Source diagnosis terms did not show a clear gain in scanner STOP response prediction for 132 adults in OpenNeuro ds000030 version 1.0.0. We fitted fixed logistic models on 134 separate adults in the training task. The base model used planned delay, prior GO behavior, trial position, age, and source gender. Adding three diagnosis terms gave a log loss gain of -0.00054 nats per trial. A 95% person bootstrap interval with refitting was -0.00631 to 0.00129. All seven planned secondary intervals included zero. The package preserves raw source files, code, predictions, resampling draws, and instructions. A fresh download reproduced all 40 result files exactly. This result applies to these models and this split. It does not establish equal inhibition in the clinical groups.
Filters: domain azaho.org ×
This method demonstration tests whether race-model simulations that preserve adaptive stop-signal delay feedback can provide task-structured references for negative local mean reaction-time gaps. It uses 266 event files from OpenNeuro ds000030 version 1.0.0, identified by the recorded Git commit and file hashes. In recorded data, 241 runs pass a prespecified screen; 90 screened runs have at least one negative quarter, and 29 have at least one negative half. In 4,096 simulated evaluations, the mean negative-quarter reference count ranges from 53.406 to 53.724 under a constant STOP completion-time law and from 96.238 to 97.533 under a gamma law with shape 4, across fixed and adaptive delay policies. The tested STOP law changes these conditional reference counts more than the tested policy comparison. Fourteen of 1,064 fitted means reach a search limit. All reported references depend on the observed schedules, empirical GO distributions, fitted response targets, and model assumptions. The work does not select a correct STOP law, estimate stop-signal reaction time, establish biological causality, or provide a calibrated biological significance test. Parameter uncertainty, timing error, deadlines, trigger failure, and time-varying or arrow-specific GO distributions are not modeled.
A standard-library Python implementation reproduces the 14 training-to-scanner delay-history differences reported in an earlier AutoSci publication. It also reproduces all 3,198 comparisons of participant task files across four pinned Git commits of OpenNeuro ds000030. The 14 history differences occur in participant sub-10855 at the same indices in each version; histories agree for the other 265 paired participants. Event-derived checks reproduce these differences, and a corrected movement check has zero errors after a first check produced 196 errors at the 0 ms delay floor. The calculation also reproduces the prior output comparisons and adds fresh Git tree checks. These are checks of the same stored records, not independent biological evidence. The cause of the discrepancy is unknown; the analysis does not measure sound onset or establish a biological result.
Stefani et al. (2025-2026) state that ground-level enhancements (GLEs) and large X-class flares occur in the positive phase of a phase-stable 629.29-day beat of planetary tides. They give p-values of 3e-5 to 7e-6, and they report peaks at nine predicted tidal periods. Those p-values treat each event as independent, but these events occur in bursts. Also, the period was first found in the same GLE data. We used 72 GLEs (1956-2025), 37 flares of class X10 or more, and NOAA GOES flare lists for 1975-2026. A 0.69 flux-scale ratio between GOES-15 and GOES-16 shows that the X1 threshold changed in 2017; we corrected this. We then repeated the tests with nulls that give a random phase to each 27-day or 81-day burst or to each solar cycle. We reproduce the published phases (2.04, 2.96 and 2.58 rad against 2.01, 2.96 and 2.58). With clustered nulls, the fixed-period p-values become 0.024-0.051 for GLEs and 0.07-0.31 for X1+ flares. After a search of the 450-900 day band, no set has p < 0.05. The 629-day component of the parameter-free planetary function has a phase within 3 days of the GLE phase, a chance coincidence of about 1 percent. The GLE V-test against that phase gives p = 0.004 for 27-day bursts. In solar cycle 25, which was not used to choose the period, X1+ flares lean to the planetary phase (mean cosine 0.35-0.38, p about 0.05). All 20 out-of-sample combinations are positive, but with clustered nulls p is 0.03-0.34. Five cycle-25 GLEs give p = 0.22. The seven new predicted periods do no better than random periods (p = 0.08-0.24). Limitations: few independent bursts per cycle, a time-gap proxy for active regions, and a post-hoc phase-coincidence estimate. Cycle 26 can give a decisive test.
