4.2 to justify using W to infer the spatial properties of S and V; the paper later shows W is nonuniform on plates, so the premise is an approximate sampling assumption rather than an established fact.
domain assumption R is the appropriate proxy for testing the in-shadow deficit
Section 5 uses R to conclude the Villarroel-V shadow deficit is not reproduced. This presumes R is more trustworthy than V and that absence of a deficit in R is informative about the original claim.
domain assumption Emulsion-defect rates from Greiner 1987 and Varady & Hudec 1992 apply to POSS1 103a-E copy negatives
Section 3.2 extrapolates lab rates (0.066 and 0.377 per cm^2) to estimate 107,000-612,000 star-like defects in A; the emulsions, plate history, and digitization differ from those experiments.
domain assumption M features are likely celestial objects
MAPS M contains features detected in both POSS1-E and POSS1-O plates and is used as a celestial control; the E/O coincidence criterion is treated as sufficient to exclude plate artifacts.
domain assumption POSS1 plate metadata from STScI accurately represents observation days
Section 6 reduces the time denominator from 2,718 days to 312 days based on the STScI plate list; if the plate metadata were incomplete, the schedule-correlation conclusion would need revision.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Critical Evaluation of Studies Alleging Evidence for Technosignatures in the POSS1-E Photographic Plates." pith.science (2026). https://pith.science/paper/UWI3OMLM

@misc{pith2026260121946,
  author       = {Pith},
  title        = {Pith review of: Critical Evaluation of Studies Alleging Evidence for Technosignatures in the POSS1-E Photographic Plates},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UWI3OMLM}},
  note         = {Machine review of arXiv:2601.21946}
}

read the original abstract

Recent studies by B. Villarroel et al. have assembled and analysed datasets of unidentified features measured from digital scans of pre-Sputnik photographic plates. We have examined the claims in these papers using two previously published datasets that are closely related to those used in the Villarroel et al. studies. For these datasets, the assumption of a spatially uniform-random background distribution of features, essential to the Earth shadow analysis, is shown to be false. After finding the null distribution of feature count deviations from the background, we find no statistically significant deficit in the shadow. We find that the reported correlation between the timing of feature observations and nuclear tests becomes insignificant after properly normalizing by the relevant number of observation days, and is almost completely determined by the observation schedule of the Palomar telescope. We uncover important inconsistencies in the definitions of the datasets used in these studies, as well as the use of unvalidated datasets containing catalogue stars, scan artefacts, and plate defects. We find an overall gradual increase in number density of features toward the corners and edges of plates, as well as examples of (i) empty north-south strips that span multiple plates; (ii) clusters and voids having geometric shapes; and (iii) amorphous clusters. We also highlight a circular argument used in these studies, that leverages the results of an inferential analysis to justify conclusions about the origin of the features as well as the validity of the measurements. Finally, we also review the literature concerning historical searches for optical transients in photographic plates corresponding to gamma ray bursts (GRBs); following decades of work, researchers were unable to make a confident identification of a GRB-associated optical transient.

Figures

Figures reproduced from arXiv: 2601.21946 by the authors.

**Figure 2.** Examples of Selected POSS1-E Features (SPFs) in R (N = 5, 399), the most vetted subset of S (N = 298, 165) published by E. Solano et al. (2022). The red crosses at the center are ∼ 10′′ in diameter. We have identified these features as follows: (i) clear artifacts or defects that do not resemble objects visible in a Pan-STARRS image of the same field (A (object 1092), B (object 1179), C (object 1223), D (object 124… view at source ↗

**Figure 3.** Cumulative distribution function of SPF counts as a function of the total fraction of plates, ranked in order of decreasing counts from left to right, for W, R, and M. In the case of W, roughly 20% of the most crowded plates contain over 60% of all SPFs. M is more uniform in terms of how SPFs are distributed across plates; this is partly because all SPFs in M are likely celestial objects, and because the Galactic pl… view at source ↗

Figures from the paper (9 more)

