Evidence · tedana

t2smap

t2smap is tedana’s fit of the T2* and S0 maps of multi-echo fMRI and its optimal combination of the echoes. fMRIPrep runs it for every multi-echo BOLD run and carries on with the combined series. The replica (larmorx-gpl) gives tedana 26.0.3’s files, messages and exit status bit for bit in all 62 of its compared cases, out of 102, in both kept library environments and at any thread count; fMRIPrep’s four calls on three fMRIPrep runs are among them. The clean room (larmorx, from the next release) holds tedana’s results within its specification’s thresholds in 88 of its 89 compared cases, out of 125, fMRIPrep’s four calls among them, the same files at any thread count; in the other, a chaotic curve fit, it does not print SciPy’s own warnings. On fMRIPrep’s curve-fit calls the clean room runs 10 to 15 times as fast as tedana on one thread and 19 to 30 times on eight, with a quarter of tedana’s peak memory. Both correct tedana’s defects by default or refuse them with a clear error, each with tedana’s result on request.

Ports
  • replica larmorx-gpl GPL-3.0: crates larmorx-gpl-tedana and larmorx-gpl-cli, translated from tedana 26.0.3's source with the arithmetic of numpy 2.5.4 (OpenBLAS 0.3.34), SciPy 1.18.1, nibabel 5.4.2 and nilearn 0.14.1, whose permissive pieces live in larmorx itself; it runs as a separate program
  • clean room larmorx Apache-2.0: crate larmorx-tedana, written from specs/t2smap.md, tedana's documentation, the papers it names and runs of tedana 26.0.3 and of the replica, without tedana's source (2026-11-11 and 12); from the next release
Status
replica experimental; clean room experimental
Records
the replica's: standard comparison (tedana 26.0.3) · the clean room's: full comparison (tedana 26.0.3) · findings: the replica's tedana-t2smap, the clean room's tedana-t2smap-clean-room · benchmark tedana-t2smap
Reference
t2smap · API page tedana-t2smap

What is compared

A multi-echo BOLD run records every volume at several echo times. In each voxel the signal decays with the echo time TE as S(TE) = S0 · exp(−TE / T2*): S0 is the signal at TE = 0 and T2* the decay’s time constant, which the BOLD effect changes. tedana (TE-Dependent ANAlysis, the ME-ICA developers; DuPre et al., 2021, JOSS 6:3669) estimates both in every voxel and combines the echoes into one series that keeps the BOLD contrast of every echo.

Its t2smap workflow works in four steps:

  • it counts, in each voxel of the mask, the echoes whose signal is good: the adaptive mask. An echo whose mean signal drops below a third of that of the voxels the 33rd percentile marks is not good;
  • it fits the decay through those echoes, at least two. The fit is log-linear (a least-squares line through log(|S| + 1) against TE, for each count of echoes all voxels at once) or, with --fittype curvefit, a monoexponential least-squares fit started from the log-linear one, its S0 bounded below by the voxel’s lowest signal;
  • it floors and caps the maps (the limited maps leave out the voxels with one good echo) and measures the fit’s error per voxel and per volume;
  • it weighs every echo by TE · exp(−TE / T2*) with the voxel’s T2* (Posse et al., 1999), or by its temporal signal-to-noise ratio (--combmode paid; Poser et al., 2006).

fMRIPrep 25.2.6 runs it for every multi-echo BOLD run (init_bold_t2s_wf, its T2SMap interface): t2smap -d <the echoes, resampled onto the BOLD reference> -e <the echo times in ms> --mask <the BOLD mask dilated by two voxels> --fittype curvefit (--me-t2s-fit-method, curvefit by default), with one thread. It keeps the full T2* map and the optimally combined series.

The two ports run beside tedana 26.0.3’s t2smap on the same files with the same arguments. fMRIPrep’s own files are the ones it wrote in its runs, with tedana 25.0.1 in fMRIPrep 25.2.6’s image.

bash
larmorx --implementation clean-room tedana t2smap -d echo-1.nii.gz echo-2.nii.gz echo-3.nii.gz \
    -e 14.0 29.96 45.92 --mask mask.nii.gz --fittype curvefit --out-dir t2smap
python
import larmorx as lx

res = lx.tedana.t2smap(["echo-1.nii.gz", "echo-2.nii.gz", "echo-3.nii.gz"],
                       [0.014, 0.02996, 0.04592], mask="mask.nii.gz", fittype="curvefit",
                       implementation="clean-room")
res.t2star, res.s0, res.optcom, res.adaptive_mask, res.fit_failures, res.confounds

Both choose the clean room, which comes with the next release; in 0.1.0rc1 t2smap is the replica only, in larmorx-gpl. The API page describes every option. How to read what follows: how to read the evidence.

Its two ports

tedana comes under the GNU LGPL version 2.1, so LARMORX keeps t2smap in two ports (licensing).

The replica lives in its own package, larmorx-gpl (pip install larmorx-gpl, or pip install "larmorx[all]"), under the GNU GPL version 3, which the LGPL’s section 3 allows. It runs as its own program.

  • The arithmetic it relies on is permissive and lives in LARMORX itself: numpy’s least squares (LAPACK 3.12’s dgelsd over each group of voxels at once, with OpenBLAS’s Haswell kernels), SciPy’s bounded trust-region fit (least_squares’ TRF method, its finite-difference Jacobian and its SVD), numpy’s float32 log (numpy’s own kernel), its sums, means and percentiles, glibc 2.35’s float64 exp and log (what numpy calls on CPUs without AVX-512), nilearn’s mask when none is given, and nibabel’s and nilearn’s writers (the quaternion in x87’s extended precision included).
  • It divides the voxels between threads in fixed pieces, so --n-threads never changes a bit.
  • In its default, fix, it names the maths entries NP-03 and GLIBC-07 exactly where the correctly rounded functions change a written value. To do so it runs the stages from the mask to the combination again with the other choice of functions, writing nothing: up to four passes, so the replica is as fast as tedana at one thread.

The clean room is part of LARMORX itself (pip install larmorx, Apache-2.0) and runs in the same process. Its implementers worked from the specification, tedana’s documentation, the papers it names and runs of tedana and of the replica as black boxes, never from tedana’s source or the replica’s code. It is built for speed:

  • the log-linear fit in closed form, voxel by voxel;
  • its own trust-region curve fit on the echoes’ means, where each evaluation costs as many operations as there are echoes, not echoes times volumes;
  • the work divided between threads in fixed pieces, so --n-threads never changes a bit.

In reproduce it gives tedana’s defects exactly and its fitted numbers within the specification’s thresholds (its §17.3), not bit for bit. The exception is the voxels whose observations are all equal, where tedana’s values are its least squares’ rounding noise (TED-46): there it computes them with LARMORX’s exact ports of numpy’s lstsq and SciPy’s curve_fit.

From the next release, implementation="auto" | "replica" | "clean-room" chooses between them, and larmorx --implementation clean-room tedana t2smap ... does the same on the command line. auto, the default, and larmorx tedana t2smap take the replica when larmorx-gpl is installed, else the clean room.

The verdicts at a glance

The replica is a translation of tedana’s source, so the question for it is whether it gives tedana’s bytes: every image (header and data), the confounds table, the JSON files, the messages and the exit status.

Agreement, replica against the original: tedana t2smap 26.0.3 (standard comparison, 102 cases of the record), by category and verdict
fMRIPrep's calls5
synthetic inputs32
failures and refusals24
file formats20
tedana's own variability3
random testing18

Select a segment to list its cases below.

The replica’s record runs tedana 26.0.3 as its program in both kept library environments, with numpy’s OpenBLAS on its Haswell kernels, numpy’s AVX-512 kernels off and one BLAS thread:

  • the newest releases (numpy 2.5.4 with OpenBLAS 0.3.34, SciPy 1.18.1), LARMORX’s reference;
  • the shared ones (numpy 2.3.5 with OpenBLAS 0.3.30, SciPy 1.15.3).

It compares the replica with each in reproduce (upstream_defects="reproduce"), with that environment’s libraries. Of its 102 cases:

  • Bit-identical, 62. Every file, message and exit status the same in both environments.
  • Rejected by both, 40. tedana refuses the input, and the replica does the same, with the same messages and exit status.

None differs. The cases cover:

  • fMRIPrep’s four calls on three fMRIPrep runs (ds003592’s also with --fittype loglin; ds000210’s and ds000258’s from the fMRIPrep runs made with OpenBLAS’s Haswell kernels);
  • every option: masks given and computed, every mask type, both fit types and fit modes, both combinations, dummy scans and excluded volumes, both naming conventions, prefixes, --verbose, runs made twice with and without --overwrite;
  • every data type and file form: float32, float64, scaled int16, uint16 and uint8, big-endian, .nii, .nii.gz, pairs, NIfTI-2, z-concatenated echoes, and 20 cases of ANALYZE 7.5 and MGH/MGZ;
  • NaN, infinite, zero and negative samples, one volume and nine echoes;
  • every failure case, with tedana’s chained tracebacks;
  • tedana’s own variability, and 18 cases from random testing.

Each case also runs the replica’s default, fix, where LARMORX corrects tedana’s deterministic defects and names them (below).

The clean room is written from the specification, not translated from tedana’s source, so the question for it is whether it behaves as tedana does within the specification’s thresholds (§17.3, by its check.py):

  • its exit statuses, files, headers, data types, messages and the defects it names are tedana’s in reproduce (A) and the replica’s default’s in fix (B);
  • its fitted numbers are judged value by value; where a curve fit is ill-conditioned (a flat objective or a start on the feasibility boundary), by the fit’s own objective, with at most 3 % of a call’s fit voxels judged that way.
Agreement, clean room against the original: tedana t2smap 26.0.3 (full comparison, 125 cases of the record), by category and verdict
the specification's probe117
thread counts2
fMRIPrep's calls6

Select a segment to list its cases below.

The clean room’s record runs 125 cases of its full comparison against tedana 26.0.3 (through the cache of the original’s runs), and each case’s fix against the fix of a replica built from the same branch:

  • Bit-identical, 65. tedana’s values bit for bit.
  • Within the thresholds, 23. tedana’s values within §17.3’s thresholds.
  • Rejected by both, 36. Both programs refuse the input with the same messages and exit status. The record counts these 36 among its 124 passes, since both programs’ refusals are compared.
  • Expected divergence, 1. A chaotic curve fit, inside whose steps SciPy prints its own floating-point warnings; the clean room does not print them, since its solver is its own.

None fails. The cases cover the specification’s black-box probe and more:

  • every option: masks given and computed, both fit types and modes, both combinations, dummy scans and excluded volumes, both naming conventions, prefixes, --verbose;
  • every data type and file form: NIfTI-1 and -2, pairs, z-concatenated echoes, ANALYZE 7.5, MGH/MGZ;
  • NaN, infinite, zero and negative samples, and the saturated voxels of TED-46 in both fits and both modes;
  • every failure case, and thread invariance at 1, 4 and 8 threads;
  • fMRIPrep’s four calls on three fMRIPrep runs (ds003592’s also with --fittype loglin and with --verbose; ds000210’s two runs and ds000258’s from the fMRIPrep runs made with OpenBLAS’s Haswell kernels).