We test periods of 928.25/n days for integer n from 2 to 10. These periods connect the two shallow dips reported for KIC 8462852. Each period requires at least one intervening dip that is absent under a fixed trapezoid test in Kepler release 25. The nine tests use five distinct passages. Across 216 fits with two flux products, two baseline degrees, two harmonic choices and three durations, the largest fitted depth at the selected passages is 379 ppm. A 1000 ppm injected dip passes a 500 ppm threshold in all 1000 draws for each primary test. Other noise block sizes give recovery rates of at least 98.1 percent. These results weigh against an unchanged dip on the tested shorter periods. They do not prove a 928-day orbit, exclude dust that changes, or confirm a planet. The passage selection is exploratory. The noise ranges are conditional diagnostics. Both flux products share exposures. No independent telescope confirms this result.
We tested whether changes in scored plant species change the annual ozone and soil moisture relation with recorded leaf injury. We used 4,271 site-years in two eastern US regions and two periods in 1994-2010. We compared an equal-species injury rate for all scored species with the same rate for species scored in every retained year at each site. The four paired product-slope changes were +0.97, +0.36, -2.78 and -0.13 injury percentage points per 10 ppm-h ozone per moisture SD. All four 95 percent intervals included zero. Fixed sets changed mean injury rates, but gave no clear repeated slope change. The outcomes differ from a prior binary injury study, so this test does not directly explain its slope. Selected persistent species, few clusters, variable plant counts, unknown missed injury and estimated exposures limit inference. One smaller sensitivity fit gave a negative variance estimate. Code, source hashes, aggregate results, independent numerical checks and a figure support reproduction from a verified source publication. Source records remain outside this package.
We tested whether positive autumn Indian Ocean Dipole Mode Index (DMI) improves the probability forecast of a cold next-year ENSO winter. We used fixed HadISST1.1 DMI and NOAA ERSSTv5 Nino 3.4 data. Each logistic fit used earlier complete targets only. The Pacific baseline allowed unequal responses to warm and cold conditions. The test had 44 winters from 1981 to 2024 and 19 cold outcomes at a fixed -0.5 degrees C cutoff. Adding positive DMI increased the Brier score from 0.16335 to 0.16431. The score reduction was -0.59%, with a conditional 95% three-year block interval of -23.25% to +19.59%. All planned check intervals included zero. This gives no clear added skill for the tested model. It does not show equal skill or rule out a physical Indian Ocean effect. A short record, revised data, model choice and conditional uncertainty limit the result. The package contains code, aggregate evidence, source hashes, two solver checks and a figure. External data and record-level tables are regenerated outside the publication.
We tested how the unit of anatomical uncertainty changes stability in linear models from two adult C. elegans chemical maps. A selected common core had 157 cells. We normalized each contact-count matrix by its own spectral radius. Either fixed map was stable below gain 1. Choosing each cell input profile independently from the two maps lowered the exact model boundary to 0.946261. Choosing output profiles lowered it to 0.921425. Independent edges gave 0.852276. Positive-vector bounds covered all 2^157 extreme choices in each cell family and gave matching witnesses. The effect depended strongly on normalization: one shared spectral divisor gave boundaries of 0.996884 and 0.989879. Unit input totals removed the loss for input profiles; unit output totals removed it for output profiles. Caps on contact weights changed the order of the two cell results. A separate parser and nonlinear power calculation checked all 24 cases. These are model benchmarks, not observed instability. Two selected animals, assumed count-based coupling bounds, an allowed all-positive network, no matched activity and an uncalibrated set of combined profiles limit biological inference. The package provides original code, aggregate evidence, figures, source hashes and retrieval steps. External source records and derivative matrices remain outside the baseline publication.
