The Human Fingerprint on the Meteorite Record: A Full-Database Scan of the Meteoritical Bulletin (n = 85,775)

The Human Fingerprint on the Meteorite Record: A Full-Database Scan of the Meteoritical Bulletin (n = 85,775) | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article The Human Fingerprint on the Meteorite Record: A Full-Database Scan of the Meteoritical Bulletin (n = 85,775) Rongzhen Dai This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9687218/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The Meteoritical Bulletin Database underpins quantitative claims about meteoroid flux and composition, but its 85,775 valid and provisional entries combine specimens recovered under different search regimes. We treat the 1969 Yamato Mountains discovery and late-1990s hot-desert prospecting as two recovery-regime transitions and run a stratified scan together with adjusted-regression and pairing-uncertainty Monte-Carlo audits. Median recorded mass collapses 147-fold pre- to post-1969 at specimen level (4,400 g → 30 g; Cliff's δ = 0.81); a Monte-Carlo over literature pairing-factor ranges shrinks the fall-level mass-collapse median to 77× (95% CI [63, 96]×) but never below an order of magnitude. The iron-meteorite fraction falls from 29.9% to 0.9% at specimen level (Cohen's h = 0.97; pairing-MC fall-level: 34× with tight CI). The Antarctic-versus-Arabia/Oman lunar-fraction contrast is 11.7× at specimen level; the era-restricted 1969–1999 contrast is 6.75× ; the pairing-MC fall-level median is 4.31× with 95% CI [2.04, 9.73]× — comfortably bounded above parity but materially smaller than the unstratified figure. Decadal Fell counts average 62 ± 13 from the 1860s through the 2010s; the Mann–Kendall test returns Z = + 2.25, p = 0.024, a small but significant positive trend that a 10% reporting-improvement multiplier across 150 years already renders non-significant (deflated p = 0.08) and a 20% multiplier eliminates entirely. The Fell sub-sample is therefore a least-biased reference rather than an unbiased one. Adjusted logistic regression with era × region interaction confirms that era is the dominant predictor of iron fraction (post-1969 OR ≈ 0.02–0.18, p < 10⁻⁴⁹) and region is the dominant predictor of lunar fraction (Sahara OR = 142, Arabia/Oman OR = 26, both vs Other; p ≪ 10⁻³); both effects survive simultaneous control. The catalogue's headline negative latitudinal-mass correlation (ρ = −0.39) collapses to ρ ≈ 0 within witnessed falls — an instance of dataset-induced spurious physical inference driven by single-collection dominance of one corner of the covariate space. Users should restrict flux inferences to the Fell sub-sample with explicit acknowledgement of the documented small reporting trend, or apply the adjusted era-, region-, and pairing-stratified weighting we provide here. meteorite database Antarctic Search for Meteorites hot-desert recoveries sampling bias ANSMET lunar meteorites meteorite flux Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction The Meteoritical Bulletin Database is routinely used as a primary data source for meteoroid flux estimates, asteroid-source inversions, and planetary-meteorite census studies (Bland et al. 1996 ; Burbine 2017 ; Evatt et al. 2020 ). Yet the database conflates specimens recovered under fundamentally different search regimes — pre-1969 opportunistic collection, post-1969 Antarctic blue-ice programmes, and post-2000 hot-desert prospecting — without flagging which regime produced each entry. When downstream analyses treat the combined catalogue as a representative sample of the in-falling flux, they inherit biases that we show here to be of order 10–100× in mass, class composition, and geographic provenance. No prior study has quantified these biases at full-database resolution with effect sizes, adjusted regression, and pairing-uncertainty propagation. We provide that quantification. Meteorites carry information no other extraterrestrial sample can. They are the only macroscopic extraterrestrial matter routinely available for terrestrial laboratory analysis, and the catalogue of recovered specimens is the empirical foundation for almost every quantitative claim made about the flux, composition and origin of meteoroids reaching Earth (Grossman 1999 ; Connolly et al. 2007 ; Burbine 2017 ). The Meteoritical Bulletin Database, maintained by the Meteoritical Society at the Lunar and Planetary Institute, is the canonical curated catalogue. It combines two fundamentally different populations. The first is witnessed falls (the Fell category), where a fireball and an associated stone are unambiguously linked. The second is finds (the Found category), in which an extraterrestrial origin is established post-recovery on petrological and geochemical grounds (Bevan and Bindon 1996 ; Grady 2000 ). Witnessed falls are the least-biased available reference for the in-falling flux — modulated by population density, observational opportunity, and reporting culture, but free of the recovery-environment effects that dominate finds — while finds depend on where humans have looked and on how the local environment preserves and exposes meteoritic material. The catalogue grew slowly through the 18th, 19th and early 20th centuries, doubled roughly every 25 years up to about 1960, and then underwent two abrupt accelerations. The first was the discovery, in 1969, that the blue ice fields of East Antarctica preserve and concentrate meteorites by glacial flow against the Transantarctic Mountains and ablate them back to the surface, giving rise to the U.S. Antarctic Search for Meteorites programme (ANSMET) and to its Japanese (NIPR), Italian (PNRA), Chinese (CHINARE) and European (EUROMET) counterparts (Cassidy et al. 1992 ; Harvey 2003 ; Yamaguchi et al. 2002 ). The second, beginning in the late 1990s, was the systematic prospecting of hot-desert pavements in the Sahara, on the Arabian Peninsula, and in Oman, where flat regs and sebkhas combine with an arid climate to preserve dark cosmic stones for tens of thousands of years (Bischoff and Geiger 1995 ; Hofmann et al. 2009 ; Hutzler et al. 2016 ). These two episodes between them deposited some 80,000 new specimens into the database — roughly forty times the cumulative count of the entire previous two centuries. The growth has been understood since the 1990s as a recovery effect, not a flux effect (Bevan et al. 1998 ; Halliday et al. 1989 ; Bland and Artemieva 2006 ). The new specimens are systematically smaller, less iron-rich, and concentrated in places that pre-1969 collectors never visited. Yet the magnitude and structure of the bias is rarely quantified at the database scale, and downstream uses — flux modelling, asteroid-source inversion, isotopic-anomaly sampling — can inadvertently conflate catalogue composition with in-falling composition if recovery regime is not explicitly modelled. We quantify the magnitude of the bias at full-database resolution, with a stratified scan of the 85,775-record catalogue structured to mirror the questions a downstream user would ask before drawing inferences from any sub-sample. (i) How much of the post-1969 explosion is real flux vs. recovery effort? (We compare the Fell and Found time series.) (ii) How much smaller are post-1969 specimens, and how much less iron-rich, than pre-1969 specimens? (iii) Where, geographically, do the new specimens come from, and how does the prospecting environment shape the apparent abundance of rare classes (lunar, martian, iron)? (iv) Is the apparent latitudinal trend in recorded mass a physical signal or a search-bias artefact? (v) Do the two recovery surges define statistically distinct eras in the catalogue? The qualitative claim that the catalogue is recovery-biased has been recognised at least since the 1990s (Bevan et al. 1998 ; Halliday et al. 1989 ; Bland and Artemieva 2006 ). Our contribution is quantitative and structural . We report effect sizes (Cliff's δ for distributional contrasts, Cohen's h for proportional contrasts) with bootstrap confidence intervals throughout. We decompose the headline biases into era × region interactions to expose previously-conflated signals. We bracket the residual contribution from differential pairing under an explicit sensitivity analysis with a literature-pairing best estimate alongside two upper- and lower-bound stress tests. We provide a reproducible full-snapshot quantification of the long-recognised observation that the apparent negative correlation between recorded mass and absolute latitude is a sampling artefact that vanishes within the witnessed-fall sub-sample. We release a deterministic single-script analysis pipeline so the audit can be re-run as the Bulletin is updated. The principal finding is that any descriptive statistic computed from the combined Found+Fell sample is driven by recovery effort to first order, while the witnessed-fall sub-sample provides a substantially less-biased reference for population-level inference. Methods Data source We used a CSV export of the live Meteoritical Bulletin Database (MBDB) maintained by the Lunar and Planetary Institute on behalf of the Meteoritical Society, retrieved from https://www.lpi.usra.edu/meteor/metbull.php on 4 May 2026. The export carries 86,648 rows. Of these, 85,775 carry a Status flag of Official (n = 79,595) or Provisional (n = 6,180); we retain both because Provisional entries are typically catalogued in the Bulletin's published numbered issues and become Official on routine review, with conversion rates close to unity over multi-year windows. Records with Status = Pseudo , Doubtful , or Crater (n = 682 in total) are excluded, as their petrological status is either questioned or non-meteoritic. The MBDB schema differs from the older NASA Meteorite Landings tabulation; the field mapping is Code → identifier, Status → nametype, Fall → Fell/Found (Y or Yc → Fell), Year → year, Type → recclass, Mass → mass(g), Lat/Long → reclat/reclong, with a separate Boolean Antarctic flag we cross-check against the latitude-based assignment. Cleaning and filtering We coerced Year to a numeric integer and discarded a small number of records with corrupted year fields (one record dated to 2101 CE; two records dated before 500 CE). Table A1 in the supplementary material lists the three dropped entries by name. After cleaning, 83,895 records carry a usable year (97.8%), 85,211 carry a recorded mass (99.3%), and 52,096 carry decimal latitude and longitude (60.7%). Records lacking a mass field were retained for class-composition analyses but excluded from mass-distribution tests; records lacking coordinates were retained for time-series and mass analyses but excluded from geographic stratifications. Class taxonomy Classifying the 619 distinct Type strings in the live Bulletin into a small number of physically meaningful categories required deterministic substring rules. Iron meteorites were identified as records whose Type either began with the literal substring Iron (capturing the 12 chemical groups IAB through IVB and their structural sub-groups) or whose first whitespace-delimited token matched one of the canonical iron-group abbreviations (IAB, IIAB, IIIAB, IVA, IVB, IIICD, IIE, IIF, IIG, IIIE, IIIF, IC, IIC, IID). Lunar meteorites were identified by a case-insensitive match of Lunar. Martian meteorites by case-insensitive matches of Martian, Shergottite, Nakhlite or Chassignite — a slight broadening over the NASA snapshot taxonomy because the live Bulletin's Type strings encode Martian-grouping at the petrologic level rather than the explicit Martian token. Stony-irons (pallasites and mesosiderites) were identified by their own substrings; they form a small population (n = 614). All other records were treated as non-iron, non-lunar, non-martian for the binary contrasts. This taxonomy intentionally collapses the rich petrological structure of the chondrite classes into a single "non-iron" residual. We considered preserving finer chondrite subtypes (CM/CR/CV/CO carbonaceous, LL/L/H ordinary) but the Type strings encode subtype assignments inconsistently across decades — pre-1980 entries are often classified only to top-level group ("L" or "H") with no subtype, while modern entries carry full petrologic-grade information. Aggregating the two would create spurious era differences. The coarse taxonomy avoids that confound at the cost of compressing within-class diversity. Geographic strata Four prospecting regions were defined by simple latitude/longitude rectangles, chosen to match the operational footprint of the major recovery programmes. Antarctic is latitude < − 60° (the Transantarctic / blue-ice fields targeted by ANSMET, NIPR, PNRA, CHINARE and EUROMET). Arabia/Oman is 15° ≤ latitude ≤ 35°, 35° ≤ longitude ≤ 60° (the Rub' al Khali, Oman desert, and adjacent terrain prospected since the late 1990s). Sahara is 18° ≤ latitude ≤ 35°, − 15° ≤ longitude ≤ 35° (Algeria, Libya, Mali, Mauritania, Morocco, Niger, Western Sahara). Australia is − 40° ≤ latitude ≤ − 15°, 110° ≤ longitude ≤ 155° (Nullarbor Plain and adjacent regions covered by the early- and mid-20th-century Australian collections). The seam between the Sahara and Arabia/Oman rectangles at longitude 35° is exclusive on the Sahara side and inclusive on the Arabia side, so a record can belong to at most one rectangle. Records outside all four rectangles are aggregated into a single Other stratum dominated by Northwest Africa proper, the Atacama, the United States, and central Eurasia. The principal results are insensitive to ± 5° perturbations of the boundaries; we report the perturbation experiment in §Robustness checks. Era partition The catalogue was partitioned into three eras keyed to the two recovery transitions. Pre-1969 is year ≤ 1968, before any systematic Antarctic searches. Antarctic era is 1969 ≤ year ≤ 1999, dominated by ANSMET and partner programmes. Desert era is year ≥ 2000, in which hot-desert prospecting from Northwest Africa, the Arabian Peninsula and Oman dominates new accessions. The 1969 boundary is set by the December 1969 Yamato Mountains discovery by the Japanese Antarctic Research Expedition (Yoshida et al. 1971 ), conventionally taken as the start of the systematic Antarctic-meteorite era. The 2000 boundary is more graded; we adopt year 2000 as a notational convenience and verify (§Robustness checks) that 1995, 2000 and 2005 boundaries give qualitatively identical era-composition profiles. Statistical methods Binary contrasts (e.g. iron-meteorite fraction pre- vs. post-1969, lunar fraction Antarctic vs. Arabia) used Pearson's χ² on a 2 × 2 contingency. Mass-distribution contrasts used the Mann–Whitney U test and the two-sample Kolmogorov–Smirnov test on log-mass. Monotonic-correlation tests used Spearman ρ, with absolute latitude |φ| where the physical question was not hemisphere-resolved. Trend tests on the witnessed-fall time series used the rank-based Mann–Kendall Z statistic (Mann 1945 ; Kendall 1975 ), with a pre-specified Monte-Carlo power calculation (10⁴ Poisson realisations of a 16-decade series with the observed mean and dispersion) used to bracket the secular trends the test could and could not detect. At the sample sizes here, almost every two-group contrast is statistically significant by conventional thresholds; we therefore report effect sizes (Cliff's δ ; Cohen's h ) with bootstrap confidence intervals as the primary measure of contrast magnitude, and use p -values only as an auxiliary screen. Pairing-aware fall-level estimates apply two distinct treatments. The first is a literature-derived pairing factor for each region and class, applied as a multiplicative correction to specimen counts. We use Korotev ( 2012 )'s empirically-derived hot-desert lunar pairing factor of 2.5×, the Cassidy et al. ( 1992 ) and Harvey ( 2003 ) ANSMET-consensus Antarctic lunar pairing of ≈ 1.5× and Antarctic bulk-chondrite pairing of ≈ 3×, and the Hutzler et al. ( 2016 ) hot-desert bulk pairing of ≈ 1.75×. The second is a deterministic naming-based clustering applied as a lower-bound stress test : specimens whose name prefix (the leading three- to four-letter ANSMET/JARE site code, or hot-desert locality token such as NWA, Dhofar, or Dar al Gani) and discovery year both match are merged into a single fall unit, with year-less specimens treated as singletons. This clustering deliberately over-pairs bulk chondrites (e.g. all ALH 80xxx specimens collapse into one cluster) and is reported as an extreme upper bound on multiplicity rather than as a corrected estimate; we explicitly flag it as a stress test, not as a third independent estimator parallel to the literature-pairing best estimate. To propagate pairing uncertainty into the principal effect sizes we run a Monte-Carlo ( N = 2000) over uniform priors on each of the four pairing factors above (with width ± 50–100% of the central value), recompute each headline contrast at fall level on every draw, and report the resulting 95% bootstrap intervals alongside the point estimates (§Pairing-uncertainty propagation). Adjusted multivariate models are fit with statsmodels. Class-binary outcomes (is_iron, is_lunar) are modelled with logistic regression on era + region + fall-status; the era × region interaction is included where the design admits it (the lunar model is fit with main effects only because pre-1969 era × region cells contain near-zero lunar specimens, which causes the full-interaction model to be singular). Log₁₀ recorded mass is modelled with Huber-robust linear regression on the same predictors. Coefficients are reported as adjusted odds ratios (logistic) or dex shifts (robust regression) with 95% confidence intervals. Reproducibility All analyses were performed in Python using NumPy and SciPy. The analysis script (analysis_live.py), the figure-generation script (make_figures.py), and the cleaned, joined CSV (meteorites_clean_live.csv) are provided as supplementary material; intermediate results are serialized to JSON to avoid re-computation during figure generation, but the JSON is regenerated from scratch by analysis_live.py and is never edited by hand. No manual record-level editing or outlier removal was applied after the automated cleaning pipeline; every stratification reported here is a deterministic function of the raw MBDB CSV export of 4 May 2026. Results Catalogue composition The 85,775-record catalogue contains 1,251 Fell records (witnessed falls) and 84,524 Found records (post-fall recoveries) — a 1 : 68 ratio. Year is present for 83,895 records (97.8%) and mass for 85,211 (99.3%); decimal latitude and longitude are present for 52,096 (60.7%). Of the geolocated records, 32,952 (63.3%) lie south of latitude − 60° (the Antarctic stratum), 5,304 (10.2%) lie within the Arabia/Oman rectangle, 3,345 (6.4%) within the Sahara rectangle, and 697 (1.3%) within the Australia rectangle; the remaining 18.8% are distributed across the rest of the inhabited world (Fig. 3). Iron meteorites total 1,417 (1.65% of the full catalogue), lunar meteorites 820 (0.96%), Martian meteorites 419 (0.49%), and stony-irons 614 (0.72%). A factor-of-147 collapse in median recorded mass The most prominent feature of the catalogue is the dramatic drop in characteristic specimen mass before and after 1969. Among the 2,200 pre-1969 records with a recorded mass, the median is 4,400 g (mean log₁₀ mass = 2.55); among the 81,596 post-1969 records, the median is 30.0 g (mean log₁₀ mass = 1.32). The pre/post ratio is 147 (Fig. 2). The two distributions are nearly disjoint: Cliff's δ = 0.81 [95% CI 0.80, 0.82] (derived from Mann–Whitney U = 1.62 × 10⁸ on n₁ = 2,200, n₂ = 81,596), placing the contrast in the "very large" range. Kolmogorov–Smirnov D = 0.670 on log-mass confirms that the shift extends to distributional shape, not only to location. The pre-1969 distribution is broad and centred near 10³ g, dominated by the historical witnessed-fall and historical-find populations of multi-kilogram stones; the post-1969 distribution is sharply