We tested survey temperature and recorded bird taxa in California, Oregon, New York, and Pennsylvania. We used accepted Breeding Bird Survey runs and 7,068 consecutive-year pairs with the same route and observer. We fit separate first-difference models in western and eastern states for 1990-2006 and 2007-2024. We adjusted for date, time, duration, assistant status, observer experience, wind, sky, and state-year effects. Slopes ranged from 0.06 to 0.35 additional recorded taxa per 5 C higher start temperature. All point estimates were positive. None excluded zero after correction for four tests. All four corrected intervals were within -1 to +1 taxon per 5 C under the model. Three sensitivity checks gave the same broad result. These are conditional weather associations in accepted survey conditions. They cannot separate detection from ecological change. Missing data, roadside sampling, observer changes, and dependence across routes limit inference. We did not test pollution, extreme heat, long-term warming, or species loss. The package includes code, source hashes, filtered records, model designs, uncertainty estimates, and offline reproduction instructions.
We tested whether light from the known eclipsing source KOI-3688.01 can explain the shallow signal on KOI-1924.01. We used public Kepler quarters 8, 9, 12, and 13. A calculation that treats the source SAP depth as intrinsic requires 2.46 +/- 0.64 times its total light in quarter 8. The archive crowding correction reduces this model-dependent requirement to 0.65 +/- 0.17. All four corrected point estimates lie between zero and one, but aperture and magnitude errors are not included. A scale set at phase zero underpredicts the quarter 13 secondary-phase depth by about 2 to 2.5 estimated standard errors across two baseline windows. Thus, the flux budget is feasible under the model, but exact shared-light scaling is not established. This work tests a previously published timing pair. It establishes no new planet or source association. No pixel localization or independent telescope observation was used. The package includes all eight light curves, hashes, two source catalog rows, code, event depths, figures, and offline reproduction instructions.
We tested whether the target neuron membrane time constant improves a prediction of synaptic voltage decay more than the source neuron time constant. We used 1,576 connections in the Allen Synaptic Physiology r2.1 small database. We held out whole inferred donors and fit ridge models with cell labels, region, and cell-body geometry. At the primary penalty of 10, no gain interval excluded zero after adjustment for eight comparisons. In 473 mouse excitatory connections, target time reduced donor-weighted squared log error by 5.12% against the base model and 5.58% against the source-time model. The adjusted intervals were -1.51% to 10.80% and -1.10% to 11.44%. A penalty of 100 gave positive adjusted intervals in that group. The result therefore depends on model regularization. Human and inhibitory samples did not give a clear primary gain. The package supplies code, selected connection records, held-out predictions, a figure, and an independent score check. The intervals condition on fitted models; slice sampling, missing measurements, and inferred donor labels limit inference. This test does not establish a membrane-filtering mechanism or a rule for intact brains.
We tested a local centroid model for the known D800 and D1520 dips of KIC 8462852 in Kepler quarters 8 and 16. The model uses the exact flux-loss factor L/(1-L). Four fits with pointing terms give coefficient magnitudes of 0.02770 to 0.02802 pixel for D800 and 0.00430 to 0.00448 pixel for D1520. Thus, one fixed measured coefficient does not describe both events under these choices. Removing the pointing terms changes the estimates. The result agrees in general with earlier work on variable centroid response. It does not show that the dips occur on different stars. The aperture masks differ, both quarters use channel 56, and all PSF centroid values are missing. Conditional noise intervals do not include all detector and baseline errors. We establish no new dip, planet, or source position. The package includes the two raw light curves, exact hashes, code, all fits, figures, and offline instructions.
We tested the availability of cycle constraints on electrical coupling in Allen mouse visual cortex slices. In a stable passive reciprocal DC network, the product of voltage coupling coefficients around a loop equals the reverse product. This constraint cancels individual port resistances and survives hidden passive cells. We required two inhibitory cells, two positive electrical calls, two positive coefficients, and pulse correlation p values below .05 in both directions. The sample had 34 reciprocal edges in 30 recordings, but only one independent loop. Its products differed by a factor of 1.00490. Removing the pulse p filter gave 112 edges in 107 recordings and still only one loop. The data therefore give little support for a population test of this constraint. The single close product does not establish passive reciprocity. Finite pulse windows, state changes, measurement errors, and selection limit interpretation. We supply a graph diagnostic, a proof of its scope, source provenance, derived edges, and independent numerical checks.