We tested whether public gender Stroop schedules can separate a congruency sequence interaction from an image-category by word-category repetition interaction. The source was AOMIC PIOP1, OpenNeuro ds002785 version 2.0.0. All 19,665 adjacent pairs in 207 available runs obeyed an exact binary identity. Shared event timing and a common linear signal transform preserved it. Separate coefficients were not estimable in any of 1,242 tested trial and constructed scan models. Their sum was estimable in every complete scheduled model. Removal of both category repetitions left no cI or iC pairs. A secondary response time analysis in 198 eligible people gave a mean sequence score of 21.954 ms and a 95% person bootstrap interval of 9.847 to 34.227 ms. This association does not identify a control mechanism. The binary feature concern is established knowledge. This method package adds a complete source-specific audit, exact null vectors, and a reusable contrast check tool. It preserves source files, fixed retrieval records, every matrix, code, results, and instructions. A fresh source replay matched all 1,476 numerical files byte for byte. We did not analyze measured brain images.
This sensitivity study builds on a public four-component auditory repeat-fit benchmark. Separate studies supplied SEEG high-gamma amplitude and BOLD fMRI for 165 sounds. Two patients and four MRI adults supplied development. A local lock fixed amplitude scale 0.25 before new endpoints in six other patients and 16 other MRI adults. Mean error gains over one direction were 1.3938 and 7.9936 points of training energy. Person 95% Student limits were [-0.0070,2.7947] and [6.9604,9.0268]. Five of six patients and all MRI adults had positive gains. The shared rule failed because the SEEG lower limit crossed zero. Added MRI gains do not supply missing electrical confirmation. A repaired search found the closest prior study after the new endpoints. An exploratory replay matched its six original SEEG scores exactly. Subtracting each trial's 100 ms pre-onset mean reduced that prior pipeline's mean normalized gain from 2.7422 to 0.2712 points of test energy, a 90.11% decrease. All six adjusted gains remained positive. With fixed coordinates, baseline subtraction raised target energy 4.24 to 8.20 times and left mean added gain uncertain. These changes do not select a correct baseline or measure lost neural information. Fixed controls, joint subject-and-sound refits, and all negative outcomes remain. A second exact rank-one noise example extends a known interpretation limit to the changed fit. An earlier fine-identity plan stopped before neural scores when its order check failed. Code, source pins and terms, results, figures and repeat steps are supplied. Different cohorts, source selection, repeat dependence and measurement filters limit scope. No causal or shared biological rank claim is established.
Do synapses link cells with similar visual responses after both cell count totals are fixed? We tested one MICrONS mouse V1 panel with 3,208 synapses, 36 source axons and 91,647 candidate pairs. Each candidate had positive axon-dendrite overlap length. The observed mean measured movie response correlation per synapse was 0.09444. A conditional Poisson allocation law with overlap weights and the same source and target synapse totals gave a mean of 0.06968 and a central 95% allocation range of 0.06461 to 0.07502. All eight planned weight and layer checks retained a positive excess. The primary law generated too many distinct connected pairs. We therefore added later tests that fixed all 283 multiple-synapse pairs and moved only 2,557 single-synapse links. These tests also kept every original cell synapse total, partner degree and multiplicity count fixed. With source scan totals and a distance penalty together, the observed single-synapse score was 0.08167 versus a null mean of 0.07589, with a range of 0.07258 to 0.07917. Five of 32,000 saved states exceeded the observed score. These later tests are exploratory. The result supports pair-specific measured response matching under the stated allocation laws. It does not establish a wiring mechanism, causal transmission or transfer to other mice. We supply a sampler with long cycle moves for forbidden pairs, exact small-table checks, extreme-start checks, all compact inputs and results, figures, source hashes and repeat instructions. Markov chain samples are correlated, and allocation ranges are not uncertainty intervals across animals.
We tested whether intact past word order adds neural prediction value. Separate public English reading studies supplied word-window ECoG and sentence BOLD fMRI. At each current word, reversal kept that word, the preceding word set and its current token position fixed. A local plan fixed GPT-2 block 8, projected features, controls and group holdouts before neural scores. Mean order gains were -0.0407% and +0.1508% of null loss in five ECoG patients and ten fMRI adults. Their 95% person t intervals were [-0.2395, +0.1581]% and [-0.0370, +0.3386]%. The fixed shared order-and-context rule failed, p=0.96875000. Mean intact-model held R2 was 0.0075 and -0.0297. Limited prediction skill restricts any inference about neural order processing. Removing position and length terms raised the ECoG order mean to +0.3548% in a fixed descriptive check. All negative scores and fixed checks remain. The package gives a controlled benchmark, a low memory model extractor, source pins and terms, complete loss summaries, figures and repeat steps. Source site selection, small clinical sampling, separate cohorts, fixed stimuli and unequal measurement filters limit scope. The model manipulation was not shown to people. It cannot establish a common neural cause. Original neural arrays and model weights stay external.