peaked near 10¹ g, dominated by the hand-sized fragments that are typical of Antarctic blue-ice and hot-desert prospecting (Fig. 2a). A caveat on the absolute mass numbers is in order. The MBDB Mass field reports recorded mass as catalogued by the contributing agency. The Bulletin documentation explicitly states that recorded masses may be rounded, that they sometimes represent total known weight rather than individual stone mass, and that they should not be treated as authoritative physical measurements. For the catalogue-scale order-of-magnitude contrast we report, this rounding is irrelevant — a 147-fold ratio cannot be moved into the "no-effect" range by any plausible rounding model. The ratio is, however, a recorded-mass ratio rather than a pre-atmospheric ratio. A factor-of-34 collapse in iron-meteorite fraction The iron-meteorite fraction shows an even more dramatic collapse. Among the 2,244 pre-1969 records, 670 (29.86%) are irons; among the 81,651 post-1969 records, 715 (0.88%) are irons (Fig. 2b). The pre/post ratio is 34 , and the contingency χ² = 11,280 (1 df); Cohen's h = 0.97 [95% CI 0.93, 1.01] places this contrast in the "large" range. The simplest reading is that the historical record was selectively populated by easily-recognisable iron specimens picked up by farmers and travellers, while modern systematic searches recover the chondritic stones that dominate the in-falling population. Iron meteorites are believed to constitute 5–6% of the in-falling flux (Bland et al. 1996 ); the witnessed-fall ( Fell ) sub-sample returns an iron fraction of 49 / 1,251 = 3.9%, closer to the flux estimate than either era's Found fraction. Within the Fell sub-sample alone, the pre/post-1969 iron fraction is 41/857 = 4.78% vs 8/394 = 2.03%, a roughly two-fold contraction even after recovery-mode is fixed. Whether this Fell -internal change reflects improving stony-fall recognition over the modern era (a recognition trend) or a small drift in the in-falling class composition is not separable from the data we have here. We treat it as a residual uncertainty on the Fell sub-sample's role as a flux reference, and accept that the Fell baseline is itself not perfectly stationary. Witnessed-fall rate: a small but significant positive trend since 1860 If the post-1969 surge in the catalogue reflected an actual increase in the in-falling meteoroid flux, we would see it in the Fell counts as well as the Found counts. The Fell time series is more complicated than a simple null. Decadal Fell counts from the 1860s through the 2010s are 55, 51, 40, 51, 57, 66, 70, 92, 57, 60, 63, 61, 56, 59, 70, 84 (16 complete decades; mean 62 ± 13 per decade, coefficient of variation 20%; the partial 2020s value of 58 is excluded because the catalogue snapshot terminates in May 2026). The Mann–Kendall test for monotonic trend on this series returns Z = + 2.25, two-tailed p = 0.024: a small but statistically significant positive trend. The 1860s value (55) and the 2010s value (84) bound a roughly 50% rise across 150 years. A Monte-Carlo power analysis (10⁴ Poisson realisations of a 16-decade series with the observed mean and dispersion) gives the test ≈ 94% power against a 50% cumulative trend (∼3%/decade) and ≈ 31% power against a 20% cumulative trend (∼1.3%/decade). The detected trend sits at the edge of what the test can resolve, and is roughly the magnitude that improving reporting infrastructure alone could produce. The simplest reading of this trend is that modern reporting infrastructure — fireball camera networks, smartphone observations, social-media propagation of bolide reports — has improved markedly since the 2000s, raising the apparent Fell count without any change in the underlying flux. A true flux change of comparable magnitude cannot be ruled out from the catalogue alone. The two interpretations sit on opposite sides of the same data point and our analysis cannot distinguish them. Whatever the interpretation, the Fell trend is at least two orders of magnitude smaller than the Found series across the same window. The decadal Found counts are 39, 40, 83, 75, 83, 88, 89, 220, 141, 155, 341, 5,673, 7,963, 14,441, 26,050, 20,599 across the 1860s through the 2010s (the 2020s value of 6,467 is partial). The pattern is unambiguous: a long pre-1960s baseline of 50 to 350 finds per decade, a step of order 30× in the 1970s coincident with the start of ANSMET, and a further step of order 4× in the 1990s and 2000s coincident with the rise of hot-desert prospecting. The catalogue surge is therefore overwhelmingly attributable to recovery and cataloguing effort rather than to any plausible flux change. We can quantify how much of the Fell trend a reporting-only model would absorb. Suppose the observed Fell count is true_flux × r(t) , with r(t) a linear ramp from 1 in 1865 to 1 + δ in 2015. Deflating the observed series by r(t) and re-running the Mann–Kendall test gives Z = + 1.76 (p = 0.079) at δ = 10%, Z = + 0.77 (p = 0.44) at δ = 20%, and Z = + 0.27 (p = 0.79) at δ = 30%. A reporting-improvement multiplier of about 10% across 150 years is therefore enough to render the apparent Fell trend statistically non-significant; a multiplier of 20% eliminates it. The infrastructure changes documented in the literature plausibly produce a multiplier in this range, so we treat the Fell sub-sample as the least-biased available reference rather than as an unbiased one, and we carry the residual uncertainty forward in §Adjusted models. Geographic recovery hotspots The geolocated catalogue is dominated by four prospecting regions (Fig. 3, Table 2 ). Antarctic (latitude < − 60°): n = 32,952, 63.3% of geolocated records, median recorded mass 9.8 g , iron fraction 0.44%, lunar fraction 0.124%, Martian fraction 0.049%. Arabia/Oman (the Rub' al Khali and surrounds): n = 5,304, 10.2%, median 162.0 g, iron 0.23%, lunar 1.45%, Martian 0.38%. Sahara (Northwest Africa, including Morocco): n = 3,345, 6.4%, median 305.6 g, iron 1.20%, lunar 3.65% , Martian 1.14%. Australia (Nullarbor and surrounds): n = 697, 1.3%, median 126.0 g, iron 10.91% , lunar 0.29%, Martian 0. Table 2 Geographic-stratum composition (geolocated records only) Geolocated set n = 52,096; percentages are within the geolocated total. Mass medians exclude records without a recorded mass (within each region). Region n % of geolocated Median mass (g) % Iron % Lunar % Martian Antarctic 32,952 63.3 9.8 0.44 0.12 0.05 Arabia/Oman 5,304 10.2 162.0 0.23 1.45 0.38 Sahara 3,345 6.4 305.6 1.20 3.65 1.14 Australia 697 1.3 126.0 10.91 0.29 0.00 Other 9,798 18.8 151.0 8.64 0.04 0.09 Several patterns stand out. First, the regional median masses span more than a factor of 30: Antarctic specimens are an order of magnitude smaller than hot-desert specimens, which in turn are about half the size of Australian and rest-of-world specimens. This is a direct fingerprint of the prospecting mode — Antarctic blue-ice fields concentrate small stones by ablation, hot-desert pavements expose larger stones over millennial preservation timescales, and the Australian Nullarbor collection was historically dominated by the multi-kilogram irons of the early-20th-century iron-collecting era. Second, the iron fraction varies by a factor of ~ 47 across the regions, from 0.23% in Arabia/Oman to 10.91% in Australia (Fig. 4a) — overwhelmingly a regional-history artefact rather than a regional flux signal. Third, the specimen-level lunar fraction is 11.7 times higher in Arabia/Oman than in the Antarctic across the full catalogue (1.45% vs. 0.124%; χ² = 257.5, Cohen's h = 0.17 [0.14, 0.19]; Fig. 4b). This headline number is, however, sensitive to two confounds we address in turn: differential era sampling, and differential pairing. The Sahara stratum's even higher lunar fraction (3.65%) reinforces the picture that hot-desert pavements have, at the catalogue level, returned a higher recovered-fraction of lunar specimens than blue-ice fields have. Era confound. Antarctic recoveries are concentrated in 1969–1999 (most of the Antarctic stratum is from the ANSMET era) and Arabia/Oman recoveries in 2000+ (most of the Arabia/Oman stratum is from the desert era), so the unstratified comparison conflates regional environment with recovery-era effects. Restricting both regions to a common temporal window: within the 1969–1999 era the Antarctic-vs-Arabia lunar-fraction ratio is 6.75× (Antarctic 0.091%, Arabia 0.617%), and within the 2000 + era it is 8.86× (Antarctic 0.167%, Arabia 1.479%). Pairing confound. The catalogue's specimen-level counts treat paired fragments of single falls as independent events. Antarctic blue-ice fields and hot-desert pavements pair at different rates and for different classes, so a pairing-naïve ratio over-states the environmental component. Applying literature-derived pairing factors — Korotev ( 2012 )'s hot-desert lunar pairing of ≈ 2.5×, Cassidy et al. ( 1992 ) and Harvey ( 2003 ) ANSMET-consensus Antarctic lunar pairing of ≈ 1.5×, the same authors' Antarctic bulk-chondrite pairing of ≈ 3×, and Hutzler et al. ( 2016 ) hot-desert bulk pairing of ≈ 1.75× — gives fall-level lunar fractions of about 0.16% (Antarctic) and 0.97% (Arabia/Oman), a ratio of ≈ 6×. We apply the lunar-specific pairing factor to lunar specimens explicitly and do not substitute the bulk-chondrite factor for them, because lunar specimens pair at lower rates than bulk chondrites in Antarctic strewn fields by a factor of about two; using the 3× bulk factor would over-correct lunar pairing in the Antarctic stratum and is not justified by the literature. An aggressive naming-based clustering (treating every specimen sharing a name prefix and discovery year as a single fall; rule given in Methods ) collapses the contrast to ≈ 1×, but this is reported only as a stress-test lower bound on multiplicity rather than as a third independent estimator. The literature-pairing best estimate of ≈ 6× and the era-restricted 1969–1999 specimen-level contrast (also 6.75×) sit at the same order of magnitude; their numerical agreement may be partly coincidental (both estimators respond to the temporal concentration of recoveries in different post-1969 sub-eras, so they are not fully independent estimators). We conclude that the environmental component of the lunar contrast is plausibly 5–10× under realistic pairing assumptions — substantial, but materially smaller than the 11.7× headline number. The Sahara stratum is intermediate at the specimen level for Arabia/Oman comparisons but is the highest individual-region lunar yield in the catalogue (3.65%, much higher than the previously-reported NASA-snapshot value because the live MBDB has captured the post-2010 surge of Northwest African lunar recoveries). The Australia stratum sits at 0.29%, reflecting the older Australian collecting era predating systematic lunar-meteorite recognition (Korotev 2005 ). Plausible mechanisms for the residual environmental contrast include differential pairing rates (Antarctic specimens are more likely to be paired fragments of a single fall), differential recognition (the fusion crust of lunar feldspathic specimens has higher visual contrast against desert pavement than against ice), and differential dwell-times before recovery (hot-desert pavements integrate the in-falling flux over 10⁴–10⁵ y, whereas Antarctic blue ice integrates over 10⁵–10⁶ y, so a transient enrichment of lunar specimens following a particular lunar-impact ejection event would show up more strongly in the shorter-dwell-time hot-desert sample). The latter mechanism is consistent with the absence of comparable enrichment in iron meteorites. Latitude vs. mass: a dataset-induced spurious physical inference The full catalogue exhibits a strongly negative Spearman correlation between absolute latitude and log-recorded-mass: ρ = −0.393 , n = 52,022; the rank correlation accounts for ρ² ≈ 15% of the mass variance. Naively interpreted, this would imply that the in-falling meteoroid mass distribution is latitude-dependent, with smaller stones falling at high latitudes — a conclusion with no obvious physical mechanism, since meteoroids are decoupled from any latitude-resolved process at the relevant size scales (Bland and Artemieva 2006 ; Halliday et al. 1989 ). The result reverses sign under stratification (Fig. 5). * Restricted to witnessed falls ( Fell ): ρ = +0.031 , n = 1,213 , p = 0.273. The correlation is small and not statistically significant. Within the Fell reference population, there is no meaningful latitude-mass relationship. Excluding the Antarctic stratum: ρ = +0.142 , n = 19,078. Among non-Antarctic recoveries, |φ| is positively correlated with mass — large stones come from mid-latitudes, small stones from low latitudes. This is the population-density signal, modulated by 19th- and 20th-century iron-collection biases. Within the Antarctic stratum alone: ρ = +0.144 (in |φ| terms), reflecting the spatial structure of the blue-ice prospecting fields against the Transantarctic Mountains. Full catalogue (combining the above) *: the negative ρ = −0.393 emerges because the Antarctic stratum, with its distinctive |φ| ≥ 60° and median mass 9.8 g, dominates the high-|φ| portion of the joint distribution and pulls the global rank correlation negative. The headline negative correlation is therefore an instance of dataset-induced spurious physical inference : a sub-population that dominates one corner of a covariate space generates a pooled correlation whose sign is the opposite of the within-stratum correlation, and whose physical interpretation has no mechanistic basis. The pattern is the meteoritic-database analogue of Simpson's paradox in epidemiology and of the ecological-fallacy correlations long known in geographic statistics. The diagnostic procedure is the same as in those fields. Whenever a community catalogue is dominated, in any covariate corner, by a single recovery campaign, any pooled correlation involving that covariate must be re-checked against the least-biased reference sub-sample. In our case, the Antarctic-led ρ = −0.39 collapses to ρ ≈ 0 within the Fell sub-sample, and the negative correlation should be removed from the list of properties of the in-falling meteoroid flux. We expect that other community datasets — paleontological collections, exoplanet catalogues, sky surveys with different completeness limits — show analogous spurious correlations driven by single-collection dominance, and we suggest that explicit Fell -style stratified checks should be standard practice rather than an after-the-fact artefact diagnosis. Three statistically distinct eras The combination of the median-mass collapse, the iron-fraction collapse, and the geographic shift partitions the catalogue into statistically distinct eras (Table 1 , Fig. 6). Pre-1969 (n = 2,244): median mass 4,400 g, iron 29.86%, lunar 0.045%, Martian 0.267%. Antarctic era 1969–1999 (n = 28,323): median mass 15.2 g, iron 0.91%, lunar 0.092%, Martian 0.071%, Antarctic share dominant. Desert era 2000+ (n = 53,328): median mass 45.0 g, iron 0.86%, lunar 1.483% , Martian 0.735% , Arabia/Sahara share dominant. Table 1 Era-stratified composition of the meteorite catalogue All percentages and counts are based on records with Status ∈ {Official, Provisional} and a usable year (n = 83,895 in total). Era n Median mass (g) % Iron % Lunar % Martian Pre-1969 2,244 4,400.0 29.86 0.045 0.267 Antarctic era (1969–1999) 28,323 15.2 0.91 0.092 0.071 Desert era (2000+) 53,328 45.0 0.86 1.483 0.735 Two observations are particularly notable. The Antarctic share of the catalogue peaked in the Antarctic era and has fallen in the desert era — not because Antarctic recoveries have stopped, but because the absolute pace of hot-desert prospecting has overtaken them. The lunar fraction increased by a factor of 16 between the Antarctic era and the desert era (0.092% → 1.483%), and the Martian fraction by a factor of 10 (0.071% → 0.735%), almost entirely because the Sahara, Arabia/Oman and Other (mostly Northwest African) strata over-yield these planetary classes relative to Antarctica. Estimates of Solar-System provenance abundances drawn from the catalogue therefore require weighting by recovery region as well as by recovery era. Pairing-uncertainty propagation The pairing factors used so far are point estimates drawn from the literature. To propagate their uncertainty into the principal effect sizes, we ran a Monte-Carlo ( N = 2000) over independent uniform priors on each of four pairing factors — Antarctic lunar (1.0–2.0×), hot-desert lunar (1.5–4.0×), Antarctic bulk (2.0–4.0×), hot-desert bulk (1.5–2.5×) — and recomputed each headline contrast at fall level on every draw. The results are summarised below. The Arabia/Oman-versus-Antarctic lunar-fraction contrast at fall level has a median of 4.31× with a 95% bootstrap interval of [2.04, 9.73]×. The interval comfortably excludes parity, so the environmental component of the lunar contrast is bounded above zero even under aggressive pairing assumptions; it is also bounded below the 11.7× specimen-level figure under any plausible pairing assumption. The pre-versus-post-1969 mass collapse at fall level has a median of 77× with a 95% interval of [63, 96]× — the 147× specimen-level figure shrinks substantially when post-1969 stones are aggregated into falls, but the contrast remains an order of magnitude even at the upper-pairing end. The pre-versus-post-1969 iron-fraction collapse is 34× with a tight 95% interval; pre-1969 records are dominated by historical non-systematic finds with effective pairing close to 1×, so the iron contrast is essentially insensitive to post-1969 pairing factors. We therefore report the principal contrasts as ranges rather than point estimates wherever pairing matters. Adjusted models: era × region × fall-status The stratified contrasts of § 3.1–§ 3.6 expose every pairwise effect but do not adjust each contrast for the others simultaneously. We fit three adjusted models on the n ≈ 84 k records that carry year, mass, and a region assignment. The first is a logistic regression for P (iron) on era + region + fall-status with a full era × region interaction. The second is a logistic regression for P (lunar) with main effects only (the full interaction is singular because lunar-cell counts in pre-1969 era × region intersections are zero or one). The third is a Huber-robust linear regression on log₁₀ recorded mass against the same predictors with the era × region interaction. Table 3 reports the headline coefficients. The reference cell is Pre-1969 × Other × Found . Three findings stand out. (i) Era is the dominant adjusted predictor of iron fraction. Holding region and fall-status fixed, post-1969 records have ≈ 50× lower iron odds than pre-1969 records (Antarctic-era OR = 0.18, p ≈ 10⁻⁴⁹; Desert-era OR = 0.024, p ≈ 10⁻²⁷¹). The era × region interactions further depress Antarctic-era Antarctic, Sahara, and Australia cells, but the era main effect alone delivers most of the contrast. (ii) Region drives the lunar fraction, era supplies an additional desert-era boost. Holding era and fall-status fixed, the Sahara stratum has 142× higher lunar odds than the Other reference (p ≈ 10⁻²²), Arabia/Oman has 26×, NoLoc has 39×, Antarctic has 4.5×, Australia has 21× (with wider CIs because Australia counts are small). The Desert-era main effect adds roughly 5× over Antarctic-era (p = 0.13) but the bulk of the lunar contrast is regional. (iii) Mass is set by era, with a smaller region modifier. The robust regression returns era main-effect coefficients of − 0.92 ± 0.04 dex (Antarctic era) and − 1.68 ± 0.03 dex (Desert era) — i.e. a 7 to 50-fold drop in adjusted mean log-mass, depending on era. The Antarctic-era × Antarctic-region interaction is a further − 1.54 dex, consistent with the blue-ice fragment population. The Fell main effect is + 0.27 dex (a slight upward bias of witnessed-fall mass over found mass at the same era × region), consistent with witnessed falls preserving larger stones. Table 3 reports