In fMRIPrep: the T2* and S0 maps

fMRIPrep’s own call on its run of OpenNeuro ds003592 sub-280 (three echoes, 200 volumes) was given to tedana 26.0.3, to both ports and to tedana itself in other settings: numpy’s AVX-512 kernels on (this CPU’s default), the shared environment’s releases, four threads, and tedana 25.0.1 in fMRIPrep’s image, with numpy’s AVX-512 kernels off and on. The replica ran in reproduce at 1 and 14 threads and in fix; the clean room in reproduce at one thread and in fix at eight and at one. fMRIPrep keeps the full T2* map, so its own file is compared with that map.

The input panels are the first volume of each echo as fMRIPrep gives it to t2smap, and fMRIPrep’s mask over the first echo. The outputs, chosen above the viewer, are the full and the limited maps, the adaptive mask and the curve fit’s failures. The viewer opens on the full T2* map, where the default’s correction TED-05 is largest. tedana 26.0.x caps only the limited map (values more than ten times its 99.5th percentile are set to it), so a few voxels of the full map, the map fMRIPrep keeps and the combination uses, hold values of a hundred thousand seconds. tedana 25.0.1, fMRIPrep’s, capped the full map, and LARMORX’s default does.

Worked example

fMRIPrep's T2* and S0 maps of ds003592's multi-echo run

What this shows: In reproduce the replica gives every value of tedana 26.0.3’s maps, at 1 and 14 threads alike, and the clean room holds them within the thresholds of specs/t2smap.md §17.3: 11,381,033 of the 11,406,278 values of the run’s files are identical, and 0.305% of the curve fit’s voxels are judged by the fit’s own objective (its files are the same at 1 and 8 threads). In fix, their default, both correct TED-05 and TED-08 and the replica also the maths entries NP-03 and GLIBC-07; the two defaults agree by the same thresholds (11,380,746 of 11,406,278 values identical, 0.259%). fMRIPrep’s own T2* map, made by tedana 25.0.1, differs from the replica’s reproduce in 2,531 of its 176,128 values, tedana’s own step between its versions; run here in fMRIPrep’s image with numpy’s AVX-512 kernels off, as in fMRIPrep’s run, tedana 25.0.1 gives it bit for bit.

fMRIPrep 25.2.6’s t2smap call on OpenNeuro ds003592 sub-280’s rest run (3 echoes (TE 14, 29.96, 45.92 ms) of 64 × 64 × 43 voxels of 3.38 × 3.38 × 3 mm, 200 volumes (TR 3 s), float32): tedana’s adaptive mask (how many echoes have good signal in each voxel), the monoexponential fit of S0 and T2* in every voxel (log-linear first, then SciPy’s bounded curve fit from it, --fittype curvefit as fMRIPrep calls it), the full and the limited maps and the fit’s failures (the RMSE map and the confounds table are in the table of every file below). tedana 26.0.3 and the replica compute every map; fMRIPrep’s own T2* map (tedana 25.0.1, the only map fMRIPrep keeps) and further runs of tedana are compared with them value by value. LARMORX’s clean room computes every map too, in reproduce at one thread and in fix, its default, at eight threads and at one; its comparisons with tedana 26.0.3 and with the replica’s default are judged by specs/t2smap.md §17.3’s thresholds.

Data
OpenNeuro ds003592 (Neurocognitive aging data release), sub-280, task rest (CC0-1.0): the echoes as fMRIPrep’s node was given them; shown inside fMRIPrep’s mask dilated by one voxel, so no face shows
Node
fMRIPrep 25.2.6, ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/t2smap_node (bold_t2smap_wf, its T2SMap interface)
Command
t2smap -d ‘/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-280_ses-1_task-rest_echo-1_bold_valid_tshift_xformresampled.nii.gz’ ‘/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-280_ses-1_task-rest_echo-2_bold_valid_tshift_xformresampled.nii.gz’ '/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold…
Maps
tedana’s full T2* map ranges from 0 to 2.398e+05 s; its adaptive mask keeps 52,826 voxels; the curve fit failed in 1,549
Programs
tedana 26.0.3 as its program (t2smap) in both kept environments, tedana 25.0.1 in fMRIPrep 25.2.6’s image; LARMORX’s replica, the larmorx-gpl program, at commit 48278c64fe24; numpy’s OpenBLAS on its Haswell kernels, one thread
Clean room
LARMORX’s clean room (crate larmorx-tedana, experimental) through its binding larmorx._core.tedana_t2smap_cleanroom, as its own example runs it, at commit 55dace9da91f; these are the runs of its own example below, the same call
Grid
64 × 64 × 43 voxels of 3.38 × 3.38 × 3 mm; shown: 53 × 59 × 43 around the brain
Inputds003592: echo 1 (14 ms)volume 1 of 200, as fMRIPrep gives it to t2smap (the echo resampled onto the BOLD reference)
Originaltedana t2smap 26.0.3numpy 2.5.4, SciPy 1.18.1, AVX-512 kernels off, one thread: LARMORX's reference
Replicareplicalarmorx-gpl, GPL-3.0 · 1 thread, reproduce · against tedana t2smap 26.0.3
Replicareplicalarmorx-gpl, GPL-3.0 · 14 threads, fix (the default)
Clean roomclean roomlarmorx, Apache-2.0 · 1 thread, reproduce · against tedana t2smap 26.0.3
Clean roomclean roomlarmorx, Apache-2.0 · 8 threads, fix (the default)
the original against itself: numpy's AVX-512 kernels on and offtedana t2smap: numpy's AVX-512 kernels on − offscale ±2.4e5

Bit-identical176,128 of 176,128 voxels equal

the replica against the originalreplica − tedana t2smap 26.0.3
the replica against itself: fix and reproducereplica: fix − reproducescale ±2.4e5
the clean room against the originalclean room (reproduce) − tedana t2smap 26.0.3scale ±2.4e5