We tested the pixel pattern of the known shallow signal on KOI-3006.01. At the nearby source eclipse times, the lost-light centroid is displaced from the mean-light centroid toward KIC 8758161 by 1.77 pixels in Kepler quarter 5 and 1.23 pixels in quarter 9. Block bootstrap 95 percent intervals are 1.31 to 2.45 and 0.84 to 1.62 pixels. The positive direction remains in time halves, with a wider baseline window, and with linear pointing predictors. Quarter 2 does not resolve the spatial displacement. These results add pixel evidence for the earlier light contamination explanation. The source lies outside the pixel stamp. We do not fit a pixel response function or identify the true binary host. All data are from Kepler, so this is not independent telescope confirmation. No new planet is established. The package includes all used pixel files, hashes, derived event images, code, and reproduction instructions.
We tested whether the depth correlation reported in Kepler quarters 4 to 7 of KIC 12557548 repeats in later data. Fixed extraction settings give 970 accepted transits and 905 consecutive-orbit pairs in quarters 8 to 15. The pooled correlation is 0.140, with a 95% block interval of 0.061 to 0.214. A fitted slow trend and fixed rotation terms reduce it to 0.048, with a conditional interval of -0.029 to 0.121. Quarter 15 has the largest raw value, 0.452. Quarters 10 and 14 revisit the detector channel used in quarter 6 but give a pooled value of 0.051. Gaussian simulations with the actual gaps show that fitting the slow model can bias the residual correlation below zero. Thus, an interval that contains zero does not establish independent events. The data support a weak and time-dependent depth relation. They do not identify dust survival as its cause. The package contains raw files, hashes, code, event tables, figures, and offline reproduction steps. No new object or independent telescope confirmation is claimed.
We tested timing in closed excitatory three-cell circuits from Allen Synaptic Physiology r2.1. We compared each direct response delay with the sum of two indirect delays. The sample had 17 routes in mouse visual cortex and 12 in human temporal cortex, from 10 and eight inferred donors. The direct delay was shorter in 15 mouse routes and all human routes. Two mouse routes had indirect sums earlier by 0.011 and 0.128 ms. These differences exclude the time needed for the relay cell to fire. With measured delays fixed, relay time of 0.15 ms removes both apparent advantages in an additive model. If each edge can have 0.05 ms of error, neither positive margin gives a guaranteed ordering. The package supplies a measured-delay benchmark and exact conditional error bounds. Small selected slice circuits, shared cells, unknown delay errors, and absent relay-spike measurements limit the result. It does not establish actual route speed or a general species rule.
We audited changes in the ozone site assigned to repeated bird surveys in California, Oregon, New York, and Pennsylvania in 2005-2019. Only 28 of 1,394 eligible consecutive-year pairs with the same route and observer changed sites. None had at least 24 valid days at both sites in both annual exposure windows. Thus, the planned two-site comparison could not separate annual ozone change from site substitution. An exploratory one-site comparison was possible for 23 pairs. Site substitution contributed at least 2 ppb to 11 differences, and five differences reversed sign with a fixed site. A common-relative-day check reduced sign reversals to three of 22 pairs. Same-site matching removed few pairs in this selected panel, but this does not bound ecological model bias. Missing overlap, sparse switches, exposure error, and selection limit inference. We do not estimate an ozone or climate effect on birds. The package includes source hashes, daily evidence, pair records, code, figures, and offline checks.
We tested signed third central moments of two known deep dips of KIC 8462852 in Kepler release 25 data. For D1520, the moment stayed negative in 36 tested combinations of flux product, quality selection, local baseline, and event window. Its range was -0.576 to -0.390 day cubed. For D800, 14 of 36 values were positive. The primary noise perturbation ranges did not capture this baseline sensitivity. A symmetric-kernel calculation explains a conditional link to the direction of an exponential tail. A hidden-wing calculation shows why finite local moments cannot reject a full dust-tail model. The reported background changes are much smaller than the dips. Both quarters use the same detector channel. We establish no new object or dust-tail direction. No independent telescope data confirm these moment values. The package includes raw inputs, hashes, code, all analysis choices, figures, and reproduction instructions.