the full coefficient set. The adjusted-model framework supplies the answer to the most direct objection an external reviewer can raise: that the stratified contrasts in § 3.1–§ 3.6 are confounded with each other. They are, but the adjusted-OR and adjusted-RLM coefficients show that the headline effects survive simultaneous control. The era effect on iron and on mass is order-of-magnitude even after adjusting for region and fall-status; the region effect on lunar is order-of-magnitude even after adjusting for era and fall-status. The headline contrasts are therefore not a single-confounder artefact, and the recovery-bias structure has independent era and region components. Table 3 Adjusted-model summary. Columns are odds-ratios (logistic models for iron and lunar) and dex coefficients (robust linear regression for log-mass) with 95% CIs. p -values are reported only where ≥ 10⁻³; smaller values are simply marked as ≪ 10⁻³. Predictor Iron OR [CI] Lunar OR [CI] log-mass β (dex) [CI] Antarctic era (vs Pre-1969) 0.18 [0.15, 0.23] 0.37 [0.05, 3.00] −0.92 [− 1.00, − 0.85] Desert era (vs Pre-1969) 0.024 [0.02, 0.03] 4.99 [0.64, 39.07] −1.68 [− 1.73, − 1.62] Antarctic region (vs Other) 1.16 [0.16, 8.29] 4.48 [1.60, 12.55] + 0.10 [− 0.85, + 1.06] Arabia/Oman region (vs Other) 0.34 [0.14, 0.84] 26.1 [9.5, 71.5] −0.70 [− 1.03, − 0.36] Sahara region (vs Other) 0.83 [0.32, 2.18] 142 [52, 387] + 0.11 [− 0.29, + 0.51] Australia region (vs Other) 0.68 [0.49, 0.95] 20.6 [3.6, 117] + 0.22 [+ 0.07, + 0.37] Fell (vs Found) 0.07 [0.05, 0.10] unstable (rare event) + 0.27 [+ 0.21, + 0.34] Antarctic era × Antarctic 0.024 [0.001, 0.18] — −1.54 [− 2.50, − 0.59] Desert era × Sahara 0.93 [0.32, 2.70] — + 0.64 [+ 0.24, + 1.04] Desert era × Arabia/Oman 0.12 [0.03, 0.46] — + 1.01 [+ 0.68, + 1.35] Robustness checks We verified that the qualitative results survive several reasonable perturbations of the analysis. First, restricting to Official records (excluding the 6,180 Provisional records) changes the iron-fraction percentages by less than 0.05 percentage points and the median masses by less than 1%. Second, moving the Antarctic-era boundary to 1965, 1970, or 1975 changes the era-summary percentages by less than 0.5 percentage points; the same is true for moving the desert-era boundary across 1995, 2000, 2005. Third, redefining the Arabia/Oman rectangle to exclude the Persian Gulf coast (latitude ≥ 22° rather than ≥ 15°) reduces the Arabia/Oman count from 5,304 to about 4,800 but raises the lunar fraction from 1.45% to roughly 1.65%, sharpening rather than weakening the contrast with Antarctica. Fourth, the Spearman ρ between |φ| and log-mass is stable under bin-aggregating mass at 0.25-dex resolution and under dropping the smallest 1% and largest 1% of masses. Discussion What the catalogue *can* and *cannot* be used for The Fell sub-sample is the appropriate baseline for flux-based questions. With n = 1,251 and a near-stationary ≈ 62-per-decade rate over the past 150 years (with a small positive trend documented in §Witnessed-fall rate), it is the least-biased reference population available for the in-falling meteoroid flux at the kilogram-mass scale, suitable for flux-based, class-frequency, or population-mean inference. Its biases — population-density modulation of observational opportunity, geographic variation in reporting culture, under-recovery of small specimens, and the smaller-magnitude recognition trend that produces the documented Mann–Kendall significance — are smaller than the recovery-environment biases that dominate the Found sub-sample. The numbers line up. Its iron fraction (3.9%) falls within the 4–7% range estimated by independent flux models (Bland et al. 1996 ; Bland and Artemieva 2006 ; Halliday et al. 1989 ); its median mass (2.6 kg) and class distribution are likewise consistent with those models within the resolution of the available statistics. The full catalogue, by contrast, is dominated by recovery effort, recovery environment, and recovery-era effects. None of the following are reliably extractable from the full catalogue without explicit recovery-bias correction: the absolute flux of meteoroids reaching Earth's surface as a function of class; the latitudinal dependence of any property of the in-falling flux; the temporal variation of the in-falling flux (which the Fell time series places strong upper limits on, contradicting any claim of order-of-magnitude flux variability over the past 200 years); and the relative abundance of rare classes (lunar, Martian, ungrouped achondrites) without weighting by recovery region. The full catalogue is, however, well-suited for searches for petrologically novel material (where coverage matters more than statistical representativeness), for pairing studies and strewn-field reconstructions within recovery regions, and for studies of recovery itself — i.e. of how prospecting environment, climate, and collector behaviour interact to shape the apparent record (the present paper being one such study). Implications for asteroid-source inversion A common use of the catalogue is to invert the relative abundance of meteorite classes against the relative abundance of asteroid spectral types in order to constrain meteorite-asteroid genetic links (Burbine 2017 ; Bottke et al. 2007 ). Such inversions tend to assume — implicitly more often than explicitly — that the catalogue's class composition reflects the in-falling flux. The factor-of-34 collapse in iron-meteorite fraction across the 1969 boundary, and the order-of-magnitude variation in iron fraction across recovery regions, suggest that this assumption is violated by orders of magnitude in the pre-1969 sub-sample and by factors of several even within post-1969 hot-desert collections. Inversions that pool all eras and regions therefore inherit the bias. The remedy is direct. Perform the inversion against the Fell sub-sample only, accepting the smaller sample size in exchange for a less-biased estimator. A stronger approach combines the Fell sub-sample for absolute calibration with the Found sub-sample for class-internal substructure, weighting by recovery region and era to remove the gross biases we have documented. We have not attempted this multivariate weighting. The present scope is restricted to quantifying the magnitude of the bias and supplying the statistical infrastructure for follow-up work. Implications for lunar and Martian sample yields Lunar and Martian meteorites are scientifically valuable beyond their statistical weight because they provide a planetary surface sample at near-zero mission cost. The 11.7× contrast in lunar fraction between Arabia/Oman and the Antarctic, together with the large absolute counts in the desert era (about 70 lunar specimens recovered in Arabia/Oman post-2000 and a comparable number from the Sahara stratum vs. about 25 from the Antarctic), indicates that hot-desert collections have historically contributed a higher catalogued fraction of lunar specimens per recovered stone than Antarctic collections have. The Martian fractions show a similar but slightly weaker contrast (Arabia/Oman 0.38% vs. Antarctic 0.05%; Sahara 1.14% vs. Antarctic 0.05%). The mechanistic reasons are several — better visual contrast against desert pavement, shorter terrestrial dwell times that preserve the chemistry of a transient lunar-impact event, lower pairing rates per recovered specimen — and the empirical catalogue-level contrast is large. We do not have a search-effort denominator (km² covered, person-hours, total mass collected) that would be needed to convert the catalogue contrast into a per-effort yield, so we phrase the contrast as a recovered-fraction contrast rather than as a yield-per-effort recommendation; that conversion is left to operations-level studies that have access to the relevant denominator. That said, the Antarctic record provides a unique long-dwell-time integration of the planetary-meteorite flux that hot deserts cannot match. For studies that depend on the longest-possible delivery-time integral (e.g. searches for very-old planetary surface material, or for material from a particular impact event millions of years past) the Antarctic strata remain irreplaceable. The pre-1969 record as a window into pre-modern observation The pre-1969 record contains 2,244 specimens accumulated over ~ 250 years. Its 30% iron fraction and 4.4-kg median mass tell us less about the in-falling flux than they tell us about what humans noticed and bothered to keep in the pre-systematic era: visually distinctive metallic stones large enough to be picked up by hand and small enough to be carried by a single farmer. Iron meteorites, in particular, were prized as a free source of nickel-iron ore and as curiosities throughout the Bronze, Iron, and Industrial ages (King 1865 ; Buchwald 1975 ), and the historical record is enormously skewed toward them. The witnessed-fall sub-sample of the pre-1969 era, however, has an iron fraction of 4.78% (41 of 857), reduced further to 2.03% (8 of 394) in the post-1969 witnessed-fall sub-sample — both far below the 30% pre-1969 Found figure, confirming that the historical 30% iron fraction is a feature of the pre-1969 Found sub-population only and not of the underlying in-falling flux. This is itself an empirical lesson: even within a single era, the recovery channel (witnessed vs. found) imposes a class-composition bias that swamps any era-to-era physical signal. A note on historical chronologies The catalogue extends back to 860 CE (the Nōgata fall, Japan) and includes a small handful of mediaeval and early-modern witnessed falls. These records are valuable as cultural-historical evidence but are statistically negligible: there are 14 records dated before 1700 in the entire catalogue. The pre-1700 sub-sample is dominated by the few falls of sufficient magnitude or social impact to be recorded in surviving documents, and any quantitative claim about pre-1700 rates is dominated by Poisson noise on a one-digit count base. We have included the pre-1700 records in our time series for completeness but draw no inference from their decadal counts. Limitations Several limitations should be borne in mind. The catalogue continues to grow; the precise era-end percentages will shift as further 2020s recoveries accumulate, although the qualitative pattern (Antarctic-era → desert-era transition centred near 2000) is sufficiently well-established that further accessions will reinforce rather than reverse the structure. Our class taxonomy collapses 619 Type strings into a small number of bins; users interested in the substructure within iron meteorites (e.g. IAB-MG vs. IIIAB) or within carbonaceous chondrites (e.g. CM vs. CR vs. CV) should re-derive their own taxonomy from meteorites_clean_live.csv. The four geographic strata are coarse rectangles; users interested in finer regional differentiation (e.g. ANSMET vs. NIPR vs. EUROMET, or Acfer vs. Dar al Gani vs. Hammadah al Hamra) will need the original Bulletin metadata, which carries finer locality fields than we have used here. The most consequential limitation concerns pairing. The analysis treats individual catalogue entries as independent specimens, but fragmented falls — particularly in Antarctic blue-ice fields, where one bolide can deposit many small stones across a strewn field — violate this assumption, and pairing factors differ across regions and across classes. We did not have access to the Bulletin's authoritative pairing decisions, so we report a sensitivity range rather than a single corrected number. A literature-derived pairing-factor correction, applied per region and per class, gives an environmental component of the Antarctic-vs-Arabia/Oman lunar contrast of approximately 6×, against an unstratified 11.7× upper bound. An aggressive naming-based clustering collapses the contrast to about 1× as a stress-test lower bound; this is not a believable estimate but bounds the family of plausible pairings. The literature-pairing best estimate of ≈ 6× and the era-restricted 1969–1999 specimen-level contrast (6.75×) sit at the same order of magnitude; their numerical agreement should not be over-interpreted, since both estimators respond to the temporal concentration of recoveries in different post-1969 sub-eras, so they are correlated estimators rather than independent ones. We therefore conclude that the environmental component of the Antarctic-vs-Arabia/Oman lunar contrast is plausibly 5–10× under realistic pairing assumptions — substantial, but materially smaller than the 11.7× headline number, and bracketed rather than pinned down. The mass and iron-fraction collapses, by contrast, are large enough (Cliff's δ = 0.81; Cohen's h = 0.97) that no plausible pairing correction reverses them; only their interval estimates change. A definitive pairing-corrected re-analysis using the Bulletin's published pairing decisions remains outstanding and would tighten all of these estimates. Mass is the second worry. Small post-1969 specimens recovered from hot deserts have integrated 10⁴–10⁵ years of weathering and ablation; their pre-atmospheric masses were almost certainly larger than what we measure today, so part of the apparent post-1969 mass collapse is recovered-mass attrition rather than recovered-population shift. The Bulletin's Mass field is a recorded mass, not a pre-atmospheric mass, and is documented by the Bulletin maintainers as approximate and sometimes representative of total known weight rather than individual stone mass. The 147-fold catalogue-scale contrast we report is too large to be erased by any plausible mass-rounding model. Its precise magnitude is a recorded-mass contrast rather than an in-falling-population contrast. A further limitation concerns the Fell time series itself. The Mann–Kendall result on this series returns Z = + 2.25, p = 0.024 — a small positive trend that we attribute provisionally to improving modern reporting infrastructure (fireball camera networks, smartphone observations, social-media propagation) but that we cannot, with the catalogue alone, distinguish from a true small flux change. The Monte-Carlo power analysis bounds the size of any such effect: a 50% cumulative trend would have been detected with 94% statistical power, and the trend we now detect is at the edge of that 50% threshold. The Fell sub-sample is therefore the least-biased reference available, but it is not free of secular biases, and any quantitative use of it should accommodate the residual uncertainty documented here. Future work Several follow-up analyses are suggested by the present scan. First, a quantitative pairing-corrected re-analysis of the Antarctic and hot-desert lunar/Martian fractions, using the published Bulletin pairing decisions, would clarify how much of the 11.7× lunar contrast is a true delivery-time effect and how much is a recognition / pairing artefact. Second, a temporal analysis of the Fell sub-sample's class composition over the 19th and 20th centuries would test whether recognition of new classes (e.g. ureilites, brachinites) is real flux or recognition-driven, and would also help separate the small Fell trend documented here into its reporting-improvement and flux-change components. Third, a forward-model of expected Found class fractions given a hypothesised in-falling flux and a parametric recovery-bias model could be fit to our era- and region-stratified data, providing the first formal estimate of the human fingerprint on the catalogue against which future recovery campaigns can be benchmarked. Fourth, an adjusted regression model — log-mass and class-binary outcomes regressed jointly on era × region × fall-status — would convert the per-question contrasts we report into a single set of adjusted odds ratios; we leave this synthesis for a follow-up paper. Conclusions A stratified scan of the 85,775-record live Meteoritical Bulletin Database, supplemented with adjusted-regression and pairing-uncertainty Monte-Carlo audits, confirms at population scale that the apparent meteorite record is dominated by recovery effort rather than by in-falling flux. The principal findings are these. First, the mass and iron contrasts. The median recorded mass collapses by a factor of 147 between pre-1969 and post-1969 records at specimen level (4,400 g → 30 g; Cliff's δ = 0.81). Pairing-uncertainty Monte-Carlo deflates the fall-level mass collapse to a median of 77× (95% CI [63, 96]×); the iron-meteorite fraction collapses by 34× at specimen level and is essentially insensitive to pairing because pre-1969 records pair near 1×. Huber-style regression confirms that era is the dominant adjusted predictor of mass (post-1969 dex shifts of − 0.92 ± 0.04 and − 1.68 ± 0.03) and that the era main effect on iron survives simultaneous control for region and fall-status (Antarctic-era OR = 0.18, Desert-era OR = 0.024, both p ≪ 10⁻³). Now the witnessed-fall side. The Fell sub-sample sits at 62 ± 13 events per decade since the 1860s; the Mann–Kendall test returns Z = + 2.25, p = 0.024, a small but significant positive trend that a 10% reporting-improvement multiplier across 150 years already renders non-significant. We treat the Fell sub-sample accordingly as a least-biased reference rather than an unbiased one. The post-1969 catalogue surge is overwhelmingly attributable to recovery and cataloguing effort rather than to any plausible flux change. Second, geographic recovery hotspots impose a strong second-order bias on the apparent record. The headline 11.7× Arabia-vs-Antarctic specimen-level lunar contrast shrinks to a pairing-MC fall-level median of 4.31× (95% CI [2.04, 9.73]×) — comfortably bounded above parity but materially smaller than the unstratified figure. Adjusted logistic regression confirms that region is the dominant predictor of lunar fraction (Sahara OR = 142, Arabia/Oman OR = 26, vs Other; p ≪ 10⁻³), with an additional Desert-era main-effect boost of order 5×. Both era and region effects therefore survive simultaneous control. Third, the catalogue's headline negative latitudinal-mass correlation (Spearman ρ = −0.39 against |φ|) collapses to ρ ≈ 0 within the witnessed-fall sub-sample. This is an instance of dataset-induced spurious physical inference. A sub-population that dominates one corner of a covariate space generates a pooled correlation whose sign is the opposite of the within-stratum correlation, with no mechanistic basis. Downstream analyses claiming a latitudinal-flux dependence should restrict themselves to the Fell reference population, and analogous Fell -style stratified diagnostics should be standard practice for any community catalogue dominated by a single recovery campaign in any covariate corner. The database has been doubled, and then doubled again, by two distinct human prospecting campaigns. Their statistical fingerprints — small-stone Antarctic blue-ice recoveries (1969–1999) and hot-desert pavements (2000+) — are large enough to dominate every descriptive statistic computable from the full catalogue. The witnessed-fall sub-sample remains the appropriate reference for flux-based questions; the Found sub-sample remains essential for material searches and for studies of the recovery process itself. The two should not be pooled without explicit bias correction. We release all input data, the joined and cleaned catalogue, the analysis pipeline, the adjusted-regression results, the pairing-uncertainty Monte-Carlo, and the figure-generation script as supplementary material so that every test reported here can be reproduced from raw data in under one minute, and so that the analysis can be re-run as the Bulletin Database is updated. Declarations Competing interests The author declares no competing interests. Author contributions All sections of this paper — study design, data acquisition and cleaning, statistical analysis, software, figure preparation, and manuscript writing — were carried out by the corresponding author. Data availability The Meteoritical Bulletin Database is publicly