Axial slice 37 of 43; the crosshair where the default's T2* map differs most from reproduce's; choose the output above the images: the full T2* map (fMRIPrep's output) and S0, the limited maps, the adaptive mask (good echoes per voxel), the curve fit's failures. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The original against itself: the yardstick
tedana t2smap: numpy's AVX-512 kernels on − offreferenceA reference row (the original against itself): numpy’s float64 exp and log are Intel’s SVML kernels on CPUs with AVX-512 (NP-04), glibc 2.35’s otherwise; LARMORX’s reference is the path without them (decision D2): 4,555 of 1,056,768 values differ. NP-04T2* (s)8.12e41,675 of 176,128 (0.951 %)3580.3390.0621
S00.00711604 of 176,128 (0.343 %)1.76e−54.3e−75.53e−16
T2*, limited (s)0.002161,676 of 176,128 (0.952 %)6.76e−60.0001549.31e−10
S0, limited0.00711600 of 176,128 (0.341 %)1.76e−54.65e−75.6e−16
Adaptive mask00 of 176,128000
Fit failures00 of 176,128000
tedana t2smap: numpy 2.3.5 / SciPy 1.15.3 − numpy 2.5.4 / SciPy 1.18.1referenceA reference row (the original against itself): SciPy 1.15.3 → 1.18.1 changed least_squares’ trust-region reflective method and its finite differences (the order of their steps, the SVD factors’ memory order): the curve fit’s last bits, which the full T2* map magnifies where T2* is large: 5,909 of 1,056,768 values differ.T2* (s)8.65e42,182 of 176,128 (1.24 %)3280.3610.0512
S00.000244776 of 176,128 (0.441 %)4.33e−61.48e−83.33e−17
T2*, limited (s)0.0007152,183 of 176,128 (1.24 %)2.34e−65.09e−51.16e−10
S0, limited0.000244768 of 176,128 (0.436 %)4.28e−61.6e−83.3e−17
Adaptive mask00 of 176,128000
Fit failures00 of 176,128000
tedana t2smap: --n-threads 4 − 1referenceA reference row (the original against itself): tedana’s --n-threads 4 runs the curve fits in four joblib processes and lets OpenBLAS use four threads: every one of the 1,056,768 values is the same.every outputidentical: 0 of 176,128 values differ
tedana t2smap 25.0.1 (fMRIPrep's image, here) − 26.0.3referenceA reference row (the original against itself): tedana’s own step from 25.0.1 (fMRIPrep 25.2.6’s) to 26.0.3: tedana 26.0.x caps only the limited T2* map (TED-05; 25.0.1 capped the full map, as LARMORX’s default does), loads the data as float32 (25.0.1: float64), applies the T2* floors to the full map and no longer clamps negative values; and fMRIPrep’s image has other libraries (numpy 2.2.6, which sums more than 8,192 elements in chunks, SciPy 1.15.2, conda-forge’s OpenBLAS): 7,357 of 704,512 values differ.T2* (s)2.4e52,531 of 176,128 (1.44 %)99710.966
S00.001041,160 of 176,128 (0.659 %)5.8e−66.28e−85.98e−17
T2*, limited (s)13.62,514 of 176,128 (1.43 %)0.1090.9670.293
S0, limited0.001041,152 of 176,128 (0.654 %)5.8e−66.79e−86.06e−17
tedana t2smap 25.0.1 here, numpy's AVX-512 kernels on (as in fMRIPrep's run) − fMRIPrep's outputreferenceA reference row (the original against itself): tedana 25.0.1 in fMRIPrep’s image, here with numpy’s AVX-512 kernels on (this CPU’s default: fMRIPrep’s runs set only OpenBLAS’s Haswell kernels), against fMRIPrep’s own run of it, made on the project’s earlier main machine, an Intel Core i7-8750H (AVX2, no AVX-512) (12 logical CPUs), where numpy computes float64 exp and log with glibc’s functions: 1,620 of 176,128 values differ.T2* (s)7.32e−51,620 of 176,128 (0.92 %)2.97e−71.64e−54.3e−12
tedana t2smap 25.0.1 here, numpy's AVX-512 kernels off − fMRIPrep's output (25.0.1 in its run)referenceA reference row (the original against itself): tedana 25.0.1 in fMRIPrep’s image, here with numpy’s AVX-512 kernels off (LARMORX’s reference setting: glibc 2.35’s float64 exp and log, NP-04), against fMRIPrep’s own run of it, made on the project’s earlier main machine, an Intel Core i7-8750H (AVX2, no AVX-512) (12 logical CPUs), where numpy computes float64 exp and log with glibc’s functions: every one of the 176,128 values is the same. NP-04T2* (s)identical: 0 of 176,128 values differ
The replica against the original and fMRIPrep's output
replica − tedana t2smap 26.0.3Bit-identicalEvery value of the maps shown is tedana 26.0.3’s (1,056,768 values): the adaptive mask, the log-linear fit (numpy’s lstsq: LAPACK’s dgelsd over each group of voxels at once, with OpenBLAS’s Haswell kernels), SciPy’s bounded trust-region curve fit voxel by voxel, the floors and the cap, written as nibabel and nilearn write them.every outputidentical: 0 of 176,128 values differ
replica − fMRIPrep 25.2.6's own file (tedana t2smap 25.0.1)Within the original's own variabilityThe replica is tedana 26.0.3 bit for bit, so this is tedana’s own step from 25.0.1 (fMRIPrep’s) to 26.0.3 (2,531 of 176,128 values): tedana 26.0.x caps only the limited T2* map (TED-05; 25.0.1 capped the full map, as the replica’s default, fix, does), loads the data as float32 (25.0.1: float64), applies the T2* floors to the full map and no longer clamps negative values; and fMRIPrep’s image has other libraries (numpy 2.2.6, which sums more than 8,192 elements in chunks, SciPy 1.15.2, conda-forge’s OpenBLAS). On this machine tedana 25.0.1 in fMRIPrep’s image differs from tedana 26.0.3 in 7,357 values.T2* (s)2.4e52,531 of 176,128 (1.44 %)9975.38e40.966
replica, fix − fMRIPrep 25.2.6's own fileDeliberate correctiontedana’s step from 25.0.1 to 26.0.3 (the replica’s reproduce against fMRIPrep’s output) and the default’s corrections (TED-08, TED-05, NP-03, GLIBC-07): 2,591 of 176,128 values differ. The full T2* map’s cap (TED-05) is 25.0.1’s behaviour again. TED-08 TED-05 NP-03 GLIBC-07T2* (s)13.62,591 of 176,128 (1.47 %)0.1093.040.292
The replica against itself: thread counts, fix and reproduce
replica: 14 threads − 1 threadBit-identicalThe same bytes at 1 and 14 threads: the replica divides the voxels between threads in fixed pieces, and the log-linear fit runs over each group at once as tedana’s one lstsq call does.every outputidentical: 0 of 176,128 values differ
replica: fix − reproduceDeliberate correctionThe default corrects TED-08, TED-05, NP-03, GLIBC-07 and names each where it changes a written value: 7,561 of 1,056,768 values change. TED-05: 34 voxels of the full T2* map more than ten times the cap, which tedana 26.0.x leaves uncapped (25.0.1 capped them), get the limited map’s cap. NP-03: the log-linear fit’s float32 log correctly rounded (numpy’s own kernel is not); GLIBC-07: float64 exp, log and the curve fit’s powers correctly rounded (glibc 2.35’s are not): the curve fit starts from other last bits and ends within its tolerances elsewhere. upstream_defects='reproduce' gives tedana’s outputs. TED-08 TED-05 NP-03 GLIBC-07T2* (s)2.4e52,635 of 176,128 (1.5 %)99710.929
S00.007111,151 of 176,128 (0.654 %)1.78e−54.3e−75.6e−16
T2*, limited (s)0.002172,620 of 176,128 (1.49 %)6.07e−60.0001547.47e−10
S0, limited0.007111,141 of 176,128 (0.648 %)1.78e−54.65e−75.68e−16
Adaptive mask00 of 176,128000
Fit failures114 of 176,128 (0.00795 %)0.0089210.00453
The clean room against the original, fMRIPrep's output and the replica
clean room (reproduce) − tedana t2smap 26.0.3Within the original's own variabilityspecs/t2smap.md §17.3 by its check.py, over the run’s files: 11,381,033 of 11,406,278 compared values identical; the curve fit’s (b)/(c) share 0.305% of the fit voxels (the clean room’s own trust-region solver, its log-linear fit in closed form: the fitted numbers within the thresholds, not bit for bit) §17.3T2* (s)1.38e52,108 of 176,128 (1.2 %)4720.5770.0497
S00.000488639 of 176,128 (0.363 %)4.34e−62.96e−83.35e−17
T2*, limited (s)0.001822,096 of 176,128 (1.19 %)4.5e−60.000134.13e−10
S0, limited0.000488630 of 176,128 (0.358 %)4.3e−63.19e−83.33e−17
Adaptive mask00 of 176,128000
Fit failures00 of 176,128000
clean room, fix − fMRIPrep 25.2.6's own fileDeliberate correctiontedana’s own step from 25.0.1 to 26.0.3, the default’s corrections (TED-05, TED-08) and the clean room’s own solvers, within §17.3’s thresholds of tedana 26.0.3: 2,127 of 176,128 values differ. The full T2* map’s cap (TED-05) is tedana 25.0.1’s behaviour again. TED-05 TED-08T2* (s)13.62,127 of 176,128 (1.21 %)0.1093.040.292
clean room − replica, both fixWithin the original's own variabilityThe two ports’ defaults by §17.3 (check.py), over the run’s files: 11,380,746 of 11,406,278 compared values identical; the (b)/(c) share 0.259%. §17.3T2* (s)0.0008762,157 of 176,128 (1.22 %)2.63e−66.24e−51.43e−10
S00.00711693 of 176,128 (0.393 %)1.76e−54.3e−75.51e−16
T2*, limited (s)0.0008762,147 of 176,128 (1.22 %)2.63e−66.24e−51.43e−10
S0, limited0.00711687 of 176,128 (0.39 %)1.76e−54.65e−75.58e−16
Adaptive mask00 of 176,128000
Fit failures00 of 176,128000
The clean room against itself: thread counts, fix and reproduce
clean room: fix − reproduceDeliberate correctionThe default’s corrections, each named by its notice: TED-05 (the full T2* map capped as the limited one); TED-08 (the failed fits of voxels with one good echo marked): 521 of 1,056,768 values change. TED-05 caps 34 voxels of the full T2* map. TED-05 TED-08T2* (s)2.81e537 of 176,128 (0.021 %)1.3e310.927
S00.000122234 of 176,128 (0.133 %)1.43e−67.39e−93.61e−18
T2*, limited (s)4.77e−73 of 176,128 (0.0017 %)1.14e−93.4e−82.74e−17
S0, limited0.000122233 of 176,128 (0.132 %)1.43e−67.99e−93.66e−18
Adaptive mask00 of 176,128000
Fit failures114 of 176,128 (0.00795 %)0.0089210.00453
clean room, 1 thread − 8 threadsBit-identicalThe same bits at 1 and 8 threads.every outputidentical: 0 of 176,128 values differ

Δ is A − B over all 176,128 voxels of each output file, not only those shown; fMRIPrep keeps the full T2* map only, so its comparisons cover that map; tedana at four threads and tedana 25.0.1 are compared over their files; the RMSE map and the confounds table in the table of every file. The replica’s record holds each call to tedana 26.0.3’s files, messages and exit status bit for bit in reproduce. The clean room’s comparisons with tedana 26.0.3 and with the replica’s default are judged over the runs’ files by specs/t2smap.md §17.3’s check.py, its record’s thresholds (tedana’s files as stored for the call, with the empty figures folder tedana writes).

Provenance of this example
  • Data: OpenNeuro ds003592 (Neurocognitive aging data release), sub-280, CC0-1.0
  • Inputs: echo 1 (sha256 2181945e16c46e0c…); echo 2 (sha256 81aa91a1100b9abb…); echo 3 (sha256 38c75634bf28d967…); mask (sha256 0b43353da4487e08…); fMRIPrep’s own t2s (sha256 21940313de6a1c60…); fMRIPrep’s own optcom (sha256 ab7e95ea476cbf8a…)
  • Runs: figures cache entry tedana-t2smap/call-ds003592-sub-280-rest-curvefit, computed 2026-11-08T20:29:30+00:00; 4 of its 9 runs made again since, last on 2026-11-12T04:08:06+00:00, with larmorx 48278c64fe24; tedana 26.0.3 in oracles/python-newest (numpy 2.5.4 (OpenBLAS 0.3.34), SciPy 1.18.1: larmorx’s reference) and in the shared .venv (numpy 2.3.5 (OpenBLAS 0.3.30), SciPy 1.15.3); tedana 25.0.1 in fMRIPrep 25.2.6’s image; larmorx-gpl 0.1.0rc1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • The clean room’s runs: figures cache entry tedana-t2smap/fmriprep-ds003592-curvefit-original, computed 2026-11-11T22:22:43+00:00; 4 of its 5 runs made again since, last on 2026-11-12T05:06:43+00:00, with larmorx 55dace9da91f; tedana 26.0.3 (oracles/python-newest, the launcher of the spec’s probe; through the cache of the original’s runs); larmorx-gpl built from the clean-room part’s branch, its program as a black box; the clean room (crate larmorx-tedana) through larmorx._core.tedana_t2smap_cleanroom; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in tedana-t2smap.json, exported by python -m larmorx_validation evidence tedana-t2smap.

The optimally combined series

Worked example

The optimally combined series of ds003592's run

What this shows: In reproduce the replica gives every value of tedana 26.0.3’s combined series, at 1 and 14 threads alike, and the clean room holds them within the thresholds of specs/t2smap.md §17.3: 11,381,033 of the 11,406,278 values of the run’s files are identical, and 0.305% of the curve fit’s voxels are judged by the fit’s own objective (its files are the same at 1 and 8 threads). In fix, their default, both correct TED-05 and TED-08 and the replica also the maths entries NP-03 and GLIBC-07; the two defaults agree by the same thresholds (11,380,746 of 11,406,278 values identical, 0.259%). fMRIPrep’s own combined series, made by tedana 25.0.1, differs from the replica’s reproduce in 39,366 of its 35,225,600 values, tedana’s own step between its versions; run here in fMRIPrep’s image with numpy’s AVX-512 kernels off, as in fMRIPrep’s run, tedana 25.0.1 gives it bit for bit.

The series fMRIPrep carries on with: tedana’s weighted average of the 3 echoes in every voxel and volume, each echo weighted by TE · exp(−TE / T2*) with the voxel’s full T2* map (Posse et al., 1999), over the echoes of good signal of the adaptive mask. 3 of the 200 volumes are shown; the metrics cover every volume. fMRIPrep’s own combined series (tedana 25.0.1) is a panel. LARMORX’s clean room combines the echoes too, in reproduce at one thread and in fix, its default, at eight threads and at one.

