We tested one claim out of sample: the Gutenberg-Richter b-value decreases when the tidal shear stress amplitude increases (Ide et al. 2016). We used 15,100 shallow GCMT earthquakes (depth 70 km or less, Mw 5.5 or more) from 1976 to 2026. The test period, 2017-01-01 to 2026-11-05, has 3,155 events and starts after the claim. For each event we calculated the body-tide shear stress on the fault plane. We used three metrics: the 24-hour stress range, the rank of that range against the 15 previous days, and a spring-tide index. We compared b in the high and low thirds of each metric. The null model was 300 random time shifts of the full catalogue, which keep location, mechanism and clustering. In the test period, the predicted decrease did not occur: delta b was +0.062 for stress range (null mean +0.069, p = 0.40) and -0.027 for rank (null mean 0.000, p = 0.32). The 1976-2016 data also did not show a decrease, so our model did not reproduce the original result. For 1976 to 2026, the data exclude a tidal decrease in b between thirds larger than 0.010 to 0.039, depending on the metric. The largest events (Mw 7.0 or more) did not occur preferentially at high tidal rank or near new or full moon. A raw comparison by absolute stress range gave a positive contrast in 1976-2016 (+0.056, bootstrap interval +0.008 to +0.108). The shift null shows that this contrast comes mostly from location and fault type, not from tidal timing. Main limitations: no ocean tide loading, a simple spherical Earth model, metric definitions that are not the same as in the original study, and low power for small effects.
Question: does the point-to-point axon density between two positions in mouse visual cortex, predicted from Allen anterograde tracer data, tell which cross-area unit pairs have a short-latency directed spike link? Approach: We made a 100 um flatmap kernel-regression model of visual-cortex projections from 289 Allen tracer injections (leave-one-out r of log density 0.88 at a 200 um kernel; 0.63 without position). For each directed cross-correlogram test between units in two of six visual areas, the model gave the predicted density D from the source unit position at the target unit position. Data: Allen Visual Coding Neuropixels (passive; 3.37 million tests, 5,921 links, 50 mice) and Allen Visual Behavior Neuropixels (task; 4.84 million tests, 33,831 links, 34 mice). We used Poisson models with absorbed fixed effects and mouse-clustered errors. Result: Without a distance control, D predicted links in Visual Coding (rate ratio 2.18 per 10-fold D, z = 4.97). A spline of inter-unit distance removed most of this association (1.22, 95% CI 0.92-1.62). In Visual Behavior, D had no association at any level of adjustment (main model 0.94, 0.77-1.16). Distance alone was a strong predictor in both datasets: the link rate at 3 mm was 0.19 and 0.13 of the rate at 0.5 mm. In five variants (strict links, wild-type tracers, kernels of 100 and 400 um), Visual Coding estimates were always above 1 (3 of 10 significant) and Visual Behavior estimates were always below 1. No pooled estimate differed from 1. The reverse-direction density was not a better predictor. Limitation: D is a model prediction at tracer resolution (injections of about 0.5-1 mm) from other animals. Probe placement limits the range of D. A CCG link is not proof of a monosynaptic connection.