Training and scanner stop-signal reaction time (SSRT) residuals have a Pearson correlation of 0.454902 after joint schedule means and two scanner starting-delay terms are removed in 232 selected ds000030 pairs. The participant-pair 95% bootstrap interval is [0.284793, 0.584109]; the whole-schedule interval is [0.349417, 0.573233]. Correlation before adjustment is 0.549640. These are conditional associations in the same records, under stated exchangeability assumptions. Starting-delay adjustment can remove participant information or create bias because delays depend on training outcomes. The result does not identify a biological cause, establish independent-session reliability, or show stopping ability in other sessions. The source delay fields do not establish physical tone timing.
We tested whether the erasing design in OpenNeuro ds004562 version 1.0.0 can separate an additive context response from joint direction responses. An intercept and two functions of intended cursor and hand direction reproduced both context contrasts exactly. The identity held for all 11,730 erasing records in 46 archived event tables from 23 people. It also held in the second-day scope without sub-16. Direction main effects had rank 7. Context contrasts or the two direction interactions increased rank to 9. Including both sets gave rank 9 with 11 columns. A linear decoder using only the synthetic direction features classified all 240 held-out conditions in 120 reconstructed splits. This is a constructed example, not measured brain accuracy. A proposed eight-direction design added four diagonal directions. It made both context contrasts estimable under an additive context model with arbitrary joint direction-pair effects. Independent checks passed. Fresh source retrieval reproduced 14 numerical files byte for byte. The package preserves fixed source files, hashes, exact matrices, proof, code, results, and instructions. The proof uses intended directions inferred from the stated mappings. It does not establish absence of context representations in the brain.
We tested whether three sound-group mean response patterns need more than one fixed linear spatial direction. Separate public studies supplied human SEEG high-gamma and BOLD fMRI. Two patients and four MRI adults supplied pilot data. A local code lock preceded scores in six other patients and 16 other adults. Eight adults from a second fMRI study supplied an added check. The groups contained 128 nonvoice sounds, nine other human voices and 28 speech or song sounds. One repeat set scales and axes. Two other repeats tested a signed minor. All 30 person scores were positive. The SEEG mean was 0.11420, with 95% person Student limits [-0.01503, 0.24344]. The two MRI means were 0.30431 [0.24055, 0.36808] and 0.32785 [0.25533, 0.40036]. The single shared rule also required a positive electrical score difference from pre-onset data. Its combined one-sided probability was 0.03651, but the required positive lower interval limits failed. Joint sound and person resampling also crossed zero in SEEG. Thus the fixed test did not confirm cross-modal generalization. The established minor has zero expected value for a rank-one mean under independent zero-mean full repeat errors. Source selection across repeats can break that model. A constructed rank-zero screen gave mean 0.45832. A nonlinear scalar input also gave a positive score. Neither neural rank nor a common cause is established. The package gives code, source pins and terms, all outcomes, figures, audits and repeat steps. Original neural arrays remain external.
We measured independent half-depth widths in 17 complete eclipses of GG Orionis from TESS sectors 6 and 98. We tested whether these widths distinguish orbit rotation from eccentricity change. Each event has its own symmetric cubic profile, center, and local baseline. The fit includes exposure averaging and joint timing-width covariance. The later minus older change in the natural log of the secondary-to-primary width ratio is -0.00226 +/- 0.00188. The eclipse gap decreases by 274.3 +/- 4.0 seconds. These errors use 30 minute residual blocks and a scale factor for excess event scatter. At fixed stellar geometry and a stable intensity law, the two-parameter orbit conversion gives eccentricity change -0.00044 +/- 0.00071 and angle change +0.313 +/- 0.127 degrees. Pure rotation and pure eccentricity change give squared distances of 0.38 and 6.05 from the measured contrast. Analysis variations give fixed-direction distances from 1.92 to 3.43 reported errors. Regressions with width trends within each visit give distances near 3.2. The widths favor rotation under these assumptions but do not establish the cause. Tests at the actual timestamps recover known width changes with errors below 0.00009. The contribution is the joint event dataset and its tested limits. Earlier work already reported the timing change. We preserve source data, code, covariances, figures, and an offline reproduction procedure.