Data
OpenNeuro ds003592 (Neurocognitive aging data release), sub-280, task rest (CC0-1.0); shown inside fMRIPrep’s mask dilated by one voxel
Node
fMRIPrep 25.2.6, ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/t2smap_node
Volumes
shown: file volumes 0, 100, 199
Programs
tedana 26.0.3 (both kept environments), tedana 25.0.1 (fMRIPrep’s image), larmorx-gpl at commit 48278c64fe24
Clean room
LARMORX’s clean room (crate larmorx-tedana, experimental) through its binding larmorx._core.tedana_t2smap_cleanroom, as its own example runs it, at commit 55dace9da91f; these are the runs of its own example below, the same call
Grid
64 × 64 × 43 × 200 voxels of 3.38 × 3.38 × 3 mm; shown: 53 × 59 × 43 around the brain
Inputds003592: echo 1 (14 ms)as fMRIPrep gives it to t2smap
Originaltedana t2smap 26.0.3numpy 2.5.4, SciPy 1.18.1, AVX-512 kernels off, one thread: LARMORX's reference
fMRIPrep's outputfMRIPrep's outputfMRIPrep 25.2.6's own file: tedana 25.0.1 in fMRIPrep's image, in its run
Replicareplicalarmorx-gpl, GPL-3.0 · 1 thread, reproduce · against tedana t2smap 26.0.3
Clean roomclean roomlarmorx, Apache-2.0 · 1 thread, reproduce · against tedana t2smap 26.0.3
Clean roomclean roomlarmorx, Apache-2.0 · 8 threads, fix (the default)

Bit-identical35,225,600 of 35,225,600 voxels equal

the replica against the originalreplica − tedana t2smap 26.0.3
the replica against fMRIPrep's outputreplica − fMRIPrep 25.2.6's own file (tedana t2smap 25.0.1)scale ±1.4
the replica against itself: fix and reproducereplica: fix − reproducescale ±1.4
the clean room against the originalclean room (reproduce) − tedana t2smap 26.0.3scale ±1.4

Axial slice 21 of 43, volume 1 of 3; file volume 0; the crosshair where the default's combination differs most from reproduce's. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The original against itself: the yardstick
tedana t2smap: numpy's AVX-512 kernels on − offreferenceA reference row (the original against itself): numpy’s float64 exp and log are Intel’s SVML kernels on CPUs with AVX-512 (NP-04), glibc 2.35’s otherwise; LARMORX’s reference is the path without them (decision D2): 17,772 of 35,225,600 values differ. NP-040.0012517,772 of 35,225,600 (0.0505 %)2.06e−66.1e−72.66e−17
tedana t2smap: numpy 2.3.5 / SciPy 1.15.3 − numpy 2.5.4 / SciPy 1.18.1referenceA reference row (the original against itself): SciPy 1.15.3 → 1.18.1 changed least_squares’ trust-region reflective method and its finite differences (the order of their steps, the SVD factors’ memory order): the curve fit’s last bits, which the full T2* map magnifies where T2* is large: 22,132 of 35,225,600 values differ.0.00012222,132 of 35,225,600 (0.0628 %)7.75e−75.95e−83.78e−18
tedana t2smap: --n-threads 4 − 1referenceA reference row (the original against itself): tedana’s --n-threads 4 runs the curve fits in four joblib processes and lets OpenBLAS use four threads: 1 of 35,225,600 values differ.6.1e−51 of 35,225,600 (2.84e−6 %)1.03e−82.98e−86.65e−22
tedana t2smap 25.0.1 (fMRIPrep's image, here) − 26.0.3referenceA reference row (the original against itself): tedana’s own step from 25.0.1 (fMRIPrep 25.2.6’s) to 26.0.3: tedana 26.0.x caps only the limited T2* map (TED-05; 25.0.1 capped the full map, as LARMORX’s default does), loads the data as float32 (25.0.1: float64), applies the T2* floors to the full map and no longer clamps negative values; and fMRIPrep’s image has other libraries (numpy 2.2.6, which sums more than 8,192 elements in chunks, SciPy 1.15.2, conda-forge’s OpenBLAS): 39,366 of 35,225,600 values differ.1.5639,366 of 35,225,600 (0.112 %)0.003840.0007589.25e−11
tedana t2smap 25.0.1 here, numpy's AVX-512 kernels on (as in fMRIPrep's run) − fMRIPrep's outputreferenceA reference row (the original against itself): tedana 25.0.1 in fMRIPrep’s image, here with numpy’s AVX-512 kernels on (this CPU’s default: fMRIPrep’s runs set only OpenBLAS’s Haswell kernels), against fMRIPrep’s own run of it, made on the project’s earlier main machine, an Intel Core i7-8750H (AVX2, no AVX-512) (12 logical CPUs), where numpy computes float64 exp and log with glibc’s functions: 16,966 of 35,225,600 values differ.0.00012216,966 of 35,225,600 (0.0482 %)6.89e−75.95e−82.99e−18
tedana t2smap 25.0.1 here, numpy's AVX-512 kernels off − fMRIPrep's output (25.0.1 in its run)referenceA reference row (the original against itself): tedana 25.0.1 in fMRIPrep’s image, here with numpy’s AVX-512 kernels off (LARMORX’s reference setting: glibc 2.35’s float64 exp and log, NP-04), against fMRIPrep’s own run of it, made on the project’s earlier main machine, an Intel Core i7-8750H (AVX2, no AVX-512) (12 logical CPUs), where numpy computes float64 exp and log with glibc’s functions: every one of the 35,225,600 values is the same. NP-04identical: 0 of 35,225,600 values differ
The replica against the original and fMRIPrep's output
replica − tedana t2smap 26.0.3Bit-identicalEvery value of the optimally combined series is tedana 26.0.3’s (35,225,600 values): the adaptive mask, the log-linear fit (numpy’s lstsq: LAPACK’s dgelsd over each group of voxels at once, with OpenBLAS’s Haswell kernels), SciPy’s bounded trust-region curve fit voxel by voxel, the floors and the cap, the optimal combination, written as nibabel and nilearn write them.identical: 0 of 35,225,600 values differ
replica − fMRIPrep 25.2.6's own file (tedana t2smap 25.0.1)Within the original's own variabilityThe replica is tedana 26.0.3 bit for bit, so this is tedana’s own step from 25.0.1 (fMRIPrep’s) to 26.0.3 (39,366 of 35,225,600 values): tedana 26.0.x caps only the limited T2* map (TED-05; 25.0.1 capped the full map, as the replica’s default, fix, does), loads the data as float32 (25.0.1: float64), applies the T2* floors to the full map and no longer clamps negative values; and fMRIPrep’s image has other libraries (numpy 2.2.6, which sums more than 8,192 elements in chunks, SciPy 1.15.2, conda-forge’s OpenBLAS). On this machine tedana 25.0.1 in fMRIPrep’s image differs from tedana 26.0.3 in 39,366 values.1.5639,366 of 35,225,600 (0.112 %)0.003840.0007589.25e−11
replica, fix − fMRIPrep 25.2.6's own fileDeliberate correctiontedana’s step from 25.0.1 to 26.0.3 (the replica’s reproduce against fMRIPrep’s output) and the default’s corrections (TED-08, TED-05, NP-03, GLIBC-07): 39,589 of 35,225,600 values differ. The combination weighs the echoes with the full T2* map, which the default caps as 25.0.1 did (TED-05). TED-08 TED-05 NP-03 GLIBC-071.5639,589 of 35,225,600 (0.112 %)0.003190.0007586.4e−11
The replica against itself: thread counts, fix and reproduce
replica: 14 threads − 1 threadBit-identicalThe same bytes at 1 and 14 threads: the replica divides the voxels between threads in fixed pieces, and the log-linear fit runs over each group at once as tedana’s one lstsq call does.identical: 0 of 35,225,600 values differ
replica: fix − reproduceDeliberate correctionThe default corrects TED-08, TED-05, NP-03, GLIBC-07 and names each where it changes a written value: 31,387 of 35,225,600 values change. NP-03: the log-linear fit’s float32 log correctly rounded (numpy’s own kernel is not); GLIBC-07: float64 exp, log and the curve fit’s powers correctly rounded (glibc 2.35’s are not): the curve fit starts from other last bits and ends within its tolerances elsewhere. upstream_defects='reproduce' gives tedana’s outputs. TED-08 TED-05 NP-03 GLIBC-070.45731,387 of 35,225,600 (0.0891 %)0.0009340.0002235.49e−12
The clean room against the original, fMRIPrep's output and the replica
clean room (reproduce) − tedana t2smap 26.0.3Within the original's own variabilityspecs/t2smap.md §17.3 by its check.py, over the run’s files: 11,381,033 of 11,406,278 compared values identical; the curve fit’s (b)/(c) share 0.305% of the fit voxels (the clean room’s own trust-region solver, its log-linear fit in closed form: the fitted numbers within the thresholds, not bit for bit) §17.30.00012218,923 of 35,225,600 (0.0537 %)7.42e−75.95e−83.46e−18
clean room, fix − fMRIPrep 25.2.6's own fileDeliberate correctiontedana’s own step from 25.0.1 to 26.0.3, the default’s corrections (TED-05, TED-08) and the clean room’s own solvers, within §17.3’s thresholds of tedana 26.0.3: 33,948 of 35,225,600 values differ. The combination weighs the echoes with the full T2* map, which the default caps as tedana 25.0.1 did (TED-05). TED-05 TED-081.5633,948 of 35,225,600 (0.0964 %)0.003190.0007586.4e−11
clean room − replica, both fixWithin the original's own variabilityThe two ports’ defaults by §17.3 (check.py), over the run’s files: 11,380,746 of 11,406,278 compared values identical; the (b)/(c) share 0.259%. §17.30.0012518,991 of 35,225,600 (0.0539 %)2.07e−66.1e−72.7e−17
The clean room against itself: thread counts, fix and reproduce
clean room: fix − reproduceDeliberate correctionThe default’s corrections, each named by its notice: TED-05 (the full T2* map capped as the limited one); TED-08 (the failed fits of voxels with one good echo marked): 6,810 of 35,225,600 values change. TED-05 TED-080.4576,810 of 35,225,600 (0.0193 %)0.0009340.0002235.49e−12
clean room, 1 thread − 8 threadsBit-identicalThe same bits at 1 and 8 threads.identical: 0 of 35,225,600 values differ

Δ is A − B over all 35,225,600 values of the 64 × 64 × 43 × 200 series (every volume); runs that are no panel are compared over their files. The replica’s record holds each call to tedana 26.0.3’s files, messages and exit status bit for bit in reproduce. The clean room’s comparisons with tedana 26.0.3 and with the replica’s default are judged over the runs’ files by specs/t2smap.md §17.3’s check.py, its record’s thresholds (tedana’s files as stored for the call, with the empty figures folder tedana writes).