available. The Lunar and Planetary Institute hosts the live database at https://www.lpi.usra.edu/meteor/metbull.php , including a CSV export option as of April 2026. The CSV snapshot used here was retrieved on 4 May 2026 and is included with the supplementary material for this paper. The cleaned, joined and class-tagged dataset (meteorites_clean_live.csv), the analysis script (analysis_live.py), the figure-generation script (make_figures.py), and the cached numerical results (results_live.json) are provided alongside. Code availability The pipeline is small. Analysis (Python 3.11, NumPy 1.26, SciPy 1.13) is provided as analysis_live.py and figure generation as make_figures.py, with all intermediate results cached to results_live.json for replay. Total runtime is roughly 60 seconds on a standard laptop. No external services or random seeds are required. References Bevan AWR, Bindon P (1996) Australian Aborigines and meteorites. Records Western Australian Museum 18:93–101 Bevan AWR, Bland PA, Jull AJT (1998) Meteorite flux on the Nullarbor region, Australia. Geol Soc Spec Publ 140:59–73 Bischoff A, Geiger T (1995) Meteorites from the Sahara: find locations, shock classification, degree of weathering and pairing. Meteoritics 30:113–122 Bland PA, Smith TB, Jull AJT, Berry FJ, Bevan AWR, Cloudt S, Pillinger CT (1996) The flux of meteorites to the Earth over the last 50,000 years. Mon Not R Astron Soc 283:551–565 Bland PA, Artemieva NA (2006) The rate of small impacts on Earth. Meteorit Planet Sci 41:607–631 Bottke WF, Vokrouhlický D, Nesvorný D (2007) An asteroid breakup 160 Myr ago as the probable source of the K/T impactor. Nature 449:48–53 Buchwald VF (1975) Handbook of Iron Meteorites: Their History, Distribution, Composition and Structure. University of California Press Burbine TH (2017) Asteroids: Astronomical and Geological Bodies. Cambridge University Press Cassidy W, Harvey RP, Schutt J, Delisle G, Yanai K (1992) The meteorite collection sites of Antarctica. Meteoritics 27:490–525 Connolly HC, Smith C, Benedix G, Folco L, Righter K, Zipfel J, Yamaguchi A, Aoudjehane HC (2007) The Meteoritical Bulletin, 92, 2007 September. Meteorit Planet Sci 42:1647–1694 Evatt GW, Smedley ARD, Joy KH, Hunter L, Tey WH, Abrahams ID, Gerrish L (2020) The spatial flux of Earth's meteorite falls found via Antarctic data. Geology 48:683–687 Grady MM (2000) Catalogue of Meteorites, 5th edn. Cambridge University Press Grossman JN (1999) The Meteoritical Bulletin, 83, 1999 July. Meteorit Planet Sci 34:A169–A186 Halliday I, Griffin AA, Blackwell AT (1989) The flux of meteorites on the Earth's surface. Meteoritics 24:173–178 Harvey R (2003) The origin and significance of Antarctic meteorites. Chemie der Erde — Geochem 63:93–147 Hofmann BA, Lorenzetti S, Eugster O, Krähenbühl U, Herzog G, Serefiddin F, Gnos E, Eggimann M, Wasson JT (2009) The Twannberg (Switzerland) IIG iron meteorites: mineralogy, chemistry, and CRE ages. Meteorit Planet Sci 44:187–199 Hutzler A, Gattacceca J, Rochette P, Braucher R, Carro B, Christensen EJ, Cournède C, Gounelle M, Laridhi Ouazaa N, Martinez R, Valenzuela M, Warner M, Bourlès D (2016) Description of a very dense meteorite collection area in western Atacama: insight into the long-term composition of the meteorite flux to Earth. Meteorit Planet Sci 51:468–482 Kendall MG (1975) Rank Correlation Methods, 4th edn. Charles Griffin, London King CW (1865) The Natural History of Precious Stones and of the Precious Metals. Bell and Daldy, London Korotev RL (2005) Lunar geochemistry as told by lunar meteorites. Chemie der Erde — Geochem 65:297–346 Korotev RL (2012) Lunar meteorites from Oman. Meteorit Planet Sci 47:1365–1402 Mann HB (1945) Nonparametric tests against trend. Econometrica 13:245–259 Yamaguchi A, Kojima H, Imae N (2002) Antarctic meteorites collected by NIPR's expeditions to the Asuka site. Antarct Meteor Res 15:1–11 Yoshida M, Ando H, Omoto K, Naruse R, Ageta Y (1971) Discovery of meteorites near Yamato Mountains, East Antarctica. Antarct Record 39:62–65 Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9687218","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":638741022,"identity":"8c88ed89-1dcb-4051-9b4a-2b8028cea8fa","order_by":0,"name":"Rongzhen Dai","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+ElEQVRIiWNgGAWjYDCCA4wNYNqAgYH5wQcgg42dBC1shjNAWpgJaoHSQC0M0jwgFiEtfMcPt334uKPW3py994Cxza9t8nzMDIwfPubg1iJ5JrF55swzx5kte84lPM7tu23YxszALDlzG24tBgcSm5l5246xGdzIMTDO7bnNCNTCxsyLT8v5h2AtPAb33xhIW/bcties5QbYlhoJgxs8BtIMP24nEtQieeNhM+PMtgMGBmdyzAx7G24ntzEzNuP1C9/59McMH9vq7A2OnzF+8OPPbdv57c0HP3zEowUKDkMoxjYw2UBQPRDUQek/xCgeBaNgFIyCkQYAvw9Uo5YlMo0AAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0005-7165-0080","institution":"Independent Researcher","correspondingAuthor":true,"prefix":"","firstName":"Rongzhen","middleName":"","lastName":"Dai","suffix":""}],"badges":[],"createdAt":"2026-05-12 06:10:38","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9687218/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9687218/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109274209,"identity":"0d823a08-bdd8-455a-8952-85dc668f322b","added_by":"auto","created_at":"2026-05-14 14:39:29","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":205422,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eThe recovery surge is in \u003c/em\u003eFound\u003cem\u003e, not \u003c/em\u003eFell\u003cem\u003e. (a) Decadal counts of recovered meteorites since 1700 on a log scale. The Antarctic era (1969–1999) and the Desert era (2000+) are shaded. \u003c/em\u003eFound\u003cem\u003e counts (blue) increase by more than two orders of magnitude across the 20th century; \u003c/em\u003eFell\u003cem\u003e counts (red) sit on a near-flat baseline of about 60 per decade. (b) The same \u003c/em\u003eFell\u003cem\u003e counts on a linear scale, with the 1860s–2010s mean (62 per decade) marked. The Mann–Kendall trend test on this 16-decade series returns \u003c/em\u003eZ\u003cem\u003e= +2.25, \u003c/em\u003ep* = 0.024: a small but statistically significant positive trend, consistent with documented improvements in modern reporting infrastructure. The post-1969 catalogue surge remains dominated by recovery effort.\u003c/p\u003e","description":"","filename":"Fig1timeline.png","url":"https://assets-eu.researchsquare.com/files/rs-9687218/v1/e851b1d5b9bb4ecc9dc2ddb8.png"},{"id":109274374,"identity":"da63e657-9392-4efc-b8a6-ec2e1807a501","added_by":"auto","created_at":"2026-05-14 14:39:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":223158,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMass and iron-fraction collapse across the 1969 boundary.\u003c/strong\u003e (\u003cstrong\u003ea\u003c/strong\u003e) Log-mass histograms for the three eras; the pre-1969 distribution is broad and centred near 10³ g, the post-1969 distributions sharply peaked near 10¹ g. KS-test on pre-1969 vs. post-1969 returns \u003cem\u003eD\u003c/em\u003e = 0.670 (Cliff's \u003cem\u003eδ\u003c/em\u003e = 0.81). (\u003cstrong\u003eb\u003c/strong\u003e) Median recorded mass (left axis, log scale) and iron-meteorite fraction (right axis) for the three eras. The 147-fold mass drop and 34-fold iron-fraction drop are both clear.\u003c/p\u003e","description":"","filename":"Fig2eramassiron.png","url":"https://assets-eu.researchsquare.com/files/rs-9687218/v1/1810296ccb63c7efdaf44c76.png"},{"id":109274196,"identity":"046becd9-d67b-436e-b91e-26df77c945da","added_by":"auto","created_at":"2026-05-14 14:39:28","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":505213,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGeographic distribution of recoveries.\u003c/strong\u003e (\u003cstrong\u003ea\u003c/strong\u003e) World-map scatter of all geolocated recoveries by era, with the four prospecting hotspot rectangles (Antarctic, Arabia/Oman, Sahara, Australia) overlaid. The dominance of the Antarctic stratum is visually obvious, as are the desert-era Arabia/Oman and Sahara concentrations. (\u003cstrong\u003eb\u003c/strong\u003e) Bar chart of regional shares of the geolocated catalogue (52,096 records).\u003c/p\u003e","description":"","filename":"Fig3geography.png","url":"https://assets-eu.researchsquare.com/files/rs-9687218/v1/ee7bd13058a33f6105ca5530.png"},{"id":109274496,"identity":"2afb0254-6e98-41eb-8c73-71947f787d7a","added_by":"auto","created_at":"2026-05-14 14:40:09","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":178653,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClass composition by region.\u003c/strong\u003e (\u003cstrong\u003ea\u003c/strong\u003e) Specimen-level lunar, Martian and iron fractions for the four prospecting regions plus Other. The Australia stratum's anomalously high iron fraction (10.9 %) reflects the early-20th-century Nullarbor iron-collection era. (\u003cstrong\u003eb\u003c/strong\u003e) Lunar fractions in Antarctica vs. Arabia/Oman: 0.12 % vs. 1.45 %, an 11.7× specimen-level contrast (χ² = 257.5, Cohen's \u003cem\u003eh\u003c/em\u003e = 0.17). The headline 11.7× is partly era-amplified (era-restricted contrasts: 6.75× for 1969–1999, 8.86× for 2000+) and partly pairing-amplified (literature-pairing-corrected fall-level contrast: ≈ 6×); the environmental component is plausibly 5–10× under realistic pairing assumptions.\u003c/p\u003e","description":"","filename":"Fig4classbyregion.png","url":"https://assets-eu.researchsquare.com/files/rs-9687218/v1/b9a41ee21478e2a9869d19e2.png"},{"id":109274208,"identity":"2bc3e954-867a-4802-9199-302041e4dc74","added_by":"auto","created_at":"2026-05-14 14:39:29","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":606975,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe latitude–mass correlation is a search-bias artefact.\u003c/strong\u003e (\u003cstrong\u003ea\u003c/strong\u003e) Hexbin density of all geolocated recoveries with measured mass in (|latitude|, log-mass) space. Spearman ρ = −0.393 (ρ² ≈ 15 % of mass variance); the apparent gradient is generated by the high-|φ|, small-mass Antarctic stratum. (\u003cstrong\u003eb\u003c/strong\u003e) The same plot for the witnessed-fall (\u003cem\u003eFell\u003c/em\u003e) sub-sample only: ρ = +0.031, \u003cem\u003ep\u003c/em\u003e = 0.273 — essentially flat. The full-catalogue gradient does not survive the \u003cem\u003eFell\u003c/em\u003e control and is therefore not a physical signal.\u003c/p\u003e","description":"","filename":"Fig5latmass.png","url":"https://assets-eu.researchsquare.com/files/rs-9687218/v1/378b623d75737aa5c5ba5620.png"},{"id":109274365,"identity":"bfe8ed33-d739-43e9-bdd5-cdddbcba5c22","added_by":"auto","created_at":"2026-05-14 14:39:46","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":103364,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComposition of the meteorite record across three eras.\u003c/strong\u003e Bar chart (symmetric-log axis) of iron, lunar, Martian percentages for each era, illustrating the three-era structure: pre-1969 dominated by historical iron biases (29.9 % iron), Antarctic era dominated by Antarctica, Desert era with elevated lunar (1.48 %) and Martian (0.74 %) fractions.\u003c/p\u003e","description":"","filename":"Fig6eracomposition.png","url":"https://assets-eu.researchsquare.com/files/rs-9687218/v1/34d6395caa79ac6e89fb6009.png"},{"id":109274637,"identity":"8f3041f6-8c10-4acc-b0a0-ab4b3272e963","added_by":"auto","created_at":"2026-05-14 14:40:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2024983,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9687218/v1/56df81da-93d7-4693-875f-2b5ee2ba05e5.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eThe Human Fingerprint on the Meteorite Record: A Full-Database Scan of the Meteoritical Bulletin (n = 85,775)\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe Meteoritical Bulletin Database is routinely used as a primary data source for meteoroid flux estimates, asteroid-source inversions, and planetary-meteorite census studies (Bland et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e1996\u003c/span\u003e; Burbine \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Evatt et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Yet the database conflates specimens recovered under fundamentally different search regimes \u0026mdash; pre-1969 opportunistic collection, post-1969 Antarctic blue-ice programmes, and post-2000 hot-desert prospecting \u0026mdash; without flagging which regime produced each entry. When downstream analyses treat the combined catalogue as a representative sample of the in-falling flux, they inherit biases that we show here to be of order 10\u0026ndash;100\u0026times; in mass, class composition, and geographic provenance. No prior study has quantified these biases at full-database resolution with effect sizes, adjusted regression, and pairing-uncertainty propagation. We provide that quantification.\u003c/p\u003e \u003cp\u003eMeteorites carry information no other extraterrestrial sample can. They are the only macroscopic extraterrestrial matter routinely available for terrestrial laboratory analysis, and the catalogue of recovered specimens is the empirical foundation for almost every quantitative claim made about the flux, composition and origin of meteoroids reaching Earth (Grossman \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e1999\u003c/span\u003e; Connolly et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Burbine \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). The Meteoritical Bulletin Database, maintained by the Meteoritical Society at the Lunar and Planetary Institute, is the canonical curated catalogue. It combines two fundamentally different populations. The first is \u003cem\u003ewitnessed falls\u003c/em\u003e (the \u003cem\u003eFell\u003c/em\u003e category), where a fireball and an associated stone are unambiguously linked. The second is \u003cem\u003efinds\u003c/em\u003e (the \u003cem\u003eFound\u003c/em\u003e category), in which an extraterrestrial origin is established post-recovery on petrological and geochemical grounds (Bevan and Bindon \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1996\u003c/span\u003e; Grady \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2000\u003c/span\u003e). Witnessed falls are the least-biased available reference for the in-falling flux \u0026mdash; modulated by population density, observational opportunity, and reporting culture, but free of the recovery-environment effects that dominate finds \u0026mdash; while finds depend on where humans have looked and on how the local environment preserves and exposes meteoritic material.\u003c/p\u003e \u003cp\u003eThe catalogue grew slowly through the 18th, 19th and early 20th centuries, doubled roughly every 25 years up to about 1960, and then underwent two abrupt accelerations. The first was the discovery, in 1969, that the blue ice fields of East Antarctica preserve and concentrate meteorites by glacial flow against the Transantarctic Mountains and ablate them back to the surface, giving rise to the U.S. Antarctic Search for Meteorites programme (ANSMET) and to its Japanese (NIPR), Italian (PNRA), Chinese (CHINARE) and European (EUROMET) counterparts (Cassidy et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e1992\u003c/span\u003e; Harvey \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Yamaguchi et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). The second, beginning in the late 1990s, was the systematic prospecting of hot-desert pavements in the Sahara, on the Arabian Peninsula, and in Oman, where flat regs and sebkhas combine with an arid climate to preserve dark cosmic stones for tens of thousands of years (Bischoff and Geiger \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e1995\u003c/span\u003e; Hofmann et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Hutzler et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). These two episodes between them deposited some 80,000 new specimens into the database \u0026mdash; roughly forty times the cumulative count of the entire previous two centuries.\u003c/p\u003e \u003cp\u003eThe growth has been understood since the 1990s as a recovery effect, not a flux effect (Bevan et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Halliday et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Bland and Artemieva \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). The new specimens are systematically smaller, less iron-rich, and concentrated in places that pre-1969 collectors never visited. Yet the magnitude and structure of the bias is rarely quantified at the database scale, and downstream uses \u0026mdash; flux modelling, asteroid-source inversion, isotopic-anomaly sampling \u0026mdash; can inadvertently conflate catalogue composition with in-falling composition if recovery regime is not explicitly modelled. We quantify the magnitude of the bias at full-database resolution, with a stratified scan of the 85,775-record catalogue structured to mirror the questions a downstream user would ask before drawing inferences from any sub-sample.\u003c/p\u003e \u003cp\u003e(i) How much of the post-1969 explosion is real flux vs. recovery effort? (We compare the \u003cem\u003eFell\u003c/em\u003e and \u003cem\u003eFound\u003c/em\u003e time series.) (ii) How much smaller are post-1969 specimens, and how much less iron-rich, than pre-1969 specimens? (iii) Where, geographically, do the new specimens come from, and how does the prospecting environment shape the apparent abundance of rare classes (lunar, martian, iron)? (iv) Is the apparent latitudinal trend in recorded mass a physical signal or a search-bias artefact? (v) Do the two recovery surges define statistically distinct \u003cem\u003eeras\u003c/em\u003e in the catalogue?\u003c/p\u003e \u003cp\u003eThe qualitative claim that the catalogue is recovery-biased has been recognised at least since the 1990s (Bevan et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Halliday et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Bland and Artemieva \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Our contribution is \u003cb\u003equantitative and structural\u003c/b\u003e. We report effect sizes (Cliff's \u003cem\u003eδ\u003c/em\u003e for distributional contrasts, Cohen's \u003cem\u003eh\u003c/em\u003e for proportional contrasts) with bootstrap confidence intervals throughout. We decompose the headline biases into era \u0026times; region interactions to expose previously-conflated signals. We bracket the residual contribution from differential pairing under an explicit sensitivity analysis with a literature-pairing best estimate alongside two upper- and lower-bound stress tests. We provide a reproducible full-snapshot quantification of the long-recognised observation that the apparent negative correlation between recorded mass and absolute latitude is a sampling artefact that vanishes within the witnessed-fall sub-sample. We release a deterministic single-script analysis pipeline so the audit can be re-run as the Bulletin is updated. The principal finding is that any descriptive statistic computed from the combined \u003cem\u003eFound+Fell\u003c/em\u003e sample is driven by recovery effort to first order, while the witnessed-fall sub-sample provides a substantially less-biased reference for population-level inference.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData source\u003c/h2\u003e \u003cp\u003eWe used a CSV export of the live Meteoritical Bulletin Database (MBDB) maintained by the Lunar and Planetary Institute on behalf of the Meteoritical Society, retrieved from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.lpi.usra.edu/meteor/metbull.php\u003c/span\u003e\u003cspan address=\"https://www.lpi.usra.edu/meteor/metbull.