Question: does a cold winter decrease bird counts in the next breeding season, and only in the species that were present in that winter? Approach: we used 86,809 same-observer pairs of consecutive-year North American Breeding Bird Survey counts on 4,508 routes (winters 1968-2019) and the December-February temperature anomaly of the GHCN_CAMS 0.5 degree cell of each route. We compared the change in a resident-species index (130 species) with the change in a long-distance-migrant index (114 species) on the same survey. The migrants were absent in winter, so they control for observer, date, weather and route effects. Models had route and year fixed effects, a spring-anomaly covariate and standard errors clustered by year. Result: each +1 C of winter anomaly increased the resident-minus-migrant change by 0.0133 log units (95% CI 0.0097 to 0.0169). Residents responded (0.0128) and migrants did not (-0.0004, CI -0.0027 to 0.0019). The response was asymmetric: winters more than 3 C colder than normal were followed by a resident index approximately 7% lower (CI -9.0% to -4.5%), with no migrant change. A placebo exposure (the winter after the survey) had no effect. The slope replicated in eastern and western, northern and southern routes, in 1968-1993 and 1994-2019, and with four alternative guild definitions. Canadian routes alone gave an uncertain slope. Of 83 resident species, 77% had positive slopes (12 clearly positive, 0 clearly negative); migrant species were centred on zero. Body mass did not predict sensitivity. An out-of-sample test on Winter Storm Uri (February 2021) was inconclusive: on 110 severe southern routes the 2019-2021 contrast fell by 0.110 relative to milder routes, but mainly because migrants increased, and similar contrasts occurred in 2 of 20 placebo year pairs. Limitations: monthly gridded temperature does not resolve short extremes; resident-specific detection changes after cold winters cannot be fully excluded; guild lists use expert knowledge of winter ranges because raw AVONET migration codes misclassify many North American breeders.
Question: is the change of total land water storage (TWS) symmetric in time, and does rain skewness explain any asymmetry? Data: CSR GRACE/GRACE-FO RL06.3 mascons, 2002-04 to 2026-06, 2,168 land cells of 2.5 degrees (56 S to 75 N, no Greenland or Antarctica); GPCP v2.3 rain; GHCN-CAMS temperature. Method: after removal of a linear trend and seasonal harmonics, we calculated the skewness g(L) of storage changes x(t+L) - x(t) for L = 1-24 months. A reversible process gives g = 0. We used 1,000 common-phase Fourier surrogates (reversible, with the observed spectra and cross-spectra) and a sign-flip test on 111 blocks of 15 degrees. We also compared development and recovery times of deficit and surplus events, and drove single-store, two-pole ARX and state-dependent ARX models with observed rain. Results: the area-weighted mean g was positive at all lags (0.076 at L = 1, 0.099 at L = 12; surrogate 95% ranges -0.018 to 0.019 and -0.038 to 0.039; p = 0.001; block p < 0.001). Each mission showed it separately (GRACE g(12) = 0.088, GRACE-FO 0.074). Storage rose faster than it fell. The effect was large in arid (g(12) = 0.19) and semi-arid (0.22) cells and absent in humid cells (0.013, p = 0.66). With a linear trend, deficits ended about 6% faster than they formed (mean ln(R/D) = -0.059, p = 0.010), but with a quadratic trend this fell to -0.026, inside the null range. Thus the deficit result is not robust. Surpluses formed faster than they decayed (+0.098, p = 0.003). Monthly rain anomalies were positively skewed on 97% of the land area. Rain-driven models gave more skewness than observed at L = 1 (0.26-0.37), but only 0.018-0.041 at L = 12 and below zero at L = 24 (observed 0.074). The rain gain did not depend on the storage state (difference -0.013, 95% CI -0.063 to 0.037). Conclusion: rain pulses through a store explain the short-lag asymmetry, but less than half of the year-scale asymmetry. Untested candidate causes include human water use, lakes and reservoirs, and nonlinear recharge. Limitations: approx. 300 km resolution with correlated cells; 24-year record; Gaussian surrogates; simple rain models; complete surrogate tests only for the linear trend because the host was overloaded.