Provenance of this example
  • Data: OpenNeuro ds003592 (Neurocognitive aging data release), sub-280, CC0-1.0
  • Inputs: echo 1 (sha256 2181945e16c46e0c…); echo 2 (sha256 81aa91a1100b9abb…); echo 3 (sha256 38c75634bf28d967…); mask (sha256 0b43353da4487e08…); fMRIPrep’s own t2s (sha256 21940313de6a1c60…); fMRIPrep’s own optcom (sha256 ab7e95ea476cbf8a…)
  • Runs: figures cache entry tedana-t2smap/call-ds003592-sub-280-rest-curvefit, computed 2026-11-08T20:29:30+00:00; 4 of its 9 runs made again since, last on 2026-11-12T04:08:06+00:00, with larmorx 48278c64fe24; tedana 26.0.3 in oracles/python-newest (numpy 2.5.4 (OpenBLAS 0.3.34), SciPy 1.18.1: larmorx’s reference) and in the shared .venv (numpy 2.3.5 (OpenBLAS 0.3.30), SciPy 1.15.3); tedana 25.0.1 in fMRIPrep 25.2.6’s image; larmorx-gpl 0.1.0rc1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • The clean room’s runs: figures cache entry tedana-t2smap/fmriprep-ds003592-curvefit-original, computed 2026-11-11T22:22:43+00:00; 4 of its 5 runs made again since, last on 2026-11-12T05:06:43+00:00, with larmorx 55dace9da91f; tedana 26.0.3 (oracles/python-newest, the launcher of the spec’s probe; through the cache of the original’s runs); larmorx-gpl built from the clean-room part’s branch, its program as a black box; the clean room (crate larmorx-tedana) through larmorx._core.tedana_t2smap_cleanroom; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in tedana-t2smap.json, exported by python -m larmorx_validation evidence tedana-t2smap.

ds000258: four echoes, big-endian, oblique

OpenNeuro ds000258 sub-21262’s resting-state run has four echoes on an oblique grid stored left–posterior–superior. The raw echoes are big-endian int16, and the echoes fMRIPrep gives t2smap keep their byte order, big-endian float32. Its fMRIPrep run was made again on 2026-11-09 on this machine with OpenBLAS’s Haswell kernels, and numpy’s AVX-512 kernels on (the run’s settings name only OpenBLAS’s). tedana 25.0.1, run here in fMRIPrep’s image the same way, gives fMRIPrep’s own T2* map bit for bit; with numpy’s AVX-512 kernels off 629 of its values differ (the two comparisons of tedana 25.0.1 with fMRIPrep’s output).

Worked example

fMRIPrep's T2* and S0 maps of ds000258's multi-echo run

What this shows: In reproduce the replica gives every value of tedana 26.0.3’s maps, at 1 and 14 threads alike, and the clean room holds them within the thresholds of specs/t2smap.md §17.3: 7,311,090 of the 7,342,822 values of the run’s files are identical, and 0.124% of the curve fit’s voxels are judged by the fit’s own objective (its files are the same at 1 and 8 threads). In fix, their default, both correct TED-05 and TED-08 and the replica also the maths entries NP-03 and GLIBC-07; the two defaults agree by the same thresholds (7,312,250 of 7,342,822 values identical, 0.076%). fMRIPrep’s own T2* map, made by tedana 25.0.1, differs from the replica’s reproduce in 852 of its 122,880 values, tedana’s own step between its versions; run here in fMRIPrep’s image with numpy’s AVX-512 kernels on, as in fMRIPrep’s run, tedana 25.0.1 gives it bit for bit.

fMRIPrep 25.2.6’s t2smap call on OpenNeuro ds000258 sub-21262’s rest run (4 echoes (TE 12, 28, 44, 60 ms) of 64 × 64 × 30 voxels of 3.75 × 3.75 × 4.4 mm, 239 volumes (TR 2.47 s), float32, big-endian): tedana’s adaptive mask (how many echoes have good signal in each voxel), the monoexponential fit of S0 and T2* in every voxel (log-linear first, then SciPy’s bounded curve fit from it, --fittype curvefit as fMRIPrep calls it), the full and the limited maps and the fit’s failures (the RMSE map and the confounds table are in the table of every file below). tedana 26.0.3 and the replica compute every map; fMRIPrep’s own T2* map (tedana 25.0.1, the only map fMRIPrep keeps) and further runs of tedana are compared with them value by value. LARMORX’s clean room computes every map too, in reproduce at one thread and in fix, its default, at eight threads and at one; its comparisons with tedana 26.0.3 and with the replica’s default are judged by specs/t2smap.md §17.3’s thresholds.

Data
OpenNeuro ds000258 (Multi-echo Cambridge), sub-21262, task rest (CC0-1.0): the echoes as fMRIPrep’s node was given them; shown inside fMRIPrep’s mask dilated by one voxel, so no face shows
Node
fMRIPrep 25.2.6, ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/t2smap_node (bold_t2smap_wf, its T2SMap interface)
Command
t2smap -d ‘/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-21262_task-rest_echo-1_bold_tshift_xformresampled.nii.gz’ ‘/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-21262_task-rest_echo-2_bold_tshift_xformresampled.nii.gz’ '/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-21262…
Maps
tedana’s full T2* map ranges from 0 to 2.181e+05 s; its adaptive mask keeps 28,977 voxels; the curve fit failed in 316
Programs
tedana 26.0.3 as its program (t2smap) in both kept environments, tedana 25.0.1 in fMRIPrep 25.2.6’s image; LARMORX’s replica, the larmorx-gpl program, at commit 48278c64fe24; numpy’s OpenBLAS on its Haswell kernels, one thread
Clean room
LARMORX’s clean room (crate larmorx-tedana, experimental) through its binding larmorx._core.tedana_t2smap_cleanroom, as its own example runs it, at commit 17c4d48995a1
Grid
64 × 64 × 30 voxels of 3.75 × 3.75 × 4.4 mm; shown: 46 × 54 × 30 around the brain
Inputds000258: echo 1 (12 ms)volume 1 of 239, as fMRIPrep gives it to t2smap (the echo resampled onto the BOLD reference)
Originaltedana t2smap 26.0.3numpy 2.5.4, SciPy 1.18.1, AVX-512 kernels off, one thread: LARMORX's reference
Replicareplicalarmorx-gpl, GPL-3.0 · 1 thread, reproduce · against tedana t2smap 26.0.3
Replicareplicalarmorx-gpl, GPL-3.0 · 14 threads, fix (the default)
Clean roomclean roomlarmorx, Apache-2.0 · 1 thread, reproduce · against tedana t2smap 26.0.3
Clean roomclean roomlarmorx, Apache-2.0 · 8 threads, fix (the default)
the original against itself: numpy's AVX-512 kernels on and offtedana t2smap: numpy's AVX-512 kernels on − offscale ±2.18e5

Bit-identical122,880 of 122,880 voxels equal

the replica against the originalreplica − tedana t2smap 26.0.3
the replica against itself: fix and reproducereplica: fix − reproducescale ±2.18e5
the clean room against the originalclean room (reproduce) − tedana t2smap 26.0.3scale ±2.18e5