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e on 4 May 2026. The export carries 86,648 rows. Of these, 85,775 carry a Status flag of \u003cem\u003eOfficial\u003c/em\u003e (n\u0026thinsp;=\u0026thinsp;79,595) or \u003cem\u003eProvisional\u003c/em\u003e (n\u0026thinsp;=\u0026thinsp;6,180); we retain both because Provisional entries are typically catalogued in the Bulletin's published numbered issues and become Official on routine review, with conversion rates close to unity over multi-year windows. Records with Status\u0026thinsp;=\u0026thinsp;\u003cem\u003ePseudo\u003c/em\u003e, \u003cem\u003eDoubtful\u003c/em\u003e, or \u003cem\u003eCrater\u003c/em\u003e (n\u0026thinsp;=\u0026thinsp;682 in total) are excluded, as their petrological status is either questioned or non-meteoritic. The MBDB schema differs from the older NASA Meteorite Landings tabulation; the field mapping is Code \u0026rarr; identifier, Status \u0026rarr; nametype, Fall \u0026rarr; Fell/Found (Y or Yc \u0026rarr; Fell), Year \u0026rarr; year, Type \u0026rarr; recclass, Mass \u0026rarr; mass(g), Lat/Long \u0026rarr; reclat/reclong, with a separate Boolean Antarctic flag we cross-check against the latitude-based assignment.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eCleaning and filtering\u003c/h3\u003e\n\u003cp\u003eWe coerced \u003cem\u003eYear\u003c/em\u003e to a numeric integer and discarded a small number of records with corrupted year fields (one record dated to 2101 CE; two records dated before 500 CE). Table A1 in the supplementary material lists the three dropped entries by name. After cleaning, 83,895 records carry a usable year (97.8%), 85,211 carry a recorded mass (99.3%), and 52,096 carry decimal latitude and longitude (60.7%). Records lacking a mass field were retained for class-composition analyses but excluded from mass-distribution tests; records lacking coordinates were retained for time-series and mass analyses but excluded from geographic stratifications.\u003c/p\u003e\n\u003ch3\u003eClass taxonomy\u003c/h3\u003e\n\u003cp\u003eClassifying the 619 distinct \u003cem\u003eType\u003c/em\u003e strings in the live Bulletin into a small number of physically meaningful categories required deterministic substring rules. \u003cb\u003eIron meteorites\u003c/b\u003e were identified as records whose \u003cem\u003eType\u003c/em\u003e either began with the literal substring Iron (capturing the 12 chemical groups IAB through IVB and their structural sub-groups) or whose first whitespace-delimited token matched one of the canonical iron-group abbreviations (IAB, IIAB, IIIAB, IVA, IVB, IIICD, IIE, IIF, IIG, IIIE, IIIF, IC, IIC, IID). \u003cb\u003eLunar meteorites\u003c/b\u003e were identified by a case-insensitive match of Lunar. \u003cb\u003eMartian meteorites\u003c/b\u003e by case-insensitive matches of Martian, Shergottite, Nakhlite or Chassignite \u0026mdash; a slight broadening over the NASA snapshot taxonomy because the live Bulletin's \u003cem\u003eType\u003c/em\u003e strings encode Martian-grouping at the petrologic level rather than the explicit Martian token. \u003cb\u003eStony-irons\u003c/b\u003e (pallasites and mesosiderites) were identified by their own substrings; they form a small population (n\u0026thinsp;=\u0026thinsp;614). All other records were treated as \u003cem\u003enon-iron, non-lunar, non-martian\u003c/em\u003e for the binary contrasts.\u003c/p\u003e \u003cp\u003eThis taxonomy intentionally collapses the rich petrological structure of the chondrite classes into a single \"non-iron\" residual. We considered preserving finer chondrite subtypes (CM/CR/CV/CO carbonaceous, LL/L/H ordinary) but the \u003cem\u003eType\u003c/em\u003e strings encode subtype assignments inconsistently across decades \u0026mdash; pre-1980 entries are often classified only to top-level group (\"L\" or \"H\") with no subtype, while modern entries carry full petrologic-grade information. Aggregating the two would create spurious era differences. The coarse taxonomy avoids that confound at the cost of compressing within-class diversity.\u003c/p\u003e\n\u003ch3\u003eGeographic strata\u003c/h3\u003e\n\u003cp\u003eFour prospecting regions were defined by simple latitude/longitude rectangles, chosen to match the operational footprint of the major recovery programmes. \u003cb\u003eAntarctic\u003c/b\u003e is latitude\u0026thinsp;\u0026lt;\u0026thinsp;\u0026minus;\u0026thinsp;60\u0026deg; (the Transantarctic / blue-ice fields targeted by ANSMET, NIPR, PNRA, CHINARE and EUROMET). \u003cb\u003eArabia/Oman\u003c/b\u003e is 15\u0026deg; \u0026le; latitude\u0026thinsp;\u0026le;\u0026thinsp;35\u0026deg;, 35\u0026deg; \u0026le; longitude\u0026thinsp;\u0026le;\u0026thinsp;60\u0026deg; (the Rub' al Khali, Oman desert, and adjacent terrain prospected since the late 1990s). \u003cb\u003eSahara\u003c/b\u003e is 18\u0026deg; \u0026le; latitude\u0026thinsp;\u0026le;\u0026thinsp;35\u0026deg;, \u0026minus;\u0026thinsp;15\u0026deg; \u0026le; longitude\u0026thinsp;\u0026le;\u0026thinsp;35\u0026deg; (Algeria, Libya, Mali, Mauritania, Morocco, Niger, Western Sahara). \u003cb\u003eAustralia\u003c/b\u003e is \u0026minus;\u0026thinsp;40\u0026deg; \u0026le; latitude\u0026thinsp;\u0026le;\u0026thinsp;\u0026minus;\u0026thinsp;15\u0026deg;, 110\u0026deg; \u0026le; longitude\u0026thinsp;\u0026le;\u0026thinsp;155\u0026deg; (Nullarbor Plain and adjacent regions covered by the early- and mid-20th-century Australian collections). The seam between the Sahara and Arabia/Oman rectangles at longitude 35\u0026deg; is exclusive on the Sahara side and inclusive on the Arabia side, so a record can belong to at most one rectangle. Records outside all four rectangles are aggregated into a single \u003cem\u003eOther\u003c/em\u003e stratum dominated by Northwest Africa proper, the Atacama, the United States, and central Eurasia. The principal results are insensitive to \u0026plusmn;\u0026thinsp;5\u0026deg; perturbations of the boundaries; we report the perturbation experiment in \u0026sect;Robustness checks.\u003c/p\u003e\n\u003ch3\u003eEra partition\u003c/h3\u003e\n\u003cp\u003eThe catalogue was partitioned into three eras keyed to the two recovery transitions. \u003cb\u003ePre-1969\u003c/b\u003e is year\u0026thinsp;\u0026le;\u0026thinsp;1968, before any systematic Antarctic searches. \u003cb\u003eAntarctic era\u003c/b\u003e is 1969\u0026thinsp;\u0026le;\u0026thinsp;year\u0026thinsp;\u0026le;\u0026thinsp;1999, dominated by ANSMET and partner programmes. \u003cb\u003eDesert era\u003c/b\u003e is year\u0026thinsp;\u0026ge;\u0026thinsp;2000, in which hot-desert prospecting from Northwest Africa, the Arabian Peninsula and Oman dominates new accessions. The 1969 boundary is set by the December 1969 Yamato Mountains discovery by the Japanese Antarctic Research Expedition (Yoshida et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e1971\u003c/span\u003e), conventionally taken as the start of the systematic Antarctic-meteorite era. The 2000 boundary is more graded; we adopt year 2000 as a notational convenience and verify (\u0026sect;Robustness checks) that 1995, 2000 and 2005 boundaries give qualitatively identical era-composition profiles.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eStatistical methods\u003c/h2\u003e \u003cp\u003eBinary contrasts (e.g. iron-meteorite fraction pre- vs. post-1969, lunar fraction Antarctic vs. Arabia) used Pearson's χ\u0026sup2; on a 2 \u0026times; 2 contingency. Mass-distribution contrasts used the Mann\u0026ndash;Whitney \u003cem\u003eU\u003c/em\u003e test and the two-sample Kolmogorov\u0026ndash;Smirnov test on log-mass. Monotonic-correlation tests used Spearman ρ, with absolute latitude |φ| where the physical question was not hemisphere-resolved. Trend tests on the witnessed-fall time series used the rank-based Mann\u0026ndash;Kendall \u003cem\u003eZ\u003c/em\u003e statistic (Mann \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1945\u003c/span\u003e; Kendall \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1975\u003c/span\u003e), with a pre-specified Monte-Carlo power calculation (10⁴ Poisson realisations of a 16-decade series with the observed mean and dispersion) used to bracket the secular trends the test could and could not detect. At the sample sizes here, almost every two-group contrast is statistically significant by conventional thresholds; we therefore report effect sizes (Cliff's \u003cem\u003eδ\u003c/em\u003e; Cohen's \u003cem\u003eh\u003c/em\u003e) with bootstrap confidence intervals as the primary measure of contrast magnitude, and use \u003cem\u003ep\u003c/em\u003e-values only as an auxiliary screen.\u003c/p\u003e \u003cp\u003ePairing-aware fall-level estimates apply two distinct treatments. The first is a literature-derived pairing factor for each region and class, applied as a multiplicative correction to specimen counts. We use Korotev (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2012\u003c/span\u003e)'s empirically-derived hot-desert lunar pairing factor of 2.5\u0026times;, the Cassidy et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e1992\u003c/span\u003e) and Harvey (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2003\u003c/span\u003e) ANSMET-consensus Antarctic lunar pairing of \u0026asymp;\u0026thinsp;1.5\u0026times; and Antarctic bulk-chondrite pairing of \u0026asymp;\u0026thinsp;3\u0026times;, and the Hutzler et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) hot-desert bulk pairing of \u0026asymp;\u0026thinsp;1.75\u0026times;. The second is a deterministic naming-based clustering applied as a \u003cem\u003elower-bound stress test\u003c/em\u003e: specimens whose name prefix (the leading three- to four-letter ANSMET/JARE site code, or hot-desert locality token such as NWA, Dhofar, or Dar al Gani) and discovery year both match are merged into a single fall unit, with year-less specimens treated as singletons. This clustering deliberately over-pairs bulk chondrites (e.g. all ALH 80xxx specimens collapse into one cluster) and is reported as an extreme upper bound on multiplicity rather than as a corrected estimate; we explicitly flag it as a stress test, not as a third independent estimator parallel to the literature-pairing best estimate. To propagate pairing uncertainty into the principal effect sizes we run a Monte-Carlo (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2000) over uniform priors on each of the four pairing factors above (with width\u0026thinsp;\u0026plusmn;\u0026thinsp;50\u0026ndash;100% of the central value), recompute each headline contrast at fall level on every draw, and report the resulting 95% bootstrap intervals alongside the point estimates (\u0026sect;Pairing-uncertainty propagation).\u003c/p\u003e \u003cp\u003eAdjusted multivariate models are fit with statsmodels. Class-binary outcomes (is_iron, is_lunar) are modelled with logistic regression on era\u0026thinsp;+\u0026thinsp;region\u0026thinsp;+\u0026thinsp;fall-status; the era \u0026times; region interaction is included where the design admits it (the lunar model is fit with main effects only because pre-1969 era \u0026times; region cells contain near-zero lunar specimens, which causes the full-interaction model to be singular). Log₁₀ recorded mass is modelled with Huber-robust linear regression on the same predictors. Coefficients are reported as adjusted odds ratios (logistic) or dex shifts (robust regression) with 95% confidence intervals.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eReproducibility\u003c/h3\u003e\n\u003cp\u003eAll analyses were performed in Python using NumPy and SciPy. The analysis script (analysis_live.py), the figure-generation script (make_figures.py), and the cleaned, joined CSV (meteorites_clean_live.csv) are provided as supplementary material; intermediate results are serialized to JSON to avoid re-computation during figure generation, but the JSON is regenerated from scratch by analysis_live.py and is never edited by hand. No manual record-level editing or outlier removal was applied after the automated cleaning pipeline; every stratification reported here is a deterministic function of the raw MBDB CSV export of 4 May 2026.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eCatalogue composition\u003c/h2\u003e \u003cp\u003eThe 85,775-record catalogue contains 1,251 \u003cem\u003eFell\u003c/em\u003e records (witnessed falls) and 84,524 \u003cem\u003eFound\u003c/em\u003e records (post-fall recoveries) \u0026mdash; a 1 : 68 ratio. Year is present for 83,895 records (97.8%) and mass for 85,211 (99.3%); decimal latitude and longitude are present for 52,096 (60.7%). Of the geolocated records, 32,952 (63.3%) lie south of latitude\u0026thinsp;\u0026minus;\u0026thinsp;60\u0026deg; (the Antarctic stratum), 5,304 (10.2%) lie within the Arabia/Oman rectangle, 3,345 (6.4%) within the Sahara rectangle, and 697 (1.3%) within the Australia rectangle; the remaining 18.8% are distributed across the rest of the inhabited world (Fig.\u0026nbsp;3). Iron meteorites total 1,417 (1.65% of the full catalogue), lunar meteorites 820 (0.96%), Martian meteorites 419 (0.49%), and stony-irons 614 (0.72%).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eA factor-of-147 collapse in median recorded mass\u003c/h2\u003e \u003cp\u003eThe most prominent feature of the catalogue is the dramatic drop in characteristic specimen mass before and after 1969. Among the 2,200 pre-1969 records with a recorded mass, the median is \u003cb\u003e4,400 g\u003c/b\u003e (mean log₁₀ mass\u0026thinsp;=\u0026thinsp;2.55); among the 81,596 post-1969 records, the median is \u003cb\u003e30.0 g\u003c/b\u003e (mean log₁₀ mass\u0026thinsp;=\u0026thinsp;1.32). The pre/post ratio is \u003cb\u003e147\u003c/b\u003e (Fig.\u0026nbsp;2). The two distributions are nearly disjoint: Cliff's \u003cem\u003eδ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.81 [95% CI 0.80, 0.82] (derived from Mann\u0026ndash;Whitney \u003cem\u003eU\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.62 \u0026times; 10⁸ on n₁ = 2,200, n₂ = 81,596), placing the contrast in the \"very large\" range. Kolmogorov\u0026ndash;Smirnov \u003cem\u003eD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.670 on log-mass confirms that the shift extends to distributional shape, not only to location. The pre-1969 distribution is broad and centred near 10\u0026sup3; g, dominated by the historical witnessed-fall and historical-find populations of multi-kilogram stones; the post-1969 distribution is sharply peaked near 10\u0026sup1; g, dominated by the hand-sized fragments that are typical of Antarctic blue-ice and hot-desert prospecting (Fig.\u0026nbsp;2a).\u003c/p\u003e \u003cp\u003eA caveat on the absolute mass numbers is in order. The MBDB \u003cem\u003eMass\u003c/em\u003e field reports recorded mass as catalogued by the contributing agency. The Bulletin documentation explicitly states that recorded masses may be rounded, that they sometimes represent total known weight rather than individual stone mass, and that they should not be treated as authoritative physical measurements. For the catalogue-scale order-of-magnitude contrast we report, this rounding is irrelevant \u0026mdash; a 147-fold ratio cannot be moved into the \"no-effect\" range by any plausible rounding model. The ratio is, however, a recorded-mass ratio rather than a pre-atmospheric ratio.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eA factor-of-34 collapse in iron-meteorite fraction\u003c/h2\u003e \u003cp\u003eThe iron-meteorite fraction shows an even more dramatic collapse. Among the 2,244 pre-1969 records, \u003cb\u003e670 (29.86%)\u003c/b\u003e are irons; among the 81,651 post-1969 records, \u003cb\u003e715 (0.88%)\u003c/b\u003e are irons (Fig.\u0026nbsp;2b). The pre/post ratio is \u003cb\u003e34\u003c/b\u003e, and the contingency χ\u0026sup2; = 11,280 (1 df); Cohen's \u003cem\u003eh\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.97 [95% CI 0.93, 1.01] places this contrast in the \"large\" range. The simplest reading is that the historical record was selectively populated by easily-recognisable iron specimens picked up by farmers and travellers, while modern systematic searches recover the chondritic stones that dominate the in-falling population. Iron meteorites are believed to constitute 5\u0026ndash;6% of the in-falling flux (Bland et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e1996\u003c/span\u003e); the witnessed-fall (\u003cem\u003eFell\u003c/em\u003e) sub-sample returns an iron fraction of 49 / 1,251\u0026thinsp;=\u0026thinsp;3.9%, closer to the flux estimate than either era's \u003cem\u003eFound\u003c/em\u003e fraction. Within the \u003cem\u003eFell\u003c/em\u003e sub-sample alone, the pre/post-1969 iron fraction is 41/857\u0026thinsp;=\u0026thinsp;4.78% vs 8/394\u0026thinsp;=\u0026thinsp;2.03%, a roughly two-fold contraction even after recovery-mode is fixed. Whether this \u003cem\u003eFell\u003c/em\u003e-internal change reflects improving stony-fall recognition over the modern era (a recognition trend) or a small drift in the in-falling class composition is not separable from the data we have here. We treat it as a residual uncertainty on the \u003cem\u003eFell\u003c/em\u003e sub-sample's role as a flux reference, and accept that the \u003cem\u003eFell\u003c/em\u003e baseline is itself not perfectly stationary.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eWitnessed-fall rate: a small but significant positive trend since 1860\u003c/h2\u003e \u003cp\u003eIf the post-1969 surge in the catalogue reflected an actual increase in the in-falling meteoroid flux, we would see it in the \u003cem\u003eFell\u003c/em\u003e counts as well as the \u003cem\u003eFound\u003c/em\u003e counts. The \u003cem\u003eFell\u003c/em\u003e time series is more complicated than a simple null. Decadal \u003cem\u003eFell\u003c/em\u003e counts from the 1860s through the 2010s are 55, 51, 40, 51, 57, 66, 70, 92, 57, 60, 63, 61, 56, 59, 70, 84 (16 complete decades; mean \u003cb\u003e62\u0026thinsp;\u0026plusmn;\u0026thinsp;13\u003c/b\u003e per decade, coefficient of variation 20%; the partial 2020s value of 58 is excluded because the catalogue snapshot terminates in May 2026). The Mann\u0026ndash;Kendall test for monotonic trend on this series returns \u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;+\u0026thinsp;2.25, two-tailed \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.024: a small but statistically significant positive trend. The 1860s value (55) and the 2010s value (84) bound a roughly 50% rise across 150 years. A Monte-Carlo power analysis (10⁴ Poisson realisations of a 16-decade series with the observed mean and dispersion) gives the test\u0026thinsp;\u0026asymp;\u0026thinsp;94% power against a 50% cumulative trend (\u0026sim;3%/decade) and \u0026asymp;\u0026thinsp;31% power against a 20% cumulative trend (\u0026sim;1.3%/decade). The detected trend sits at the edge of what the test can resolve, and is roughly the magnitude that improving reporting infrastructure alone could produce.\u003c/p\u003e \u003cp\u003eThe simplest reading of this trend is that modern reporting infrastructure \u0026mdash; fireball camera networks, smartphone observations, social-media propagation of bolide reports \u0026mdash; has improved markedly since the 2000s, raising the apparent \u003cem\u003eFell\u003c/em\u003e count without any change in the underlying flux. A true flux change of comparable magnitude cannot be ruled out from the catalogue alone. The two interpretations sit on opposite sides of the same data point and our analysis cannot distinguish them. Whatever the interpretation, the \u003cem\u003eFell\u003c/em\u003e trend is at least two orders of magnitude smaller than the \u003cem\u003eFound\u003c/em\u003e series across the same window. The decadal \u003cem\u003eFound\u003c/em\u003e counts are 39, 40, 83, 75, 83, 88, 89, 220, 141, 155, 341, 5,673, 7,963, 14,441, 26,050, 20,599 across the 1860s through the 2010s (the 2020s value of 6,467 is partial). The pattern is unambiguous: a long pre-1960s baseline of 50 to 350 finds per decade, a step of order 30\u0026times; in the 1970s coincident with the start of ANSMET, and a further step of order 4\u0026times; in the 1990s and 2000s coincident with the rise of hot-desert prospecting. The catalogue surge is therefore overwhelmingly attributable to recovery and cataloguing effort rather than to any plausible flux change.