Question: does the anatomical input of a mouse visual cortex neuron predict the reliability of its responses to repeated natural movies? Pooling of more inputs and stronger inhibition both predict a positive link. Approach: in MICrONS (12,891 coregistered neurons, V1 and higher visual areas) we related CC_max (response correlation ceiling across movie repeats) to the total input synapse count of each neuron (materialization v1507) and to the fraction of typed inputs from inhibitory cells. Controls were scan, cell type, area, 25 um depth bins, lateral position and nucleus volume. CIs are cluster-robust over 100 um cubes. We wrote the plan before we computed outcomes. In V1DD (second mouse, 180 neurons with complete input lists) we tested total input count only. Results: in MICrONS, input count gave 0.047 SD per SD (95% CI 0.011 to 0.083) and inhibitory fraction 0.040 (0.005 to 0.075). Both are below the pre-set 0.05 SD threshold and together explain about 0.2% of residual variance. The input-count effect disappears with scan-only controls, and both effects change sign between cell types. The 890 proofread-dendrite MICrONS neurons gave -0.049 (-0.169 to 0.071). In V1DD, input count gave 0.24 (-0.31 to 0.77); this sample cannot test effects of the MICrONS size. Conclusion: in MICrONS, total input synapse count is a weak determinant of response reliability; effects above about 0.08 SD per SD are excluded. Limitations: one well-powered animal; automatic segmentation adds count noise that attenuates slopes; the typed inhibitory fraction is biased by missing excitatory somata and did not correlate with a compartment-based measure; no per-cell signal-to-noise control; observational data.
Question: in the MICrONS mouse visual cortex, do connected neurons have more similar responses because of a property of the pair, or because of a property of one cell? A campaign study of the V1 Deep Dive mouse found that cell-level control removed the pair-level effect. Approach: we used the public tables of Ding et al. (2025): 148 presynaptic neurons with proofread axons, 11,035 postsynaptic neurons, 295,371 axon-dendrite cotravel pairs and 8,128 connected pairs. We wrote the analysis plan before we computed results. Pair level: after control for cotravel length, the odds ratio (OR) of connection per SD of in silico signal correlation was 1.16 (95% CI 1.13-1.19). With control for postsynaptic coupling and mean similarity, the OR was 1.14 (1.11-1.17). Conditional on the postsynaptic cell, the OR was 1.08 (1.05-1.11), or 1.12 with presynaptic controls. A linear model with presynaptic and postsynaptic fixed effects gave +0.39 percentage points per SD (base rate 2.75%). In vivo signal correlation and readout similarity gave the same result. Orientation similarity gave no clear effect. Layer/area pair types and receptive-field distance did not remove the effect (OR 1.16-1.19). Cell level: postsynaptic signal coupling predicted in-degree beyond cotravel (rate ratio 1.05 per SD, permutation p = 0.003), in V1 and L5 but not in L2/3 or higher visual areas. Presynaptic coupling did not increase out-degree. In subsamples of V1DD size (785 connections), the in vivo CI excluded 1 in 74% of draws, and 2.9% of draws gave OR <= 1.01 (the V1DD estimate). Limitations: one animal per data set; different similarity estimates and candidate pairs in the two data sets; no spontaneous or noise coupling in the release; 112 of 148 presynaptic cells are in RL and AL; some axons are proofread only for projections; subset and sensitivity analyses are exploratory.
Question: does the temperature of the preceding winter change next-summer bird counts more for species that stay at the route in winter than for species that leave it? Approach: we used North American Breeding Bird Survey counts (2025 release) and GHCN-CAMS monthly gridded temperature. The response was the change in log count of each species on each route from one year to the next, with the same observer in both years (4,927,559 pairs, 361 species, 4,097 routes). AVONET migration classes separated resident, partially migratory and migratory species. A route-year fixed effect absorbed all survey conditions shared by species (observer, date, weather, effort). Migratory species, which are absent in winter, were the negative control. Results: after a winter 1 C warmer than the 1981-2010 route mean, resident counts changed by 0.94% more than migratory counts (95% CI 0.72 to 1.17); partial migrants by 0.83% more (0.61 to 1.05). The migratory class slope alone was 0.17% per C (-0.05 to 0.39). Cold winters had larger effects than warm winters: after winters 3 C or more below normal, resident counts fell 5.0% (3.8 to 6.2) relative to migrants. The contrast was positive with a CI above zero in all six region-by-period samples (West, Central, East; 1967-1995, 1996-2024) and in all three winter-climate terciles. A placebo with the winter after the survey gave no contrast (0.12%, -0.19 to 0.43). About 60% of a one-winter change reversed in the next year. The contrast did not clearly change between periods. Limitations: BBS counts are indices and class-specific detectability changes cannot be fully excluded; AVONET classes are coarse and contain some errors; monthly 0.5 degree temperature does not capture snow, ice or the duration of cold spells.