Axial slice 18 of 30; the crosshair where the default's T2* map differs most from reproduce's; choose the output above the images: the full T2* map (fMRIPrep's output) and S0, the limited maps, the adaptive mask (good echoes per voxel), the curve fit's failures. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The original against itself: the yardstick
tedana t2smap: numpy's AVX-512 kernels on − offreferenceA reference row (the original against itself): numpy’s float64 exp and log are Intel’s SVML kernels on CPUs with AVX-512 (NP-04), glibc 2.35’s otherwise; LARMORX’s reference is the path without them (decision D2): 2,061 of 737,280 values differ. NP-04T2* (s)1.43e5694 of 122,880 (0.565 %)5750.6550.187
S00.00161345 of 122,880 (0.281 %)7.33e−62.02e−71.15e−16
T2*, limited (s)9.08e−5680 of 122,880 (0.553 %)4.17e−72.54e−55.03e−11
S0, limited0.00161342 of 122,880 (0.278 %)7.33e−62.02e−71.15e−16
Adaptive mask00 of 122,880000
Fit failures00 of 122,880000
tedana t2smap: numpy 2.3.5 / SciPy 1.15.3 − numpy 2.5.4 / SciPy 1.18.1referenceA reference row (the original against itself): SciPy 1.15.3 → 1.18.1 changed least_squares’ trust-region reflective method and its finite differences (the order of their steps, the SVD factors’ memory order): the curve fit’s last bits, which the full T2* map magnifies where T2* is large: 2,508 of 737,280 values differ.T2* (s)1.7e5846 of 122,880 (0.688 %)7290.7780.207
S00.00161411 of 122,880 (0.334 %)6.25e−62.02e−78.33e−17
T2*, limited (s)8.85e−5845 of 122,880 (0.688 %)4.23e−72.47e−55.16e−11
S0, limited0.00161406 of 122,880 (0.33 %)6.25e−62.02e−78.36e−17
Adaptive mask00 of 122,880000
Fit failures00 of 122,880000
tedana t2smap: --n-threads 4 − 1referenceA reference row (the original against itself): tedana’s --n-threads 4 runs the curve fits in four joblib processes and lets OpenBLAS use four threads: every one of the 737,280 values is the same.every outputidentical: 0 of 122,880 values differ
tedana t2smap 25.0.1 (fMRIPrep's image, here) − 26.0.3referenceA reference row (the original against itself): tedana’s own step from 25.0.1 (fMRIPrep 25.2.6’s) to 26.0.3: tedana 26.0.x caps only the limited T2* map (TED-05; 25.0.1 capped the full map, as LARMORX’s default does), loads the data as float32 (25.0.1: float64), applies the T2* floors to the full map and no longer clamps negative values; and fMRIPrep’s image has other libraries (numpy 2.2.6, which sums more than 8,192 elements in chunks, SciPy 1.15.2, conda-forge’s OpenBLAS): 2,660 of 491,520 values differ.T2* (s)2.18e5854 of 122,880 (0.695 %)98010.976
S00.000969480 of 122,880 (0.391 %)5.32e−61.22e−76.04e−17
T2*, limited (s)3.45850 of 122,880 (0.692 %)0.02610.9630.222
S0, limited0.000969476 of 122,880 (0.387 %)5.32e−61.22e−76.06e−17
tedana t2smap 25.0.1 here, numpy's AVX-512 kernels on (as in fMRIPrep's run) − fMRIPrep's outputreferenceA reference row (the original against itself): tedana 25.0.1 in fMRIPrep’s image, here with numpy’s AVX-512 kernels on (this CPU’s default: fMRIPrep’s runs set only OpenBLAS’s Haswell kernels), against fMRIPrep’s own run of it, made on this machine, where numpy uses its AVX-512 kernels by default (NP-04): every one of the 122,880 values is the same.T2* (s)identical: 0 of 122,880 values differ
tedana t2smap 25.0.1 here, numpy's AVX-512 kernels off − fMRIPrep's output (25.0.1 in its run)referenceA reference row (the original against itself): tedana 25.0.1 in fMRIPrep’s image, here with numpy’s AVX-512 kernels off (LARMORX’s reference setting: glibc 2.35’s float64 exp and log, NP-04), against fMRIPrep’s own run of it, made on this machine, where numpy uses its AVX-512 kernels by default (NP-04): 629 of 122,880 values differ. NP-04T2* (s)1.05e−5629 of 122,880 (0.512 %)3.52e−88.77e−66.36e−13
The replica against the original and fMRIPrep's output
replica − tedana t2smap 26.0.3Bit-identicalEvery value of the maps shown is tedana 26.0.3’s (737,280 values): the adaptive mask, the log-linear fit (numpy’s lstsq: LAPACK’s dgelsd over each group of voxels at once, with OpenBLAS’s Haswell kernels), SciPy’s bounded trust-region curve fit voxel by voxel, the floors and the cap, written as nibabel and nilearn write them.every outputidentical: 0 of 122,880 values differ
replica − fMRIPrep 25.2.6's own file (tedana t2smap 25.0.1)Within the original's own variabilityThe replica is tedana 26.0.3 bit for bit, so this is tedana’s own step from 25.0.1 (fMRIPrep’s) to 26.0.3 (852 of 122,880 values): tedana 26.0.x caps only the limited T2* map (TED-05; 25.0.1 capped the full map, as the replica’s default, fix, does), loads the data as float32 (25.0.1: float64), applies the T2* floors to the full map and no longer clamps negative values; and fMRIPrep’s image has other libraries (numpy 2.2.6, which sums more than 8,192 elements in chunks, SciPy 1.15.2, conda-forge’s OpenBLAS). On this machine tedana 25.0.1 in fMRIPrep’s image differs from tedana 26.0.3 in 2,660 values.T2* (s)2.18e5852 of 122,880 (0.693 %)9801.82e50.976
replica, fix − fMRIPrep 25.2.6's own fileDeliberate correctiontedana’s step from 25.0.1 to 26.0.3 (the replica’s reproduce against fMRIPrep’s output) and the default’s corrections (TED-08, TED-05, NP-03, GLIBC-07): 842 of 122,880 values differ. The full T2* map’s cap (TED-05) is 25.0.1’s behaviour again. TED-08 TED-05 NP-03 GLIBC-07T2* (s)3.44842 of 122,880 (0.685 %)0.02612.880.221
The replica against itself: thread counts, fix and reproduce
replica: 14 threads − 1 threadBit-identicalThe same bytes at 1 and 14 threads: the replica divides the voxels between threads in fixed pieces, and the log-linear fit runs over each group at once as tedana’s one lstsq call does.every outputidentical: 0 of 122,880 values differ
replica: fix − reproduceDeliberate correctionThe default corrects TED-08, TED-05, NP-03, GLIBC-07 and names each where it changes a written value: 2,709 of 737,280 values change. TED-05: 14 voxels of the full T2* map more than ten times the cap, which tedana 26.0.x leaves uncapped (25.0.1 capped them), get the limited map’s cap. NP-03: the log-linear fit’s float32 log correctly rounded (numpy’s own kernel is not); GLIBC-07: float64 exp, log and the curve fit’s powers correctly rounded (glibc 2.35’s are not): the curve fit starts from other last bits and ends within its tolerances elsewhere. upstream_defects='reproduce' gives tedana’s outputs. TED-08 TED-05 NP-03 GLIBC-07T2* (s)2.18e5888 of 122,880 (0.723 %)98010.947
S00.000946470 of 122,880 (0.382 %)5.37e−61.19e−76.16e−17
T2*, limited (s)9.35e−5883 of 122,880 (0.719 %)3.67e−72.61e−53.9e−11
S0, limited0.000946464 of 122,880 (0.378 %)5.37e−61.19e−76.18e−17
Adaptive mask00 of 122,880000
Fit failures14 of 122,880 (0.00326 %)0.0057110.00629
The clean room against the original, fMRIPrep's output and the replica
clean room (reproduce) − tedana t2smap 26.0.3Within the original's own variabilityspecs/t2smap.md §17.3 by its check.py, over the run’s files: 7,311,090 of 7,342,822 compared values identical; the curve fit’s (b)/(c) share 0.124% of the fit voxels (the clean room’s own trust-region solver, its log-linear fit in closed form: the fitted numbers within the thresholds, not bit for bit) §17.3T2* (s)1.05e5723 of 122,880 (0.588 %)4150.4830.0877
S00.000847384 of 122,880 (0.312 %)4.96e−61.06e−75.26e−17
T2*, limited (s)7.58e−5718 of 122,880 (0.584 %)4.11e−72.12e−54.89e−11
S0, limited0.000847377 of 122,880 (0.307 %)4.96e−61.06e−75.27e−17
Adaptive mask00 of 122,880000
Fit failures00 of 122,880000
clean room, fix − fMRIPrep 25.2.6's own fileDeliberate correctiontedana’s own step from 25.0.1 to 26.0.3, the default’s corrections (TED-05, TED-08) and the clean room’s own solvers, within §17.3’s thresholds of tedana 26.0.3: 731 of 122,880 values differ. The full T2* map’s cap (TED-05) is tedana 25.0.1’s behaviour again. TED-05 TED-08T2* (s)3.44731 of 122,880 (0.595 %)0.02612.880.221
clean room − replica, both fixWithin the original's own variabilityThe two ports’ defaults by §17.3 (check.py), over the run’s files: 7,312,250 of 7,342,822 compared values identical; the (b)/(c) share 0.076%. §17.3T2* (s)5.1e−5732 of 122,880 (0.596 %)2.64e−71.43e−52e−11
S00.00042412 of 122,880 (0.335 %)4.46e−65.27e−84.23e−17
T2*, limited (s)5.1e−5726 of 122,880 (0.591 %)2.64e−71.43e−52e−11
S0, limited0.00042409 of 122,880 (0.333 %)4.46e−65.27e−84.24e−17
Adaptive mask00 of 122,880000
Fit failures00 of 122,880000
The clean room against itself: thread counts, fix and reproduce
clean room: fix − reproduceDeliberate correctionThe default’s corrections, each named by its notice: TED-05 (the full T2* map capped as the limited one); TED-08 (the failed fits of voxels with one good echo marked): 110 of 737,280 values change. TED-05 caps 14 voxels of the full T2* map. TED-05 TED-08T2* (s)2.41e517 of 122,880 (0.0138 %)99910.946
S00.00012244 of 122,880 (0.0358 %)4.41e−71.53e−84.15e−19
T2*, limited (s)7.45e−93 of 122,880 (0.00244 %)2.6e−112.08e−91.96e−19
S0, limited0.00012242 of 122,880 (0.0342 %)4.4e−71.53e−84.15e−19
Adaptive mask00 of 122,880000
Fit failures14 of 122,880 (0.00326 %)0.0057110.00629
clean room, 1 thread − 8 threadsBit-identicalThe same bits at 1 and 8 threads.every outputidentical: 0 of 122,880 values differ

Δ is A − B over all 122,880 voxels of each output file, not only those shown; fMRIPrep keeps the full T2* map only, so its comparisons cover that map; tedana at four threads and tedana 25.0.1 are compared over their files; the RMSE map and the confounds table in the table of every file. The replica’s record holds each call to tedana 26.0.3’s files, messages and exit status bit for bit in reproduce. The clean room’s comparisons with tedana 26.0.3 and with the replica’s default are judged over the runs’ files by specs/t2smap.md §17.3’s check.py, its record’s thresholds (tedana’s files as stored for the call, with the empty figures folder tedana writes).

Provenance of this example
  • Data: OpenNeuro ds000258 (Multi-echo Cambridge), sub-21262, CC0-1.0
  • Inputs: echo 1 (sha256 2a201f88705f85f3…); echo 2 (sha256 8c4881352e6822b6…); echo 3 (sha256 7b34f5d6ebcedfe1…); echo 4 (sha256 1a0110c23c7591ae…); mask (sha256 b061b7fb2454e0cc…); fMRIPrep’s own t2s (sha256 6e77b65e9b8e2dc2…); fMRIPrep’s own optcom (sha256 fc4fa00a45cfa227…)
  • Runs: figures cache entry tedana-t2smap/call-ds000258-sub-21262-rest-curvefit, computed 2026-11-12T03:59:35+00:00; 3 of its 9 runs made again since, last on 2026-11-12T04:09:23+00:00, with larmorx 48278c64fe24; tedana 26.0.3 in oracles/python-newest (numpy 2.5.4 (OpenBLAS 0.3.34), SciPy 1.18.1: larmorx’s reference) and in the shared .venv (numpy 2.3.5 (OpenBLAS 0.3.30), SciPy 1.15.3); tedana 25.0.1 in fMRIPrep 25.2.6’s image; larmorx-gpl 0.1.0rc1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • The clean room’s runs: figures cache entry tedana-t2smap/clean-room-ds000258-sub-21262-rest-curvefit, computed 2026-11-13T07:01:38+00:00 with larmorx 17c4d48995a1; the clean room (crate larmorx-tedana) through larmorx._core.tedana_t2smap_cleanroom, as the clean room’s own example runs it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in tedana-t2smap.json, exported by python -m larmorx_validation evidence tedana-t2smap.

Every call

fMRIPrep 25.2.6 ran t2smap four times on three fMRIPrep runs: ds003592’s run, ds000210 sub-02’s two runs and ds000258’s; the record adds ds003592’s call with the log-linear fit. In every one the replica gives tedana 26.0.3’s files in both environments. fMRIPrep’s own files, made by tedana 25.0.1, differ from tedana 26.0.3’s for the reasons the table’s note lists, and from LARMORX’s default about as much.

fMRIPrep’s calls, the replica against the original: tedana t2smap 26.0.3 (values that differ)

Callreplica, reproduce, against tedana 26.0.3tedana: shared → newestreplica, fix against reproduceNamed by defaultfMRIPrep's files vs tedana 26.0.3against the replica's fix
ds000210 cuedSGT, curvefit3 echoes, TE 13, 27, 43 msidentical7,1107,338GLIBC-07, NP-03, TED-05, TED-086,3646,470
ds000210 rest, curvefit3 echoes, TE 13.7, 30, 47 msidentical13,66114,378GLIBC-07, NP-03, TED-0513,16413,190
ds000258 rest, curvefit4 echoes, TE 12, 28, 44, 60 msidentical32,88436,842GLIBC-07, NP-03, TED-05, TED-0838,87638,661
ds003592 rest, curvefit3 echoes, TE 14, 29.96, 45.92 msidentical28,80839,865GLIBC-07, NP-03, TED-05, TED-0841,89742,180
ds003592 rest, loglin3 echoes, TE 14, 29.96, 45.92 msidentical0164,720NP-03, TED-05––

From the record (docs/validation/tedana-t2smap-replica.md): each call’s every file, message and exit status compared with tedana 26.0.3 in both kept environments (the replica in reproduce, with each environment’s libraries); the other columns count the values of the images and the confounds table that differ: tedana’s own step between the two environments’ releases, the replica’s default against reproduce (the defects it names: the benchmark’s runs), fMRIPrep’s own T2* map and combined series (tedana 25.0.1 in fMRIPrep 25.2.6’s image; fMRIPrep keeps no other output, and its log-linear call was not run by fMRIPrep) against tedana 26.0.3 and against the default. Why fMRIPrep’s files differ: tedana 26.0.x caps only the limited T2* map (TED-05; 25.0.1 capped the full map, as the replica’s default, fix, does), loads the data as float32 (25.0.1: float64), applies the T2* floors to the full map and no longer clamps negative values; and fMRIPrep’s image has other libraries (numpy 2.2.6, which sums more than 8,192 elements in chunks, SciPy 1.15.2, conda-forge’s OpenBLAS); ds000210’s and ds000258’s runs were also made with numpy’s AVX-512 kernels on (NP-04), ds003592’s on a machine without them.