We tested prior 30-day ozone and survey temperature in 1,366 consecutive-year bird survey pairs from California, Oregon, New York, and Pennsylvania in 2005-2019. We matched route, observer, and ozone site and fit separate region-period models with survey controls and state-year effects. Interaction estimates ranged from -0.132 to +0.297 recorded taxa per 5 C and 10 ppb. All four primary intervals included zero. All 16 sensitivity intervals corrected within each four-test mode included zero. No matched survey had both a start temperature of at least 25 C and a 30-day ozone mean of at least 70 ppb. The test cannot resolve joint extremes. Records overlap earlier campaign studies. Selected coverage, exposure error, seasonal weather, and detection limit inference. The package includes code, source hashes, exposure evidence, model designs, influence checks, and reproduction instructions.
We tested long transit durations in the Kepler DR25 catalog. A screen of 346 high-score candidates selected KOI-3105.01. Its reported duration is 3.36 times the maximum central circular duration under the DR25 stellar model. The Gaia-based mass and radius from Berger et al. (2020) reduce this ratio to 1.16. Favorable changes within the quoted input errors can bridge the remaining excess. Thus, the old duration excess does not establish high eccentricity. A fixed-time check of four Kepler quarters gives a pooled dip in both SAP and PDC flux. The later quarters are weak. These data do not locate the host or independently measure the duration. The stellar-model errors lack a joint covariance, and a blended source or detector effect is not excluded. We find no new planet. The package contains the catalog screen, source records, four light curves, hashes, code, uncertainty checks, a figure, and offline reproduction instructions.
We tested whether source axons marked intact had a larger mean detected absolute resting voltage response per tested cell pair than axons marked truncated. We used Allen Synaptic Physiology r2.1 small data. We matched recording, ordered cell classes, and 100 micrometre distance bins. Pairs called unconnected received zero. Matched groups with a missing response estimate for a detected connection were excluded. In 671 mouse visual-cortex pairs from 68 inferred mice, the difference was 0.00648 mV per pair. The nominal 97.5% donor bootstrap interval was -0.00916 to 0.01989 mV. In 128 human temporal-cortex pairs from eight inferred donors, the difference was -0.01091 mV, with interval -0.05588 to 0.00097 mV. Neither primary interval excluded zero. A mouse high-spike sensitivity test gave a positive descriptive interval. About 28% of detected connections in each eligible sample lacked voltage estimates. Sparse matches, missing measurements, inferred donors, and residual cell differences limit the result. The method tests detected coupling in selected slices. It does not estimate lost synapses or intact-brain coupling.
We measured contact size concentration in one Allen V1 Deep Dive mouse volume. We reused a verified table of excitatory contacts and selected 1,519 connections with at least three observed contacts. The largest contact held a mean 46.72% of total cleft size. One contact held more than half in 499 connections. Mean size-weighted effective count divided by contact count was 0.8278. Under an independent contact dropout model with linear size weights and equal survival probability, equal weights underestimate normalized variance by a mean factor of 1.2432. Observed sizes were more balanced than broad shuffles that preserved one cell, both cell types, and contact count. Position control removed clear separation for the source shuffle and left few movable contacts. These conditional results do not isolate a connection-specific process. One animal, incomplete arbors, size as a proxy, and no matched physiology limit the result. The package includes contact evidence, source code, null draws, provenance, a figure, and an independent audit.
We tested nearby Kepler signals for matching periods and transit times. A strict timing screen found 707 pairs in a selected sample of 7,679 DR25 objects. Four pairs had no ephemeris contamination flag on either member. The two DR25 candidates in these four pairs are already false positives in the retrieved cumulative catalog. We checked KOI-3006.01 and its nearby deep eclipsing source in six Kepler quarters. The shallow target depth changes with detector season, while the source eclipse persists. SAP and PDC flux give similar results. This supports light contamination and the existing false-positive classification. It does not establish a new planet. The package includes a timing audit, phase controls, all used light curves, data hashes, code, and a reproduction script. No pixel data or independent telescope was used, so the light transfer path and the true binary host remain uncertain.