Rate-and-state friction predicts that aftershock duration is inversely proportional to the tectonic stressing rate. We tested this prediction with 976 declustered shallow (depth 40 km or less) mainshocks of magnitude 6.0 or larger in the USGS ComCat catalog from 2008 to 2024. We used the geodetic strain rate of the Global Strain Rate Model v2.1 (GSRM) as the measure of loading rate. Negative-binomial models separated the pre-mainshock background rate from the magnitude-adjusted aftershock productivity (M 4.8 or larger, days 1-99). The modeled time for the aftershock rate to decrease to background decreased as strain rate to the power -0.48. The 95% cluster-bootstrap interval was -0.73 to -0.26. At Mw 7.0, the modeled time was approximately 690, 230 and 76 days at 10, 100 and 1,000 nanostrain/yr. The predicted value -1 was outside the interval in all ten planned sensitivity checks. The main cause was the background rate, which increased only as strain rate to the power 0.37 (0.18 to 0.60). Productivity decreased slightly with strain rate (power -0.18, -0.29 to -0.07). This productivity result was not robust: its interval included 0 at magnitude thresholds 5.0 and 5.2 and for mainshocks of M 6.5 or larger. An exploratory faulting-style control did not change it. Stacks in three strain-rate groups gave times to background of 241, 92 and 46 days, with wide intervals. GSRM is a smoothed and noisy measure of fault stressing rate. This noise makes all slopes smaller, so the result is an empirical scaling relative to GSRM. It does not directly reject the physical model. Catalog detection changes, background windows that include earlier sequences, and the extrapolation with one pooled Omori exponent are further limitations. The package contains the exact catalog snapshot, derived tables, code, figures and reproduction instructions.
Question: Do faint nearby eclipsing binaries (EBs) that Gaia DR3 catalogued cause some TESS planet candidates? Approach: For 8151 TESS Objects of Interest (TOIs), we found all Gaia DR3 EBs within 3 arcmin (4591 pairs). We tested if the TOI period is equal to the EB period multiplied by a small ratio (relative deviation < 1e-3). We also tested if the TOI transit epoch agrees with the propagated Gaia eclipse time (phase gap < 0.05). We did the same search around control positions 0.5 deg away. Results: The TOI cones gave 132 period and epoch matches. The control cones gave 0. Matches occur for 121 of 604 (20%) known false positives that have an EB in the cone, and for 0 of 226 confirmed or known planets. Eleven TOIs with the disposition PC match. For 10 of them, the EB is a separate star 4.5 to 16 arcsec away, and its Gaia eclipse is deep enough to make the TESS signal after dilution. These are probable nearby-eclipsing-binary false positives: TOI 2998.01, 3063.01, 2541.01, 3258.01, 3143.01, 4870.01, 1960.01, 4876.01, 1546.01 and 5008.01. For TOI 1546.01 (a 3 Earth-radius candidate on a T = 9.9 star), ZTF photometry independently gives the EB period of 1.1328272 d and a 0.6 mag eclipse 0.41 h from the TOI epoch. TESS difference images put the signal on the EB for TOI 1960.01 and 2541.01. They put the signal on the target for two wide period-only matches (TOI 699.01 and 6496.01), so those matches are chance coincidences. For TOI 3142.01, the EB is the target star, which puts the signal on the target. Limitations: Gaia DR3 contains only a subset of EBs. Eight of the ten probable false positives have separations below 0.55 TESS pixels, so the TESS pixel data cannot resolve them. Gaia epoch photometry was not available during this work. Follow-up photometry is necessary for a final disposition.