\u003c/p\u003e \u003cp\u003eWe can quantify how much of the \u003cem\u003eFell\u003c/em\u003e trend a reporting-only model would absorb. Suppose the observed \u003cem\u003eFell\u003c/em\u003e count is \u003cem\u003etrue_flux\u003c/em\u003e \u0026times; \u003cem\u003er(t)\u003c/em\u003e, with \u003cem\u003er(t)\u003c/em\u003e a linear ramp from 1 in 1865 to 1\u0026thinsp;+\u0026thinsp;δ in 2015. Deflating the observed series by \u003cem\u003er(t)\u003c/em\u003e and re-running the Mann\u0026ndash;Kendall test gives Z\u0026thinsp;=\u0026thinsp;+\u0026thinsp;1.76 (p\u0026thinsp;=\u0026thinsp;0.079) at δ\u0026thinsp;=\u0026thinsp;10%, Z\u0026thinsp;=\u0026thinsp;+\u0026thinsp;0.77 (p\u0026thinsp;=\u0026thinsp;0.44) at δ\u0026thinsp;=\u0026thinsp;20%, and Z\u0026thinsp;=\u0026thinsp;+\u0026thinsp;0.27 (p\u0026thinsp;=\u0026thinsp;0.79) at δ\u0026thinsp;=\u0026thinsp;30%. A reporting-improvement multiplier of about 10% across 150 years is therefore enough to render the apparent \u003cem\u003eFell\u003c/em\u003e trend statistically non-significant; a multiplier of 20% eliminates it. The infrastructure changes documented in the literature plausibly produce a multiplier in this range, so we treat the \u003cem\u003eFell\u003c/em\u003e sub-sample as the \u003cem\u003eleast-biased\u003c/em\u003e available reference rather than as an unbiased one, and we carry the residual uncertainty forward in \u0026sect;Adjusted models.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eGeographic recovery hotspots\u003c/h2\u003e \u003cp\u003eThe geolocated catalogue is dominated by four prospecting regions (Fig.\u0026nbsp;3, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e). \u003cb\u003eAntarctic\u003c/b\u003e (latitude\u0026thinsp;\u0026lt;\u0026thinsp;\u0026minus;\u0026thinsp;60\u0026deg;): n\u0026thinsp;=\u0026thinsp;32,952, \u003cb\u003e63.3%\u003c/b\u003e of geolocated records, median recorded mass \u003cb\u003e9.8 g\u003c/b\u003e, iron fraction 0.44%, lunar fraction 0.124%, Martian fraction 0.049%. \u003cb\u003eArabia/Oman\u003c/b\u003e (the Rub' al Khali and surrounds): n\u0026thinsp;=\u0026thinsp;5,304, 10.2%, median 162.0 g, iron 0.23%, lunar 1.45%, Martian 0.38%. \u003cb\u003eSahara\u003c/b\u003e (Northwest Africa, including Morocco): n\u0026thinsp;=\u0026thinsp;3,345, 6.4%, median 305.6 g, iron 1.20%, lunar \u003cb\u003e3.65%\u003c/b\u003e, Martian 1.14%. \u003cb\u003eAustralia\u003c/b\u003e (Nullarbor and surrounds): n\u0026thinsp;=\u0026thinsp;697, 1.3%, median 126.0 g, iron \u003cb\u003e10.91%\u003c/b\u003e, lunar 0.29%, Martian 0.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eGeographic-stratum composition (geolocated records only)\u003c/b\u003e \u003cem\u003eGeolocated set n\u0026thinsp;=\u0026thinsp;52,096; percentages are within the geolocated total. Mass medians exclude records without a recorded mass (within each region).\u003c/em\u003e\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRegion\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e% of geolocated\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMedian mass (g)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e% Iron\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e% Lunar\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e% Martian\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAntarctic\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e32,952\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e63.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e9.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eArabia/Oman\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5,304\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e162.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e1.45\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.38\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSahara\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3,345\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e305.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e3.65\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAustralia\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e697\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e126.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e10.91\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9,798\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e18.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e151.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eSeveral patterns stand out. First, the regional median masses span more than a factor of 30: Antarctic specimens are an order of magnitude smaller than hot-desert specimens, which in turn are about half the size of Australian and rest-of-world specimens. This is a direct fingerprint of the prospecting mode \u0026mdash; Antarctic blue-ice fields concentrate small stones by ablation, hot-desert pavements expose larger stones over millennial preservation timescales, and the Australian Nullarbor collection was historically dominated by the multi-kilogram irons of the early-20th-century iron-collecting era. Second, the iron fraction varies by a factor of ~\u0026thinsp;47 across the regions, from 0.23% in Arabia/Oman to 10.91% in Australia (Fig.\u0026nbsp;4a) \u0026mdash; overwhelmingly a regional-history artefact rather than a regional flux signal. Third, the \u003cb\u003especimen-level lunar fraction is 11.7 times higher in Arabia/Oman than in the Antarctic\u003c/b\u003e across the full catalogue (1.45% vs. 0.124%; χ\u0026sup2; = 257.5, Cohen's \u003cem\u003eh\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.17 [0.14, 0.19]; Fig.\u0026nbsp;4b). This headline number is, however, sensitive to two confounds we address in turn: differential era sampling, and differential pairing. The Sahara stratum's even higher lunar fraction (3.65%) reinforces the picture that hot-desert pavements have, at the catalogue level, returned a higher recovered-fraction of lunar specimens than blue-ice fields have.\u003c/p\u003e \u003cp\u003e \u003cem\u003eEra confound.\u003c/em\u003e Antarctic recoveries are concentrated in 1969\u0026ndash;1999 (most of the Antarctic stratum is from the ANSMET era) and Arabia/Oman recoveries in 2000+ (most of the Arabia/Oman stratum is from the desert era), so the unstratified comparison conflates regional environment with recovery-era effects. Restricting both regions to a common temporal window: within the 1969\u0026ndash;1999 era the Antarctic-vs-Arabia lunar-fraction ratio is \u003cb\u003e6.75\u0026times;\u003c/b\u003e (Antarctic 0.091%, Arabia 0.617%), and within the 2000\u0026thinsp;+\u0026thinsp;era it is \u003cb\u003e8.86\u0026times;\u003c/b\u003e (Antarctic 0.167%, Arabia 1.479%).\u003c/p\u003e \u003cp\u003e \u003cem\u003ePairing confound.\u003c/em\u003e The catalogue's specimen-level counts treat paired fragments of single falls as independent events. Antarctic blue-ice fields and hot-desert pavements pair at different rates and for different classes, so a pairing-na\u0026iuml;ve ratio over-states the environmental component. Applying literature-derived pairing factors \u0026mdash; Korotev (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2012\u003c/span\u003e)'s hot-desert lunar pairing of \u0026asymp;\u0026thinsp;2.5\u0026times;, Cassidy et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e1992\u003c/span\u003e) and Harvey (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2003\u003c/span\u003e) ANSMET-consensus Antarctic lunar pairing of \u0026asymp;\u0026thinsp;1.5\u0026times;, the same authors' Antarctic bulk-chondrite pairing of \u0026asymp;\u0026thinsp;3\u0026times;, and Hutzler et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) hot-desert bulk pairing of \u0026asymp;\u0026thinsp;1.75\u0026times; \u0026mdash; gives fall-level lunar fractions of about 0.16% (Antarctic) and 0.97% (Arabia/Oman), a ratio of \u0026asymp;\u0026thinsp;6\u0026times;. We apply the lunar-specific pairing factor to lunar specimens explicitly and \u003cem\u003edo not\u003c/em\u003e substitute the bulk-chondrite factor for them, because lunar specimens pair at lower rates than bulk chondrites in Antarctic strewn fields by a factor of about two; using the 3\u0026times; bulk factor would over-correct lunar pairing in the Antarctic stratum and is not justified by the literature. An aggressive naming-based clustering (treating every specimen sharing a name prefix and discovery year as a single fall; rule given in \u003cem\u003eMethods\u003c/em\u003e) collapses the contrast to \u0026asymp;\u0026thinsp;1\u0026times;, but this is reported only as a stress-test lower bound on multiplicity rather than as a third independent estimator. The literature-pairing best estimate of \u0026asymp;\u0026thinsp;6\u0026times; and the era-restricted 1969\u0026ndash;1999 specimen-level contrast (also 6.75\u0026times;) sit at the same order of magnitude; their numerical agreement may be partly coincidental (both estimators respond to the temporal concentration of recoveries in different post-1969 sub-eras, so they are not fully independent estimators). We conclude that the environmental component of the lunar contrast is plausibly \u003cb\u003e5\u0026ndash;10\u0026times;\u003c/b\u003e under realistic pairing assumptions \u0026mdash; substantial, but materially smaller than the 11.7\u0026times; headline number.\u003c/p\u003e \u003cp\u003eThe Sahara stratum is intermediate at the specimen level for Arabia/Oman comparisons but is the highest individual-region lunar yield in the catalogue (3.65%, much higher than the previously-reported NASA-snapshot value because the live MBDB has captured the post-2010 surge of Northwest African lunar recoveries). The Australia stratum sits at 0.29%, reflecting the older Australian collecting era predating systematic lunar-meteorite recognition (Korotev \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). Plausible mechanisms for the residual environmental contrast include differential pairing rates (Antarctic specimens are more likely to be paired fragments of a single fall), differential recognition (the fusion crust of lunar feldspathic specimens has higher visual contrast against desert pavement than against ice), and differential dwell-times before recovery (hot-desert pavements integrate the in-falling flux over 10⁴\u0026ndash;10⁵ y, whereas Antarctic blue ice integrates over 10⁵\u0026ndash;10⁶ y, so a transient enrichment of lunar specimens following a particular lunar-impact ejection event would show up more strongly in the shorter-dwell-time hot-desert sample). The latter mechanism is consistent with the absence of comparable enrichment in iron meteorites.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eLatitude vs. mass: a dataset-induced spurious physical inference\u003c/h2\u003e \u003cp\u003eThe full catalogue exhibits a strongly negative Spearman correlation between absolute latitude and log-recorded-mass: ρ = \u003cb\u003e\u0026minus;0.393\u003c/b\u003e, \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;52,022; the rank correlation accounts for ρ\u0026sup2; \u0026asymp; 15% of the mass variance. Naively interpreted, this would imply that the in-falling meteoroid mass distribution is latitude-dependent, with smaller stones falling at high latitudes \u0026mdash; a conclusion with no obvious physical mechanism, since meteoroids are decoupled from any latitude-resolved process at the relevant size scales (Bland and Artemieva \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Halliday et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1989\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe result reverses sign under stratification (Fig.\u0026nbsp;5). *\u003cem\u003eRestricted to witnessed falls (\u003c/em\u003eFell\u003cem\u003e): ρ = +0.031\u003c/em\u003e, n\u0026thinsp;\u003cem\u003e=\u0026thinsp;1,213\u003c/em\u003e, p\u0026thinsp;\u003cem\u003e=\u0026thinsp;0.273. The correlation is small and not statistically significant. Within the\u003c/em\u003e Fell \u003cem\u003ereference population, there is no meaningful latitude-mass relationship. Excluding the Antarctic stratum: ρ = +0.142\u003c/em\u003e, n\u0026thinsp;\u003cem\u003e=\u0026thinsp;19,078. Among non-Antarctic recoveries, |φ| is\u003c/em\u003e positively \u003cem\u003ecorrelated with mass \u0026mdash; large stones come from mid-latitudes, small stones from low latitudes. This is the population-density signal, modulated by 19th- and 20th-century iron-collection biases. Within the Antarctic stratum alone: ρ = +0.144 (in |φ| terms), reflecting the spatial structure of the blue-ice prospecting fields against the Transantarctic Mountains. Full catalogue (combining the above)\u003c/em\u003e*: the negative ρ = \u0026minus;0.393 emerges because the Antarctic stratum, with its distinctive |φ| \u0026ge; 60\u0026deg; and median mass 9.8 g, dominates the high-|φ| portion of the joint distribution and pulls the global rank correlation negative.\u003c/p\u003e \u003cp\u003eThe headline negative correlation is therefore an instance of \u003cem\u003edataset-induced spurious physical inference\u003c/em\u003e: a sub-population that dominates one corner of a covariate space generates a pooled correlation whose sign is the opposite of the within-stratum correlation, and whose physical interpretation has no mechanistic basis. The pattern is the meteoritic-database analogue of Simpson's paradox in epidemiology and of the ecological-fallacy correlations long known in geographic statistics. The diagnostic procedure is the same as in those fields. Whenever a community catalogue is dominated, in any covariate corner, by a single recovery campaign, any pooled correlation involving that covariate must be re-checked against the least-biased reference sub-sample. In our case, the Antarctic-led ρ = \u0026minus;0.39 collapses to ρ\u0026thinsp;\u0026asymp;\u0026thinsp;0 within the \u003cem\u003eFell\u003c/em\u003e sub-sample, and the negative correlation should be removed from the list of properties of the in-falling meteoroid flux. We expect that other community datasets \u0026mdash; paleontological collections, exoplanet catalogues, sky surveys with different completeness limits \u0026mdash; show analogous spurious correlations driven by single-collection dominance, and we suggest that explicit \u003cem\u003eFell\u003c/em\u003e-style stratified checks should be standard practice rather than an after-the-fact artefact diagnosis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eThree statistically distinct eras\u003c/h2\u003e \u003cp\u003eThe combination of the median-mass collapse, the iron-fraction collapse, and the geographic shift partitions the catalogue into statistically distinct eras (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Fig.\u0026nbsp;6). \u003cb\u003ePre-1969\u003c/b\u003e (n\u0026thinsp;=\u0026thinsp;2,244): median mass 4,400 g, iron 29.86%, lunar 0.045%, Martian 0.267%. \u003cb\u003eAntarctic era 1969\u0026ndash;1999\u003c/b\u003e (n\u0026thinsp;=\u0026thinsp;28,323): median mass 15.2 g, iron 0.91%, lunar 0.092%, Martian 0.071%, Antarctic share dominant. \u003cb\u003eDesert era 2000+\u003c/b\u003e (n\u0026thinsp;=\u0026thinsp;53,328): median mass 45.0 g, iron 0.86%, lunar \u003cb\u003e1.483%\u003c/b\u003e, Martian \u003cb\u003e0.735%\u003c/b\u003e, Arabia/Sahara share dominant.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eEra-stratified composition of the meteorite catalogue\u003c/b\u003e \u003cem\u003eAll percentages and counts are based on records with Status \u0026isin; {Official, Provisional} and a usable year (n\u0026thinsp;=\u0026thinsp;83,895 in total).\u003c/em\u003e\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEra\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedian mass (g)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e% Iron\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e% Lunar\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e% Martian\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePre-1969\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2,244\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4,400.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e29.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.045\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.267\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAntarctic era (1969\u0026ndash;1999)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e28,323\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.092\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.071\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDesert era (2000+)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e53,328\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e45.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e1.483\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.735\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTwo observations are particularly notable. The Antarctic share of the catalogue \u003cem\u003epeaked\u003c/em\u003e in the Antarctic era and has fallen in the desert era \u0026mdash; not because Antarctic recoveries have stopped, but because the absolute pace of hot-desert prospecting has overtaken them. The lunar fraction increased by a factor of 16 between the Antarctic era and the desert era (0.092% \u0026rarr; 1.483%), and the Martian fraction by a factor of 10 (0.071% \u0026rarr; 0.735%), almost entirely because the Sahara, Arabia/Oman and Other (mostly Northwest African) strata over-yield these planetary classes relative to Antarctica. Estimates of Solar-System provenance abundances drawn from the catalogue therefore require weighting by recovery region as well as by recovery era.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003ePairing-uncertainty propagation\u003c/h2\u003e \u003cp\u003eThe pairing factors used so far are point estimates drawn from the literature. To propagate their uncertainty into the principal effect sizes, we ran a Monte-Carlo (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2000) over independent uniform priors on each of four pairing factors \u0026mdash; Antarctic lunar (1.0\u0026ndash;2.0\u0026times;), hot-desert lunar (1.5\u0026ndash;4.0\u0026times;), Antarctic bulk (2.0\u0026ndash;4.0\u0026times;), hot-desert bulk (1.5\u0026ndash;2.5\u0026times;) \u0026mdash; and recomputed each headline contrast at fall level on every draw. The results are summarised below. The Arabia/Oman-versus-Antarctic lunar-fraction contrast at fall level has a median of \u003cb\u003e4.31\u0026times;\u003c/b\u003e with a 95% bootstrap interval of [2.04, 9.73]\u0026times;. The interval comfortably excludes parity, so the environmental component of the lunar contrast is bounded above zero even under aggressive pairing assumptions; it is also bounded below the 11.7\u0026times; specimen-level figure under any plausible pairing assumption. The pre-versus-post-1969 mass collapse at fall level has a median of \u003cb\u003e77\u0026times;\u003c/b\u003e with a 95% interval of [63, 96]\u0026times; \u0026mdash; the 147\u0026times; specimen-level figure shrinks substantially when post-1969 stones are aggregated into falls, but the contrast remains an order of magnitude even at the upper-pairing end. The pre-versus-post-1969 iron-fraction collapse is \u003cb\u003e34\u0026times;\u003c/b\u003e with a tight 95% interval; pre-1969 records are dominated by historical non-systematic finds with effective pairing close to 1\u0026times;, so the iron contrast is essentially insensitive to post-1969 pairing factors. We therefore report the principal contrasts as ranges rather than point estimates wherever pairing matters.