We tested whether current ozone and seasonal warmth predict the absence of an accepted bird survey in the next year. We joined a published US Breeding Bird Survey panel to the full USGS survey log. The pooled model used 23,816 surveys on 2,000 routes. It included route-observer and state-year effects and survey and weather controls. At 75 F, the ozone estimate was +1.22 percentage points of absence per 10 ppb, with a 95% interval from -1.38 to +3.82. The ozone by temperature term was +0.12 percentage points per 10 ppb and 5 F, with an interval from -1.86 to +2.11. All eight intervals in four separate region and time samples included zero after correction. Route trends, closer monitors, and a broader survey outcome gave no clear link. This result does not show that missing surveys are random. Selected monitor coverage, missing controls, small western samples, and a linear model limit inference. This is an observation-process test, not a test of bird loss or climate-change effects. The package has the input panel, follow-up labels, source hashes, code, an independent numerical check, tables, and a figure.
We tested a reported change in cherry bloom temperature sensitivity with shared year effects and site trends. We reused a published panel of 4,454 Japanese and 4,319 Swiss site-years. The outcome was observed bloom day minus bloom day from an early-period forcing-only model. With site-period effects, the late-minus-early slope change in Japan was -1.03 days per degree C. Its 97.5% interval widened from -1.26 to -0.80 with site clusters to -1.90 to -0.16 with site and year clusters. With year effects and site trends, the estimate was -0.68 (-1.75 to +0.39). The Swiss daily-temperature estimate was +0.10 (-0.91 to +1.11). Five-year cluster blocks and site-coverage checks did not give clear adjusted changes. Year effects remove common regional signals, so these results do not refute a climate response. They show a limit to precision and conditional robustness. This is a reanalysis of the same data, not an independent replication. It does not measure pollution or biodiversity loss. The bundle has code, the source panel, date checks, hashes, numerical checks, tables, and a figure.
We tested whether autumn-minus-winter mean sea level changed between 1960-1989 and 1996-2025 at six US tide gauges. We used fixed NOAA monthly records and 23-30 complete years per site and era. The contrast is September-November minus January-March in the same year. All six point estimates increased, by 0.59 to 4.20 cm. All primary 95% five-year block intervals for the six-site family included zero. The largest estimate, at Sewells Point, had an interval of -0.50 to 8.90 cm. Ten-year blocks gave a positive family interval there, while three-year blocks did not. Accepting inferred values, removing a simple within-year trend contribution, and using matched 19-year eras left all family intervals across zero. The changes therefore remain uncertain under the primary test. This result does not show a fixed seasonal cycle or refute prior regional findings from other periods. Missing years, only six harbor gauges, data processing, and dependence assumptions limit inference. No physical cause or flood frequency is established. The package contains all monthly input fields, source hashes, code, saved draws, figures, and an exact offline repeat.
We tested whether lilac first-bloom dates had different temperature slopes in warm and cool springs. We used 9,015 repeated plant records in separate eastern and western US samples in 1965-1986 and 1987-2008. Models used plant and year effects, winter temperature, and separate March-April temperature slopes above and below each site mean. Warm minus cool slopes were -0.72, -2.02, -1.42, and -4.35 days per degree C. All four state-cluster intervals, adjusted for four tests, included zero. Two-way state and year intervals also included zero. Removing flagged dates, changing record-length rules, giving equal plant weights, and trimming large anomalies did not give a clear common difference. The raw-data rebuild matched the reused panel byte for byte. Monthly weather, unknown visit delays, selected sites, different lilac types by region, and few clusters limit the result. We do not test daily heat waves, pollution, biodiversity loss, or causal climate effects.
We tested whether July-September Nino 3.4 improves December-February mean sea-level forecasts beyond local summer sea level and a time trend. Separate fits used earlier complete winters only. At San Francisco and Seattle, the equal mean of added-predictor MSE skills was 8.98% for test winters 1986-2025. Its conditional 95% five-year block interval was -32.87% to 32.99%. All six individual site intervals included zero. Removing the 1998 and 2016 test winters reduced Pacific skill to 1.51%. A paired-support check gave 8.79%. The result leaves added forecast value uncertain under the tested model. It does not show that ENSO has no effect. Fixed retrospective sources, missing seasons, few large events, and conditional uncertainty limit inference. This test measures seasonal mean levels, not floods or physical causes. The package has source hashes, exact used-field inputs, all forecasts, uncertainty draws, code, figures, and a successful numerical audit and offline repeat.