Those differences are tedana’s own step between its versions, not a difference of machines. tedana 25.0.1, run here in fMRIPrep’s image on fMRIPrep’s own call, gives fMRIPrep’s files bit for bit in all four curve-fit calls once numpy computes float64 exp and log as it did in fMRIPrep’s run:

  • with its AVX-512 kernels for the three runs made on this machine (their settings name only OpenBLAS’s Haswell kernels);
  • with glibc’s functions (the AVX-512 kernels off, LARMORX’s reference setting) for ds003592’s, made on a machine without AVX-512.

With the other kernels the curve fit ends elsewhere in a few voxels (NP-04).

fMRIPrep’s output against the original run here: tedana t2smap 25.0.1 (values that differ)

fMRIPrep's call and filenumpy's AVX-512 kernels offonRun as fMRIPrep's was
ds000210 cuedSGT: T2starmap.nii.gz135,168 values; fMRIPrep’s run made on this machine, numpy’s AVX-512 kernels on1200Bit-identicalkernels on, as in fMRIPrep’s run
ds000210 cuedSGT: desc-optcom_bold.nii.gz35,143,680 values; fMRIPrep’s run made on this machine, numpy’s AVX-512 kernels on3,8950Bit-identicalkernels on, as in fMRIPrep’s run
ds000210 rest: T2starmap.nii.gz238,464 values; fMRIPrep’s run made on this machine, numpy’s AVX-512 kernels on2990Bit-identicalkernels on, as in fMRIPrep’s run
ds000210 rest: desc-optcom_bold.nii.gz48,646,656 values; fMRIPrep’s run made on this machine, numpy’s AVX-512 kernels on7,3100Bit-identicalkernels on, as in fMRIPrep’s run
ds000258 rest: T2starmap.nii.gz122,880 values; fMRIPrep’s run made on this machine, numpy’s AVX-512 kernels on6290Bit-identicalkernels on, as in fMRIPrep’s run
ds000258 rest: desc-optcom_bold.nii.gz29,368,320 values; fMRIPrep’s run made on this machine, numpy’s AVX-512 kernels on24,9130Bit-identicalkernels on, as in fMRIPrep’s run
ds003592 rest: T2starmap.nii.gz176,128 values; fMRIPrep’s run made on the project’s earlier main machine, an Intel Core i7-8750H (AVX2, no AVX-512): numpy’s kernels off there01,620Bit-identicalkernels off, as in fMRIPrep’s run
ds003592 rest: desc-optcom_bold.nii.gz35,225,600 values; fMRIPrep’s run made on the project’s earlier main machine, an Intel Core i7-8750H (AVX2, no AVX-512): numpy’s kernels off there016,966Bit-identicalkernels off, as in fMRIPrep’s run

tedana 25.0.1 in fMRIPrep 25.2.6’s image (Apptainer, through the cache of the original’s runs) on fMRIPrep’s own call of each fMRIPrep run, against the T2* map and the combined series fMRIPrep kept, every value. The fMRIPrep runs of ds000210 and ds000258 were made again on 2026-11-09 on this machine with OpenBLAS’s Haswell kernels; their settings name only those, so numpy used its AVX-512 kernels for float64 exp and log (NP-04). ds003592’s run was made on a machine without AVX-512, where numpy uses glibc 2.35’s functions, as it does here with those kernels off (LARMORX’s reference setting, decision D2). The verdict compares the run made as fMRIPrep’s was: bit-identical in every call; the other column shows how far numpy’s other kernels move the curve fit.

The clean room on fMRIPrep’s calls

The six fMRIPrep calls of the clean room’s record (fMRIPrep’s four, and ds003592’s also with --fittype loglin and with --verbose) are each judged twice by the specification’s thresholds: (A) the clean room’s reproduce against tedana 26.0.3, and (B) its fix against the replica’s fix. The share of a call’s fit voxels judged by the curve fit’s objective rather than by value is the (b)/(c) share. The log-linear call’s values are tedana’s but for one float32 unit of the combined series.

fMRIPrep’s calls, the clean room against the original: tedana t2smap 26.0.3 (A, in reproduce) and against the replica (B, both fix)

Call(A) values identical(A) (b)/(c) share (%)(B) values identical(B) (b)/(c) share (%)checks
ds003592-loglin11299199/11299200–11299200/11299200–13/13
ds003592-curvefit11381033/114062780.30511380746/114062780.25915/15
ds003592-curvefit-verbose11486292/115690340.30511485187/115690340.25915/15
ds000210-cuedSGT-curvefit9430950/943629909430980/9436299015/15
ds000210-rest-curvefit13935546/13946060013935632/139460600.00215/15
ds000258-curvefit7311090/73428220.1247312250/73428220.07615/15

specs/t2smap.md §17.3’s thresholds by its check.py; (b)/(c): the curve-fit voxels whose fit the other run’s is no better than (b), or whose start lies on the feasibility boundary (c), at most 3 % allowed. The log-linear call’s values are tedana’s bit for bit but for one float32 unit in the combined series.

ds003592: tedana, the replica and the clean room

The first example’s call was given to tedana 26.0.3, to the replica’s fix and to the clean room in reproduce at one thread and in fix at eight and at one; the first example shows these same runs of the clean room. The maps shown are the T2* and S0 maps, the RMSE map (no value where fewer than two echoes are good: those voxels are not drawn) and the adaptive mask. The comparisons with tedana and between the two ports’ fix are judged by the specification’s check.py (§17.3): value by value within its thresholds and, for 0.305 % of the curve fit’s voxels (0.259 % between the two fix), by the fit’s own objective. The clean room’s files are the same at one and eight threads.

Worked example

fMRIPrep's multi-echo call: the clean room beside tedana and the replica

What this shows: In reproduce the clean room holds tedana 26.0.3’s maps within the thresholds of specs/t2smap.md §17.3: 11,381,033 of the 11,406,278 values of the run’s files are identical, and 0.305% of the curve fit’s voxels are judged by the fit’s own objective. In fix, its default, it corrects TED-05 and TED-08, as the replica’s default does (which also corrects the maths entry NP-03), and the two defaults agree by the same thresholds (11,380,746 of 11,406,278 values identical, 0.259%); its files are the same at 1 and 8 threads.

fMRIPrep 25.2.6’s t2smap call on OpenNeuro ds003592 sub-280’s rest run (three echoes), with the monoexponential curve fit, given to tedana 26.0.3, to the replica’s default and to LARMORX’s clean room. In reproduce the clean room holds tedana’s maps within specs/t2smap.md §17.3’s thresholds (its own solvers, not tedana’s bits); by default it corrects tedana’s defects, as the replica’s default does, and agrees with it by the same thresholds.

Data
OpenNeuro ds003592, sub-280, task rest (CC0-1.0); shown inside fMRIPrep’s dilated brain mask, dilated by one voxel
Node
fMRIPrep 25.2.6, bold_t2smap_wf (its echoes and mask)
Command
t2smap -d <fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-280_ses-1_task-rest_echo-1_bold_valid_tshift_xformresampled.nii.gz <fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-280_ses-1_task-rest_echo-2_bold_valid_tshift_xformresampled.nii.gz <fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-280_ses-1_task-rest_echo-3_bold_valid_tshift_xformresampled.nii.gz -e 14.0 29.96 45.92 --mask <fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/dilate_mask/dilated_mask.nii.gz --fittype curvefit
Programs
tedana 26.0.3’s t2smap (through the cache of the original’s runs); the replica: larmorx-gpl built from the branch the clean room was written on, its program as a black box; LARMORX’s clean room (crate larmorx-tedana, experimental) from LARMORX at commit 55dace9da91f
Grid
64 × 64 × 43 × 200 voxels of 3.38 × 3.38 × 3 mm; shown: 53 × 59 × 43 around the brain
InputEcho 1, the first volumefMRIPrep's boldref-space series of the first echo (float32), one of the three echoes t2smap combines
Originaltedana t2smap 26.0.3tedana 26.0.3 (its newest environment), through the cache of the original's runs
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 1 thread · against tedana t2smap 26.0.3
Clean roomclean roomlarmorx, Apache-2.0 · fix (the default), 8 threads · against tedana t2smap 26.0.3
the clean room against the originalclean room (reproduce) − tedana t2smap 26.0.3scale ±2.4e5
the clean room against the originalclean room (fix) − tedana t2smap 26.0.3scale ±2.4e5

Axial slice 21 of 43; the T2* map in seconds; each difference on its own symmetric scale. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The replica against the original
replica (fix) − tedana t2smap 26.0.3Deliberate correctionThe default’s corrections, each named by its notice: NP-03 (the correctly rounded float32 logarithm); TED-05 (the full T2* map capped as the limited one); TED-08 (the failed fits of voxels with one good echo marked).T2* map (s)2.4e52,635 of 176,128 (1.5 %)99710.929
S0 map0.007111,151 of 176,128 (0.654 %)1.78e−54.3e−75.6e−16
RMSE map0.000122856 of 176,128 (0.486 %)7.85e−76.42e−81.82e−16
adaptive mask (good echoes)00 of 176,128000
The clean room against the original and the replica
clean room (reproduce) − tedana t2smap 26.0.3Within the original's own variabilityspecs/t2smap.md §17.3 by its check.py: 11,381,033 of 11,406,278 compared values identical; the curve fit’s (b)/(c) share 0.305% of the fit voxels (the clean room’s own trust-region solver, its log-linear fit in closed form: the fitted numbers within the thresholds, not bit for bit, decided 2026-11-11) §17.3T2* map (s)1.38e52,108 of 176,128 (1.2 %)4720.5770.0497
S0 map0.000488639 of 176,128 (0.363 %)4.34e−62.96e−83.35e−17
RMSE map0.000145637 of 176,128 (0.362 %)8.78e−77.63e−82.27e−16
adaptive mask (good echoes)00 of 176,128000
clean room (fix) − tedana t2smap 26.0.3Deliberate correctionThe default’s corrections, each named by its notice: TED-05 (the full T2* map capped as the limited one); TED-08 (the failed fits of voxels with one good echo marked).T2* map (s)2.4e52,110 of 176,128 (1.2 %)99710.929
S0 map0.000488873 of 176,128 (0.496 %)4.57e−62.96e−83.71e−17
RMSE map0.000145638 of 176,128 (0.362 %)8.78e−77.63e−82.27e−16
adaptive mask (good echoes)00 of 176,128000
clean room − replica, both fixWithin the original's own variabilityThe two ports’ defaults by §17.3 (check.py): 11,380,746 of 11,406,278 compared values identical; the (b)/(c) share 0.259%.T2* map (s)0.0008762,157 of 176,128 (1.22 %)2.63e−66.24e−51.43e−10
S0 map0.00711693 of 176,128 (0.393 %)1.76e−54.3e−75.51e−16
RMSE map0.000172672 of 176,128 (0.382 %)8.91e−79.03e−82.34e−16
adaptive mask (good echoes)00 of 176,128000
The clean room against itself: thread counts
clean room, 1 thread − 8 threadsBit-identicalThe same bits at 1 and 8 threads.every outputidentical: 0 of 176,128 values differ

Δ is A − B over every value of the files (the whole grid); the verdicts of the comparisons with tedana and between the defaults are check.py’s (§17.3). The replica’s record holds each call to tedana 26.0.3’s files, messages and exit status bit for bit in reproduce.