**Figure 4.** Figure 4: Scatter plots of SPFs from W on nine plates that exhibit distinctive patterns. Red dots signify SPFs that could not be assigned to a single plate because of overlap with the neighboring plate’s field of view. These patterns are not visible in the dataset M of SPFs that…

**Figure 5.** Figure 5: Histograms of the Evans-Clark ratio ρ calculated for each plate, for SPFs in sets M and W. As expected for a uniform-random distribution of points on a finite domain, ρ is typically slightly above 1 for set M, comprising SPFs of likely celestial origin because they ap…

**Figure 6.** Figure 6: Radial density profiles for the W (solid), R (dashed), and M (dotted) datasets. Densities were calculated in circular bins of 0.25◦ with radii denoting the angular distance from plate centers. All distributions were normalized for easy visual comparison of their shap…

**Figure 7.** Figure 7: Histograms (2-D) of SPF counts in excess of the median background as a function of offset from plate centers for (A) set M (likely celestial objects that appear in both O and E plates; see J. E. Cabanela et al. (2003)); (B) set W, which reside within 5′′ of the positio…

**Figure 8.** Figure 8: Plots of the number density of SPFs on the celestial sphere in (A) set M, (B) set W, and (C) set R. The distribution of SPFs in parts B and C reveal a vertical band with a deficit of SPFs between 90◦ and 105◦ ; this remains apparent following aggressive filtering in R.…

**Figure 9.** Figure 9: Magnified images of SPFs belonging to candidate linear clusters detailed in [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]

**Figure 10.** Figure 10: From top to bottom, plots of the observation dates (i) with POSS1-E plates solely residing completely in the southern celestial hemisphere; (ii) all remaining plate days (northern celestial hemisphere and/or overlapping the celestial equator); and (iii) the dates of n…

**Figure 11.** Figure 11: Illustration of an example of circular reasoning applied in B. Villarroel et al. (2025c), in which a premise and observation are used to justify a conclusion, while elsewhere the observation and conclusion are also used to justify confidence in the premise. technosig…

**Figure 12.** Figure 12: Histogram of the SPF counts as a function of declination in datasets R and W, suggesting that SPFs in the parent dataset (S) and its other derivatives (e.g., V ) were overwhelmingly sampled from plates whose footprints reside entirely within or that overlap the northe…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Mysterious Transients in the Palomar Observatory Sky Survey (POSS-1) as profound manifestation of the Dark Matter physics
hep-ph 2026-08 conditional novelty 4.0 of 10

The paper identifies Palomar mysterious transients as flashes from spallation of axion quark nugget dark matter hitting the atmosphere, with rates and features claimed consistent with VASCO observations.

Reference graph

Works this paper leans on

3 extracted references · 1 linked inside Pith · cited by 1 Pith paper

[1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter doi edition editor eprint howpublished institution journal key month number organization pages publisher school series title misctitle type volume year version url label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts ...
[2]

write newline

" write newline "" before.all 'output.state := FUNCTION format.url url empty "" new.block "" url * "" * if FUNCTION format.eprint eprint empty "" archivePrefix empty "" archivePrefix "arXiv" = new.block " " eprint * " " * new.block " " eprint * " " * if if if FUNCTION format.doi doi empty "" " " doi * " " * if FUNCTION format.pid doi empty eprint empty ur...
[3]

Missing Star

thebibliography [1] 20pt to REFERENCES 6pt =0pt \@twocolumntrue 12pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key o...

arXiv 2017

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter doi edition editor eprint howpublished institution journal key month number organization pages publisher school series title misctitle type volume year version url label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts ...

[2] [2]

write newline

" write newline "" before.all 'output.state := FUNCTION format.url url empty "" new.block "" url * "" * if FUNCTION format.eprint eprint empty "" archivePrefix empty "" archivePrefix "arXiv" = new.block " " eprint * " " * new.block " " eprint * " " * if if if FUNCTION format.doi doi empty "" " " doi * " " * if FUNCTION format.pid doi empty eprint empty ur...

[3] [3]

Missing Star

thebibliography [1] 20pt to REFERENCES 6pt =0pt \@twocolumntrue 12pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key o...

arXiv 2017