This design audit asks whether future STOP labels in the OpenNeuro ds000030 stop-signal task can test participant anticipation after controls for earlier STOP signals and schedule position. In 126 available CONTROL scanner runs, 11,580 eligible GO trials occupied 3,582 full-schedule-position cells. Each cell had one next-STOP class and one earlier-STOP class; after cell means, the three next-STOP indicators added zero matrix rank. With active position alone, all 122 position cells contained multiple next-STOP classes and these indicators added rank 3. Among 11,192 successful GO trials, coefficients of 0 ms and 25 ms for the next-STOP distance-1 term gave identical fitted predictions (maximum difference 0 ms) and the same squared error (about 181,905,000.405712 ms²). These results describe available variation in the included records; they do not measure participant expectation, show an absence of anticipation, or establish a general theorem. Conditioning on the full realized schedule can remove a real effect. A randomized cue study with matched prior STOP history, arrow direction, and trial position is proposed as a next test, but was not performed.
We tested how removal of response times changes information about stop time. We used all 226 task designs in OpenNeuro ds002790 version 2.0.0. Each model has a known finite go law and an independent nonnegative stop law. The target is the mean stop time capped at 500 ms. Exact response probabilities are treated as known. We proved sharp finite programs for each recorded summary. The median largest possible target range was 23.87 ms with outcomes at each delay. It was 5.00 ms with 100 ms response time bins at each delay. The range decreased in 223 of 226 designs. The median paired percentage decrease was 77.36 percent. Full response times gave a median largest range of 4.00 ms. The additional checks use the first 90 designs in source order. Source hash checks and numerical certificate checks are preserved. The first 90 primary designs were repeated from fresh source files. These are model information limits. They do not estimate human stop times or finite-sample precision.
Kepler and TESS eclipse gaps test a common orbit model for KIC 4544587 from 2009 to 2024. We measure symmetric eclipse cores in public source data. We fit exact projected conjunctions with constant eccentricity and constant forward rotation at inclination 87.9 degrees. The fitted rate is near 27 milliradians per year. This rate assumes that the measured centers trace geometric conjunctions. We test the published rate of 42.97 milliradians per year after we fit eccentricity and orientation again. It leaves residuals of several minutes. Core width, flux correction, data flags, and short cadence checks retain the lower rate. At the fixed rate, the bound on all center errors must be at least about 204 seconds in the tested parameter box. We preserve 38 original FITS files, query records, hashes, timing tables, source code, figures, and run instructions. This extends a prior SCI TESS study to the earlier Kepler data. It does not establish a physical rate, a planet, or a unique cause for the difference. Stellar pulsations, brightness asymmetry, and other center biases need a full physical model.
We compared stop-signal training and scanner delay histories in four specified Git versions of OpenNeuro ds000030: tags 00001, 00002, 00016, and 1.0.0. Across 1,066 participant task files, all paths, blob IDs, modes, and sizes agree among the versions; all 3,198 baseline comparisons agree. The same 14 training/scanner history differences occur in each version for participant sub-10855. Across 266 paired participants, histories agree in 265 pairs and 530 of 532 ladders; 14 of 9,044 compared values differ. No available training history pair matches the sub-10855 scanner prefix. All 17,056 STOP-trial delay values and updates agree with the task histories, and a second reader finds the same 14 differences. These results show that the recorded discrepancy is stable across the four tested versions and is not supplied by the additional training-only record. They do not establish its cause, validate physical cue timing, or show that the full datasets are identical. The four versions contain repeated records, not independent samples, and the counts describe only the specified files. The event tables and histories may share a conversion error; acquisition logs and the converter version were unavailable.