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eAdjusted models: era \u0026times; region \u0026times; fall-status\u003c/h2\u003e \u003cp\u003eThe stratified contrasts of \u0026sect;\u0026nbsp;3.1\u0026ndash;\u0026sect;\u0026nbsp;3.6 expose every pairwise effect but do not adjust each contrast for the others simultaneously. We fit three adjusted models on the \u003cem\u003en\u003c/em\u003e\u0026thinsp;\u0026asymp;\u0026thinsp;84 k records that carry year, mass, and a region assignment. The first is a logistic regression for \u003cem\u003eP\u003c/em\u003e(iron) on era\u0026thinsp;+\u0026thinsp;region\u0026thinsp;+\u0026thinsp;fall-status with a full era \u0026times; region interaction. The second is a logistic regression for \u003cem\u003eP\u003c/em\u003e(lunar) with main effects only (the full interaction is singular because lunar-cell counts in pre-1969 era \u0026times; region intersections are zero or one). The third is a Huber-robust linear regression on log₁₀ recorded mass against the same predictors with the era \u0026times; region interaction. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e reports the headline coefficients. The reference cell is \u003cem\u003ePre-1969 \u0026times; Other \u0026times; Found\u003c/em\u003e.\u003c/p\u003e \u003cp\u003eThree findings stand out. (i) \u003cb\u003eEra is the dominant adjusted predictor of iron fraction.\u003c/b\u003e Holding region and fall-status fixed, post-1969 records have \u0026asymp;\u0026thinsp;50\u0026times; lower iron odds than pre-1969 records (Antarctic-era OR\u0026thinsp;=\u0026thinsp;0.18, p\u0026thinsp;\u0026asymp;\u0026thinsp;10⁻⁴⁹; Desert-era OR\u0026thinsp;=\u0026thinsp;0.024, p\u0026thinsp;\u0026asymp;\u0026thinsp;10⁻\u0026sup2;⁷\u0026sup1;). The era \u0026times; region interactions further depress Antarctic-era Antarctic, Sahara, and Australia cells, but the era main effect alone delivers most of the contrast. (ii) \u003cb\u003eRegion drives the lunar fraction, era supplies an additional desert-era boost.\u003c/b\u003e Holding era and fall-status fixed, the Sahara stratum has 142\u0026times; higher lunar odds than the Other reference (p\u0026thinsp;\u0026asymp;\u0026thinsp;10⁻\u0026sup2;\u0026sup2;), Arabia/Oman has 26\u0026times;, NoLoc has 39\u0026times;, Antarctic has 4.5\u0026times;, Australia has 21\u0026times; (with wider CIs because Australia counts are small). The Desert-era main effect adds roughly 5\u0026times; over Antarctic-era (p\u0026thinsp;=\u0026thinsp;0.13) but the bulk of the lunar contrast is regional. (iii) \u003cb\u003eMass is set by era, with a smaller region modifier.\u003c/b\u003e The robust regression returns era main-effect coefficients of \u0026minus;\u0026thinsp;0.92\u0026thinsp;\u0026plusmn;\u0026thinsp;0.04 dex (Antarctic era) and \u0026minus;\u0026thinsp;1.68\u0026thinsp;\u0026plusmn;\u0026thinsp;0.03 dex (Desert era) \u0026mdash; i.e. a 7 to 50-fold drop in adjusted mean log-mass, depending on era. The Antarctic-era \u0026times; Antarctic-region interaction is a further\u0026thinsp;\u0026minus;\u0026thinsp;1.54 dex, consistent with the blue-ice fragment population. The \u003cem\u003eFell\u003c/em\u003e main effect is +\u0026thinsp;0.27 dex (a slight upward bias of witnessed-fall mass over found mass at the same era \u0026times; region), consistent with witnessed falls preserving larger stones. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e reports the full coefficient set.\u003c/p\u003e \u003cp\u003eThe adjusted-model framework supplies the answer to the most direct objection an external reviewer can raise: that the stratified contrasts in \u0026sect;\u0026nbsp;3.1\u0026ndash;\u0026sect;\u0026nbsp;3.6 are confounded with each other. They are, but the adjusted-OR and adjusted-RLM coefficients show that the headline effects survive simultaneous control. The era effect on iron and on mass is order-of-magnitude even after adjusting for region and fall-status; the region effect on lunar is order-of-magnitude even after adjusting for era and fall-status. The headline contrasts are therefore not a single-confounder artefact, and the recovery-bias structure has independent era and region components.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAdjusted-model summary. Columns are odds-ratios (logistic models for iron and lunar) and dex coefficients (robust linear regression for log-mass) with 95% CIs. \u003cem\u003ep\u003c/em\u003e-values are reported only where \u0026ge;\u0026thinsp;10⁻\u0026sup3;; smaller values are simply marked as ≪ 10⁻\u0026sup3;.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredictor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIron OR [CI]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLunar OR [CI]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003elog-mass β (dex) [CI]\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAntarctic era (vs Pre-1969)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.18 [0.15, 0.23]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.37 [0.05, 3.00]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026minus;0.92 [\u0026minus;\u0026thinsp;1.00, \u0026minus;\u0026thinsp;0.85]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDesert era (vs Pre-1969)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.024 [0.02, 0.03]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.99 [0.64, 39.07]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026minus;1.68 [\u0026minus;\u0026thinsp;1.73, \u0026minus;\u0026thinsp;1.62]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAntarctic region (vs Other)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.16 [0.16, 8.29]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.48 [1.60, 12.55]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.10 [\u0026minus;\u0026thinsp;0.85, +\u0026thinsp;1.06]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eArabia/Oman region (vs Other)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.34 [0.14, 0.84]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.1 [9.5, 71.5]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026minus;0.70 [\u0026minus;\u0026thinsp;1.03, \u0026minus;\u0026thinsp;0.36]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSahara region (vs Other)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.83 [0.32, 2.18]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e142 [52, 387]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.11 [\u0026minus;\u0026thinsp;0.29, +\u0026thinsp;0.51]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAustralia region (vs Other)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.68 [0.49, 0.95]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.6 [3.6, 117]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.22 [+\u0026thinsp;0.07, +\u0026thinsp;0.37]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFell (vs Found)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.07 [0.05, 0.10]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eunstable (rare event)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.27 [+\u0026thinsp;0.21, +\u0026thinsp;0.34]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAntarctic era \u0026times; Antarctic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.024 [0.001, 0.18]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026minus;1.54 [\u0026minus;\u0026thinsp;2.50, \u0026minus;\u0026thinsp;0.59]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDesert era \u0026times; Sahara\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.93 [0.32, 2.70]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.64 [+\u0026thinsp;0.24, +\u0026thinsp;1.04]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDesert era \u0026times; Arabia/Oman\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.12 [0.03, 0.46]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;1.01 [+\u0026thinsp;0.68, +\u0026thinsp;1.35]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eRobustness checks\u003c/h2\u003e \u003cp\u003eWe verified that the qualitative results survive several reasonable perturbations of the analysis. First, restricting to \u003cem\u003eOfficial\u003c/em\u003e records (excluding the 6,180 \u003cem\u003eProvisional\u003c/em\u003e records) changes the iron-fraction percentages by less than 0.05 percentage points and the median masses by less than 1%. Second, moving the Antarctic-era boundary to 1965, 1970, or 1975 changes the era-summary percentages by less than 0.5 percentage points; the same is true for moving the desert-era boundary across 1995, 2000, 2005. Third, redefining the Arabia/Oman rectangle to exclude the Persian Gulf coast (latitude\u0026thinsp;\u0026ge;\u0026thinsp;22\u0026deg; rather than \u0026ge;\u0026thinsp;15\u0026deg;) reduces the Arabia/Oman count from 5,304 to about 4,800 but raises the lunar fraction from 1.45% to roughly 1.65%, sharpening rather than weakening the contrast with Antarctica. Fourth, the Spearman ρ between |φ| and log-mass is stable under bin-aggregating mass at 0.25-dex resolution and under dropping the smallest 1% and largest 1% of masses.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eWhat the catalogue *can* and *cannot* be used for\u003c/h2\u003e \u003cp\u003eThe \u003cem\u003eFell\u003c/em\u003e sub-sample is the appropriate baseline for flux-based questions. With n\u0026thinsp;=\u0026thinsp;1,251 and a near-stationary\u0026thinsp;\u0026asymp;\u0026thinsp;62-per-decade rate over the past 150 years (with a small positive trend documented in \u0026sect;Witnessed-fall rate), it is the least-biased reference population available for the in-falling meteoroid flux at the kilogram-mass scale, suitable for flux-based, class-frequency, or population-mean inference. Its biases \u0026mdash; population-density modulation of observational opportunity, geographic variation in reporting culture, under-recovery of small specimens, and the smaller-magnitude recognition trend that produces the documented Mann\u0026ndash;Kendall significance \u0026mdash; are smaller than the recovery-environment biases that dominate the \u003cem\u003eFound\u003c/em\u003e sub-sample. The numbers line up. Its iron fraction (3.9%) falls within the 4\u0026ndash;7% range estimated by independent flux models (Bland et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e1996\u003c/span\u003e; Bland and Artemieva \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Halliday et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1989\u003c/span\u003e); its median mass (2.6 kg) and class distribution are likewise consistent with those models within the resolution of the available statistics.\u003c/p\u003e \u003cp\u003eThe full catalogue, by contrast, is dominated by recovery effort, recovery environment, and recovery-era effects. None of the following are reliably extractable from the full catalogue without explicit recovery-bias correction: the \u003cem\u003eabsolute\u003c/em\u003e flux of meteoroids reaching Earth's surface as a function of class; the \u003cem\u003elatitudinal\u003c/em\u003e dependence of any property of the in-falling flux; the \u003cem\u003etemporal\u003c/em\u003e variation of the in-falling flux (which the \u003cem\u003eFell\u003c/em\u003e time series places strong upper limits on, contradicting any claim of order-of-magnitude flux variability over the past 200 years); and the \u003cem\u003erelative abundance\u003c/em\u003e of rare classes (lunar, Martian, ungrouped achondrites) without weighting by recovery region. The full catalogue is, however, well-suited for searches for petrologically novel material (where coverage matters more than statistical representativeness), for pairing studies and strewn-field reconstructions within recovery regions, and for studies of recovery itself \u0026mdash; i.e. of how prospecting environment, climate, and collector behaviour interact to shape the apparent record (the present paper being one such study).\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eImplications for asteroid-source inversion\u003c/h2\u003e \u003cp\u003eA common use of the catalogue is to invert the relative abundance of meteorite classes against the relative abundance of asteroid spectral types in order to constrain meteorite-asteroid genetic links (Burbine \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Bottke et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). Such inversions tend to assume \u0026mdash; implicitly more often than explicitly \u0026mdash; that the catalogue's class composition reflects the in-falling flux. The factor-of-34 collapse in iron-meteorite fraction across the 1969 boundary, and the order-of-magnitude variation in iron fraction across recovery regions, suggest that this assumption is violated by orders of magnitude in the pre-1969 sub-sample and by factors of several even within post-1969 hot-desert collections. Inversions that pool all eras and regions therefore inherit the bias.\u003c/p\u003e \u003cp\u003eThe remedy is direct. Perform the inversion against the \u003cem\u003eFell\u003c/em\u003e sub-sample only, accepting the smaller sample size in exchange for a less-biased estimator. A stronger approach combines the \u003cem\u003eFell\u003c/em\u003e sub-sample for absolute calibration with the \u003cem\u003eFound\u003c/em\u003e sub-sample for class-internal substructure, weighting by recovery region and era to remove the gross biases we have documented. We have not attempted this multivariate weighting. The present scope is restricted to quantifying the magnitude of the bias and supplying the statistical infrastructure for follow-up work.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eImplications for lunar and Martian sample yields\u003c/h2\u003e \u003cp\u003eLunar and Martian meteorites are scientifically valuable beyond their statistical weight because they provide a planetary surface sample at near-zero mission cost. The 11.7\u0026times; contrast in lunar fraction between Arabia/Oman and the Antarctic, together with the large absolute counts in the desert era (about 70 lunar specimens recovered in Arabia/Oman post-2000 and a comparable number from the Sahara stratum vs. about 25 from the Antarctic), indicates that hot-desert collections have historically contributed a higher catalogued fraction of lunar specimens per recovered stone than Antarctic collections have. The Martian fractions show a similar but slightly weaker contrast (Arabia/Oman 0.38% vs. Antarctic 0.05%; Sahara 1.14% vs. Antarctic 0.05%). The mechanistic reasons are several \u0026mdash; better visual contrast against desert pavement, shorter terrestrial dwell times that preserve the chemistry of a transient lunar-impact event, lower pairing rates per recovered specimen \u0026mdash; and the empirical catalogue-level contrast is large. We do not have a search-effort denominator (km\u0026sup2; covered, person-hours, total mass collected) that would be needed to convert the catalogue contrast into a per-effort yield, so we phrase the contrast as a recovered-fraction contrast rather than as a yield-per-effort recommendation; that conversion is left to operations-level studies that have access to the relevant denominator.\u003c/p\u003e \u003cp\u003eThat said, the Antarctic record provides a unique long-dwell-time integration of the planetary-meteorite flux that hot deserts cannot match. For studies that depend on the longest-possible delivery-time integral (e.g. searches for very-old planetary surface material, or for material from a particular impact event millions of years past) the Antarctic strata remain irreplaceable.\u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003eThe pre-1969 record as a window into pre-modern observation\u003c/h2\u003e \u003cp\u003eThe pre-1969 record contains 2,244 specimens accumulated over ~\u0026thinsp;250 years. Its 30% iron fraction and 4.4-kg median mass tell us less about the in-falling flux than they tell us about \u003cem\u003ewhat humans noticed and bothered to keep\u003c/em\u003e in the pre-systematic era: visually distinctive metallic stones large enough to be picked up by hand and small enough to be carried by a single farmer. Iron meteorites, in particular, were prized as a free source of nickel-iron ore and as curiosities throughout the Bronze, Iron, and Industrial ages (King \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e1865\u003c/span\u003e; Buchwald \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e1975\u003c/span\u003e), and the historical record is enormously skewed toward them.\u003c/p\u003e \u003cp\u003eThe witnessed-fall sub-sample of the pre-1969 era, however, has an iron fraction of 4.78% (41 of 857), reduced further to 2.03% (8 of 394) in the post-1969 witnessed-fall sub-sample \u0026mdash; both far below the 30% pre-1969 \u003cem\u003eFound\u003c/em\u003e figure, confirming that the historical 30% iron fraction is a feature of the pre-1969 \u003cem\u003eFound\u003c/em\u003e sub-population only and not of the underlying in-falling flux. This is itself an empirical lesson: even within a single era, the recovery channel (witnessed vs. found) imposes a class-composition bias that swamps any era-to-era physical signal.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003ch2\u003eA note on historical chronologies\u003c/h2\u003e \u003cp\u003eThe catalogue extends back to 860 CE (the Nōgata fall, Japan) and includes a small handful of mediaeval and early-modern witnessed falls. These records are valuable as cultural-historical evidence but are statistically negligible: there are 14 records dated before 1700 in the entire catalogue. The pre-1700 sub-sample is dominated by the few falls of sufficient magnitude or social impact to be recorded in surviving documents, and any quantitative claim about pre-1700 rates is dominated by Poisson noise on a one-digit count base. We have included the pre-1700 records in our time series for completeness but draw no inference from their decadal counts.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eSeveral limitations should be borne in mind. The catalogue continues to grow; the precise era-end percentages will shift as further 2020s recoveries accumulate, although the qualitative pattern (Antarctic-era \u0026rarr; desert-era transition centred near 2000) is sufficiently well-established that further accessions will reinforce rather than reverse the structure. Our class taxonomy collapses 619 \u003cem\u003eType\u003c/em\u003e strings into a small number of bins; users interested in the substructure within iron meteorites (e.g. IAB-MG vs. IIIAB) or within carbonaceous chondrites (e.g. CM vs. CR vs. CV) should re-derive their own taxonomy from meteorites_clean_live.csv. The four geographic strata are coarse rectangles; users interested in finer regional differentiation (e.g. ANSMET vs. NIPR vs. EUROMET, or Acfer vs. Dar al Gani vs. Hammadah al Hamra) will need the original Bulletin metadata, which carries finer locality fields than we have used here.\u003c/p\u003e \u003cp\u003eThe most consequential limitation concerns pairing. The analysis treats individual catalogue entries as independent specimens, but fragmented falls \u0026mdash; particularly in Antarctic blue-ice fields, where one bolide can deposit many small stones across a strewn field \u0026mdash; violate this assumption, and pairing factors differ across regions and across classes. We did not have access to the Bulletin's authoritative pairing decisions, so we report a sensitivity range rather than a single corrected number. A literature-derived pairing-factor correction, applied per region and per class, gives an environmental component of the Antarctic-vs-Arabia/Oman lunar contrast of approximately 6\u0026times;, against an unstratified 11.7\u0026times; upper bound. An aggressive naming-based clustering collapses the contrast to about 1\u0026times; as a stress-test lower bound; this is not a believable estimate but bounds the family of plausible pairings. The literature-pairing best estimate of \u0026asymp;\u0026thinsp;6\u0026times; and the era-restricted 1969\u0026ndash;1999 specimen-level contrast (6.75\u0026times;) sit at the same order of magnitude; their numerical agreement should not be over-interpreted, since both estimators respond to the temporal concentration of recoveries in different post-1969 sub-eras, so they are correlated estimators rather than independent ones. We therefore conclude that the environmental component of the Antarctic-vs-Arabia/Oman lunar contrast is plausibly \u003cb\u003e5\u0026ndash;10\u0026times;\u003c/b\u003e under realistic pairing assumptions \u0026mdash; substantial, but materially smaller than the 11.7\u0026times; headline number, and bracketed rather than pinned down. The mass and iron-fraction collapses, by contrast, are large enough (Cliff's \u003cem\u003eδ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.81; Cohen's \u003cem\u003eh\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.97) that no plausible pairing correction reverses them; only their interval estimates change. A definitive pairing-corrected re-analysis using the Bulletin's published pairing decisions remains outstanding and would tighten all of these estimates.\u003c/p\u003e \u003cp\u003eMass is the second worry. Small post-1969 specimens recovered from hot deserts have integrated 10⁴\u0026ndash;10⁵ years of weathering and ablation; their pre-atmospheric masses were almost certainly larger than what we measure today, so part of the apparent post-1969 mass collapse is recovered-mass attrition rather than recovered-population shift. The Bulletin's \u003cem\u003eMass\u003c/em\u003e field is a recorded mass, not a pre-atmospheric mass, and is documented by the Bulletin maintainers as approximate and sometimes representative of total known weight rather than individual stone mass. The 147-fold catalogue-scale contrast we report is too large to be erased by any plausible mass-rounding model. Its precise magnitude is a recorded-mass contrast rather than an in-falling-population contrast.\u003c/p\u003e \u003cp\u003eA further limitation concerns the \u003cem\u003eFell\u003c/em\u003e time series itself. The Mann\u0026ndash;Kendall result on this series returns Z\u0026thinsp;=\u0026thinsp;+\u0026thinsp;2.25, p\u0026thinsp;=\u0026thinsp;0.024 \u0026mdash; a small positive trend that we attribute provisionally to improving modern reporting infrastructure (fireball camera networks, smartphone observations, social-media propagation) but that we cannot, with the catalogue alone, distinguish from a true small flux change. The Monte-Carlo power analysis bounds the size of any such effect: a 50% cumulative trend would have been detected with 94% statistical power, and the trend we now detect is at the edge of that 50% threshold. The \u003cem\u003eFell\u003c/em\u003e sub-sample is therefore the least-biased reference available, but it is not free of secular biases, and any quantitative use of it should accommodate the residual uncertainty documented here.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003ch2\u003eFuture work\u003c/h2\u003e \u003cp\u003eSeveral follow-up analyses are suggested by the present scan. First, a quantitative pairing-corrected re-analysis of the Antarctic and hot-desert lunar/Martian fractions, using the published Bulletin pairing decisions, would clarify how much of the 11.7\u0026times; lunar contrast is a true delivery-time effect and how much is a recognition / pairing artefact. Second, a temporal analysis of the \u003cem\u003eFell\u003c/em\u003e sub-sample's class composition over the 19th and 20th centuries would test whether recognition of new classes (e.g. ureilites, brachinites) is real flux or recognition-driven, and would also help separate the small \u003cem\u003eFell\u003c/em\u003e trend documented here into its reporting-improvement and flux-change components. Third, a forward-model of expected \u003cem\u003eFound\u003c/em\u003e class fractions given a hypothesised in-falling flux and a parametric recovery-bias model could be fit to our era- and region-stratified data, providing the first formal estimate of the human fingerprint on the catalogue against which future recovery campaigns can be benchmarked. Fourth, an adjusted regression model \u0026mdash; log-mass and class-binary outcomes regressed jointly on era \u0026times; region \u0026times; fall-status \u0026mdash; would convert the per-question contrasts we report into a single set of adjusted odds ratios; we leave this synthesis for a follow-up paper.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusions","content":"\u003cp\u003eA stratified scan of the 85,775-record live Meteoritical Bulletin Database, supplemented with adjusted-regression and pairing-uncertainty Monte-Carlo audits, confirms at population scale that the apparent meteorite record is dominated by recovery effort rather than by in-falling flux. The principal findings are these.\u003c/p\u003e \u003cp\u003eFirst, the mass and iron contrasts. The median recorded mass collapses by a factor of \u003cb\u003e147\u003c/b\u003e between pre-1969 and post-1969 records at specimen level (4,400 g \u0026rarr; 30 g; Cliff's \u003cem\u003eδ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.81). Pairing-uncertainty Monte-Carlo deflates the fall-level mass collapse to a median of \u003cb\u003e77\u0026times;\u003c/b\u003e (95% CI [63, 96]\u0026times;); the iron-meteorite fraction collapses by \u003cb\u003e34\u0026times;\u003c/b\u003e at specimen level and is essentially insensitive to pairing because pre-1969 records pair near 1\u0026times;. Huber-style regression confirms that era is the dominant adjusted predictor of mass (post-1969 dex shifts of \u0026minus;\u0026thinsp;0.92\u0026thinsp;\u0026plusmn;\u0026thinsp;0.04 and \u0026minus;\u0026thinsp;1.68\u0026thinsp;\u0026plusmn;\u0026thinsp;0.03) and that the era main effect on iron survives simultaneous control for region and fall-status (Antarctic-era OR\u0026thinsp;=\u0026thinsp;0.18, Desert-era OR\u0026thinsp;=\u0026thinsp;0.024, both \u003cem\u003ep\u003c/em\u003e ≪ 10⁻\u0026sup3;). Now the witnessed-fall side. The \u003cem\u003eFell\u003c/em\u003e sub-sample sits at 62\u0026thinsp;\u0026plusmn;\u0026thinsp;13 events per decade since the 1860s; the Mann\u0026ndash;Kendall test returns \u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;+\u0026thinsp;2.25, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.024, a small but significant positive trend that a 10% reporting-improvement multiplier across 150 years already renders non-significant. We treat the \u003cem\u003eFell\u003c/em\u003e sub-sample accordingly as a least-biased reference rather than an unbiased one. The post-1969 catalogue surge is overwhelmingly attributable to recovery and cataloguing effort rather than to any plausible flux change.\u003c/p\u003e \u003cp\u003eSecond, geographic recovery hotspots impose a strong second-order bias on the apparent record. The headline 11.7\u0026times; Arabia-vs-Antarctic specimen-level lunar contrast shrinks to a pairing-MC fall-level median of \u003cb\u003e4.31\u0026times;\u003c/b\u003e (95% CI [2.04, 9.73]\u0026times;) \u0026mdash; comfortably bounded above parity but materially smaller than the unstratified figure. Adjusted logistic regression confirms that region is the dominant predictor of lunar fraction (Sahara OR\u0026thinsp;=\u0026thinsp;142, Arabia/Oman OR\u0026thinsp;=\u0026thinsp;26, vs Other; \u003cem\u003ep\u003c/em\u003e ≪ 10⁻\u0026sup3;), with an additional Desert-era main-effect boost of order 5\u0026times;. Both era and region effects therefore survive simultaneous control.\u003c/p\u003e \u003cp\u003eThird, the catalogue's headline negative latitudinal-mass correlation (Spearman ρ = \u0026minus;0.39 against |φ|) collapses to ρ\u0026thinsp;\u0026asymp;\u0026thinsp;0 within the witnessed-fall sub-sample. This is an instance of dataset-induced spurious physical inference. A sub-population that dominates one corner of a covariate space generates a pooled correlation whose sign is the opposite of the within-stratum correlation, with no mechanistic basis. Downstream analyses claiming a latitudinal-flux dependence should restrict themselves to the \u003cem\u003eFell\u003c/em\u003e reference population, and analogous \u003cem\u003eFell\u003c/em\u003e-style stratified diagnostics should be standard practice for any community catalogue dominated by a single recovery campaign in any covariate corner.\u003c/p\u003e \u003cp\u003eThe database has been doubled, and then doubled again, by two distinct human prospecting campaigns. Their statistical fingerprints \u0026mdash; small-stone Antarctic blue-ice recoveries (1969\u0026ndash;1999) and hot-desert pavements (2000+) \u0026mdash; are large enough to dominate every descriptive statistic computable from the full catalogue. The witnessed-fall sub-sample remains the appropriate reference for flux-based questions; the \u003cem\u003eFound\u003c/em\u003e sub-sample remains essential for material searches and for studies of the recovery process itself. The two should not be pooled without explicit bias correction. We release all input data, the joined and cleaned catalogue, the analysis pipeline, the adjusted-regression results, the pairing-uncertainty Monte-Carlo, and the figure-generation script as supplementary material so that every test reported here can be reproduced from raw data in under one minute, and so that the analysis can be re-run as the Bulletin Database is updated.\u003c/p\u003e "},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe author declares no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor contributions\u003c/h2\u003e \u003cp\u003eAll sections of this paper \u0026mdash; study design, data acquisition and cleaning, statistical analysis, software, figure preparation, and manuscript writing \u0026mdash; were carried out by the corresponding author.\u003c/p\u003e\u003ch2\u003eData availability\u003c/h2\u003e \u003cp\u003eThe Meteoritical Bulletin Database is publicly available. The Lunar and Planetary Institute hosts the live database at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.lpi.usra.edu/meteor/metbull.php\u003c/span\u003e\u003cspan address=\"https://www.lpi.usra.edu/meteor/metbull.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, including a CSV export option as of April 2026. The CSV snapshot used here was retrieved on 4 May 2026 and is included with the supplementary material for this paper. The cleaned, joined and class-tagged dataset (meteorites_clean_live.csv), the analysis script (analysis_live.py), the figure-generation script (make_figures.py), and the cached numerical results (results_live.json) are provided alongside.\u003c/p\u003e\u003ch2\u003eCode availability\u003c/h2\u003e \u003cp\u003eThe pipeline is small. Analysis (Python 3.11, NumPy 1.26, SciPy 1.13) is provided as analysis_live.py and figure generation as make_figures.py, with all intermediate results cached to results_live.json for replay. Total runtime is roughly 60 seconds on a standard laptop. No external services or random seeds are required.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBevan AWR, Bindon P (1996) Australian Aborigines and meteorites. Records Western Australian Museum 18:93\u0026ndash;101\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBevan AWR, Bland PA, Jull AJT (1998) Meteorite flux on the Nullarbor region, Australia. Geol Soc Spec Publ 140:59\u0026ndash;73\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBischoff A, Geiger T (1995) Meteorites from the Sahara: find locations, shock classification, degree of weathering and pairing. Meteoritics 30:113\u0026ndash;122\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBland PA, Smith TB, Jull AJT, Berry FJ, Bevan AWR, Cloudt S, Pillinger CT (1996) The flux of meteorites to the Earth over the last 50,000 years. Mon Not R Astron Soc 283:551\u0026ndash;565\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBland PA, Artemieva NA (2006) The rate of small impacts on Earth. Meteorit Planet Sci 41:607\u0026ndash;631\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBottke WF, Vokrouhlick\u0026yacute; D, Nesvorn\u0026yacute; D (2007) An asteroid breakup 160 Myr ago as the probable source of the K/T impactor. Nature 449:48\u0026ndash;53\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBuchwald VF (1975) Handbook of Iron Meteorites: Their History, Distribution, Composition and Structure. University of California Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurbine TH (2017) Asteroids: Astronomical and Geological Bodies. Cambridge University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCassidy W, Harvey RP, Schutt J, Delisle G, Yanai K (1992) The meteorite collection sites of Antarctica. Meteoritics 27:490\u0026ndash;525\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConnolly HC, Smith C, Benedix G, Folco L, Righter K, Zipfel J, Yamaguchi A, Aoudjehane HC (2007) The Meteoritical Bulletin, 92, 2007 September. Meteorit Planet Sci 42:1647\u0026ndash;1694\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEvatt GW, Smedley ARD, Joy KH, Hunter L, Tey WH, Abrahams ID, Gerrish L (2020) The spatial flux of Earth's meteorite falls found via Antarctic data. Geology 48:683\u0026ndash;687\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrady MM (2000) Catalogue of Meteorites, 5th edn. Cambridge University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrossman JN (1999) The Meteoritical Bulletin, 83, 1999 July. Meteorit Planet Sci 34:A169\u0026ndash;A186\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHalliday I, Griffin AA, Blackwell AT (1989) The flux of meteorites on the Earth's surface. Meteoritics 24:173\u0026ndash;178\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarvey R (2003) The origin and significance of Antarctic meteorites. Chemie der Erde \u0026mdash; Geochem 63:93\u0026ndash;147\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHofmann BA, Lorenzetti S, Eugster O, Kr\u0026auml;henb\u0026uuml;hl U, Herzog G, Serefiddin F, Gnos E, Eggimann M, Wasson JT (2009) The Twannberg (Switzerland) IIG iron meteorites: mineralogy, chemistry, and CRE ages. Meteorit Planet Sci 44:187\u0026ndash;199\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHutzler A, Gattacceca J, Rochette P, Braucher R, Carro B, Christensen EJ, Courn\u0026egrave;de C, Gounelle M, Laridhi Ouazaa N, Martinez R, Valenzuela M, Warner M, Bourl\u0026egrave;s D (2016) Description of a very dense meteorite collection area in western Atacama: insight into the long-term composition of the meteorite flux to Earth. Meteorit Planet Sci 51:468\u0026ndash;482\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKendall MG (1975) Rank Correlation Methods, 4th edn. Charles Griffin, London\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKing CW (1865) The Natural History of Precious Stones and of the Precious Metals. Bell and Daldy, London\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKorotev RL (2005) Lunar geochemistry as told by lunar meteorites. Chemie der Erde \u0026mdash; Geochem 65:297\u0026ndash;346\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKorotev RL (2012) Lunar meteorites from Oman. Meteorit Planet Sci 47:1365\u0026ndash;1402\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMann HB (1945) Nonparametric tests against trend. Econometrica 13:245\u0026ndash;259\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYamaguchi A, Kojima H, Imae N (2002) Antarctic meteorites collected by NIPR's expeditions to the Asuka site. Antarct Meteor Res 15:1\u0026ndash;11\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoshida M, Ando H, Omoto K, Naruse R, Ageta Y (1971) Discovery of meteorites near Yamato Mountains, East Antarctica. Antarct Record 39:62\u0026ndash;65\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"INDENPENT RESEARCHER","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"meteorite database, Antarctic Search for Meteorites, hot-desert recoveries, sampling bias, ANSMET, lunar meteorites, meteorite flux","lastPublishedDoi":"10.21203/rs.3.rs-9687218/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9687218/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe Meteoritical Bulletin Database underpins quantitative claims about meteoroid flux and composition, but its 85,775 valid and provisional entries combine specimens recovered under different search regimes. We treat the 1969 Yamato Mountains discovery and late-1990s hot-desert prospecting as two recovery-regime transitions and run a stratified scan together with adjusted-regression and pairing-uncertainty Monte-Carlo audits. Median recorded mass collapses \u003cb\u003e147-fold\u003c/b\u003e pre- to post-1969 at specimen level (4,400 g \u0026rarr; 30 g; Cliff's \u003cem\u003eδ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.81); a Monte-Carlo over literature pairing-factor ranges shrinks the fall-level mass-collapse median to \u003cb\u003e77\u0026times;\u003c/b\u003e (95% CI [63, 96]\u0026times;) but never below an order of magnitude. The iron-meteorite fraction falls from \u003cb\u003e29.9% to 0.9%\u003c/b\u003e at specimen level (Cohen's \u003cem\u003eh\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.97; pairing-MC fall-level: 34\u0026times; with tight CI). The Antarctic-versus-Arabia/Oman lunar-fraction contrast is \u003cb\u003e11.7\u0026times;\u003c/b\u003e at specimen level; the era-restricted 1969\u0026ndash;1999 contrast is \u003cb\u003e6.75\u0026times;\u003c/b\u003e; the pairing-MC fall-level median is \u003cb\u003e4.31\u0026times;\u003c/b\u003e with 95% CI [2.04, 9.73]\u0026times; \u0026mdash; comfortably bounded above parity but materially smaller than the unstratified figure. Decadal \u003cem\u003eFell\u003c/em\u003e counts average 62\u0026thinsp;\u0026plusmn;\u0026thinsp;13 from the 1860s through the 2010s; the Mann\u0026ndash;Kendall test returns \u003cem\u003eZ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;+\u0026thinsp;2.25, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.024, a small but significant positive trend that a 10% reporting-improvement multiplier across 150 years already renders non-significant (deflated \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.08) and a 20% multiplier eliminates entirely. The \u003cem\u003eFell\u003c/em\u003e sub-sample is therefore a \u003cem\u003eleast-biased\u003c/em\u003e reference rather than an unbiased one. Adjusted logistic regression with era \u0026times; region interaction confirms that era is the dominant predictor of iron fraction (post-1969 OR\u0026thinsp;\u0026asymp;\u0026thinsp;0.02\u0026ndash;0.18, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;10⁻⁴⁹) and region is the dominant predictor of lunar fraction (Sahara OR\u0026thinsp;=\u0026thinsp;142, Arabia/Oman OR\u0026thinsp;=\u0026thinsp;26, both vs Other; \u003cem\u003ep\u003c/em\u003e ≪ 10⁻\u0026sup3;); both effects survive simultaneous control. The catalogue's headline negative latitudinal-mass correlation (ρ = \u0026minus;0.39) collapses to ρ\u0026thinsp;\u0026asymp;\u0026thinsp;0 within witnessed falls \u0026mdash; an instance of dataset-induced spurious physical inference driven by single-collection dominance of one corner of the covariate space. Users should restrict flux inferences to the \u003cem\u003eFell\u003c/em\u003e sub-sample with explicit acknowledgement of the documented small reporting trend, or apply the adjusted era-, region-, and pairing-stratified weighting we provide here.\u003c/p\u003e","manuscriptTitle":"The Human Fingerprint on the Meteorite Record: A Full-Database Scan of the Meteoritical Bulletin (n = 85,775)","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-14 14:38:17","doi":"10.21203/rs.3.rs-9687218/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bb62378d-52d6-4bc3-bcf5-a03d12ab611d","owner":[],"postedDate":"May 14th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-14T14:38:17+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-14 14:38:17","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9687218","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9687218","identity":"rs-9687218","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}