Provenance of this example
  • Data: OpenNeuro ds003592, sub-280, CC0-1.0
  • Inputs: echo 1 (sha256 2181945e16c46e0c…); echo 2 (sha256 81aa91a1100b9abb…); echo 3 (sha256 38c75634bf28d967…); fMRIPrep’s dilated brain mask (sha256 0b43353da4487e08…)
  • Runs: figures cache entry tedana-t2smap/fmriprep-ds003592-curvefit-original, computed 2026-11-11T22:22:43+00:00; 4 of its 5 runs made again since, last on 2026-11-12T05:06:43+00:00, with larmorx 55dace9da91f; tedana 26.0.3 (oracles/python-newest, the launcher of the spec’s probe; through the cache of the original’s runs); larmorx-gpl built from the clean-room part’s branch, its program as a black box; the clean room (crate larmorx-tedana) through larmorx._core.tedana_t2smap_cleanroom; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in tedana-t2smap.json, exported by python -m larmorx_validation evidence tedana-t2smap.

Every file

The two calls shown above, file by file: the replica’s every value is tedana’s, the RMSE map (NaN where fewer than two echoes are good) and the confounds table included, at 1 and 14 threads. Its default, fix, changes what the corrections change, and tedana’s own runs show how much tedana varies with its kernels, releases, threads and version.

Every file of the two calls shown, the replica against the original: tedana t2smap 26.0.3 (values that differ)

Filereplica − tedana 26.0.3replica: 14 − 1 threadreplica: fix − reproducetedana: AVX-512 on − offtedana: shared − newesttedana: 4 threads − 1tedana 25.0.1 − 26.0.3tedana 25.0.1: AVX-512 on − off
desc-adaptiveGoodSignal_mask.nii.gzds003592: 176,128 values000000––
desc-decayFitFailures_mask.nii.gzds003592: 176,128 values0014000––
S0map.nii.gzds003592: 176,128 values001,15160477601,160618
desc-rmse_statmap.nii.gzds003592: 176,128 values008565307240128,786563
desc-confounds_timeseries.tsvds003592: 1,800 values0061424301,42932
desc-limited_S0map.nii.gzds003592: 176,128 values001,14160076801,152612
desc-limited_T2starmap.nii.gzds003592: 176,128 values002,6201,6762,18302,5141,615
desc-optcom_bold.nii.gzds003592: 35,225,600 values0031,38717,77222,132139,36616,966
T2starmap.nii.gzds003592: 176,128 values002,6351,6752,18202,5311,620
desc-adaptiveGoodSignal_mask.nii.gzds000258: 122,880 values000000––
desc-decayFitFailures_mask.nii.gzds000258: 122,880 values004000––
S0map.nii.gzds000258: 122,880 values004703454110480343
desc-rmse_statmap.nii.gzds000258: 122,880 values00586496551097,488463
desc-confounds_timeseries.tsvds000258: 2,151 values0092849001,71280
desc-limited_S0map.nii.gzds000258: 122,880 values004643424060476341
desc-limited_T2starmap.nii.gzds000258: 122,880 values008836808450850629
desc-optcom_bold.nii.gzds000258: 29,368,320 values0033,45526,49129,735637,91724,913
T2starmap.nii.gzds000258: 122,880 values008886948460854629

Each image’s stored values and each cell of the confounds table, NaN counted as equal to NaN (the RMSE map holds NaN in the voxels with fewer than two good echoes); the verdict is the replica’s against tedana 26.0.3. tedana 25.0.1 writes no adaptive mask and no failures map in this form (‘–’). The record compares the JSON files too (dataset_description.json but for the host’s and Python’s fields, decision D6).

tedana’s own variability

The record’s spread cases measure how much tedana 26.0.3 varies on one machine when one setting changes:

  • numpy’s AVX-512 kernels for float64 exp and log (NP-04: Intel’s SVML kernels, which numpy uses by default on this CPU; LARMORX’s reference is the path without them, decision D2);
  • the library releases (SciPy’s trust-region method changed between 1.15 and 1.18);
  • four threads (tedana’s curve fits in four processes);
  • its version (25.0.1, in fMRIPrep’s image on this machine).

The replica and the clean room give the same bytes at any thread count.

The original against itself, and the replica at four threads: tedana t2smap 26.0.3 (values that differ from its reference setup)

VariantfMRIPrep's ds003592 callsynthetic, curvefitsynthetic, loglin
--n-threads 4100
numpy’s AVX-512 kernels on22,89980
tedana 25.0.1 (fMRIPrep 25.2.6’s image)176,9384471,646
the shared releases (numpy 2.3.5 (OpenBLAS 0.3.30), SciPy 1.15.3)28,80850
The replica at --n-threads 4000

The record’s spread cases: tedana 26.0.3 in its reference setup (numpy 2.5.4, SciPy 1.18.1, numpy’s AVX-512 kernels off, one thread) against itself with one setting changed, and against tedana 25.0.1 in fMRIPrep’s image on this machine; every value of every image and of the confounds table. The synthetic cases are 12 × 10 × 6 voxels, 24 volumes, four echoes in fMRIPrep’s form. The replica gives the same bytes at any thread count.

Corrections by default

Every defect found in tedana was decided by the register’s rules (A: corrected by default; B: refused with a clear error; C: kept and documented; D: measured first). upstream_defects="reproduce" (on the command line --upstream-defects reproduce from the next release, LARMORX_UPSTREAM_DEFECTS=reproduce in 0.1.0rc1) gives tedana’s behaviour, its errors and messages included (the replica’s findings).

fMRIPrep’s calls meet TED-05 (the full T2* map uncapped, every call), TED-08 (a failed fit of a voxel with one good echo not marked in the failures map, three of the four curve-fit calls) and the maths entries NP-03 and GLIBC-07, all corrected and named by default. The other entries need inputs or options fMRIPrep does not give.

The clean room’s implementers found two more:

  • TED-31 (rule B): a run whose adaptive mask removes every voxel, where tedana fails with an IndexError, stops with a clear error;
  • TED-46 (rule A): a voxel whose observations are all equal, a saturated sample, gets the slope 0 instead of tedana’s rounding noise, which makes its T2* 1 ms or 10¹⁶ ms and more. In the curve fit it gets the fit’s exact solution: S0 the samples’ value, and no decay.

Both ports make every correction by default and name each alike; the clean room’s record holds each case’s fix against the replica’s (the clean room’s findings).

Corrections by default, both ports: tedana t2smap 26.0.3’s defects and the maths entries

EntryWhat tedana doesRuleboth ports' fixThe record's cases
TED-01a NaN first-echo temporal mean (a NaN sample, or no volume left) makes the dropout method’s 33rd percentile NaN: ValueError: attempt to get argmax of an empty sequenceAthe percentile and the exemplars over the means that are not NaN (an infinite mean sorts and matches as in tedana: since 2026-11-12, it was dropped with the NaN before); none left: a clear errorfail/00-ted-01
TED-02--exclude accepts an index equal to the number of volumes (> for >=), then numpy’s IndexErrorBa clear error, as for larger indicesfail/01-ted-02
TED-03without --mask, nilearn’s compute_epi_mask may find nothing (MaskWarning, then IndexError: index -1 ... in the adaptive mask) or too few voxels (the argmax ValueError)Ba clear errorsyn/x-nine-ts and 9 more
TED-04a user mask is applied by voxel index, its affine never compared with the data’s (tedana’s TODO); measured: the four real calls’ masks have the data’s affineBnilearn’s own Mask affine ... is different from img affine error, mask against the first echofail/10-ted-04 and 6 more
TED-05only the limited T2* map is cappedAthe full map gets the limited map’s capsyn/syn-none and 8 more
TED-06a curve fit whose log-linear start is outside the bounds is a silent “failure” keeping the log-linear estimate; measured: of 1,930 failures on fMRIPrep’s four calls 1,919 are such starts (every one a T2* <= 0, a signal rising with the echo time), 11 the evaluation budget; refits from the clipped start beat the log-linear residuals in only 15 to 29 %Ckept, documented (no switch)–
TED-07--verbose writes the variance of S0 as the T2* variance map and the T2* variance as the S0 map (the covariance diagonal unpacked in the wrong order)Athe true names; reproduce swaps them on output; the notice with --verbosesyn/x-verbose-ts and 3 more
TED-08a voxel with one good echo is fitted with two and kept in the full maps, but its failure is not marked in the failures map, nor its variances in the verbose mapsAmarked and written; the notice when such a fit failed (or, with --verbose, was fitted)syn/syn-inf and 4 more
TED-09one infinite sample (or an integer scaled beyond float32) makes T2* and S0 NaN for every voxel of its adaptive-mask group (one lstsq per group)Athat voxel NaN alone, the rest solved without itsyn/syn-inf and 2 more
TED-10an infinite echo time is accepted (taken as milliseconds) and fails in lstsq: LinAlgError: SVD did not converge in Linear Least Squares (numpy 2.3.5 after OpenBLAS’s DLASCL parameter messages on stdout)Ba clear errorfail/03-ted-10
TED-11--n-independent-echos 0: Python’s ZeroDivisionErrorBa clear errorfail/09-ted-11
TED-12a single echo is accepted, then an unrelated numpy error (IndexError in dropout, column_stack’s ValueError in the log-linear fit)Ba clear errorfail/05-ted-12 and 2 more
TED-13--dummy-scans removing every volume (the argmax ValueError after “Mean of empty slice”) or negative (shifts --exclude’s indices up)Ba clear errorfail/08-ted-13 and 1 more
TED-14--debug’s T2* floor message counts echoes times voxels as voxelsAthe voxel count (a message only)–
TED-15--combmode paid with a zero temporal deviation in every good echo of a voxel (every voxel of a one-volume series): ZeroDivisionError: Weights sum to zero, can't be normalized for the whole runA (B for one volume)those voxels get equal weights; a one-volume series a clear errorsyn/x-paid-const and 1 more
TED-16--exclude removing every volume: the adaptive mask from no volume, the log-linear fit of no rows, T2* = 1/0 (then 500 ms) and S0 = 1 in every voxel, written with exit status 0 (random testing)Ba clear errordifftest/ted16-exclude-every-volume
TED-17a single voxel left for the fit: fit_loglinear’s squeezed mask column is 0-dimensional, IndexError: too many indices for array (both fit types and modes; random testing)Afitteddifftest/ted17-single-voxel
TED-18the limited T2* map’s cap is its 99.5th percentile, 0 when more than 99.5 % of its values are 0 (voxels of one good echo): every positive limited T2* set to 0, the RMSE computed with T2* = 0 (random testing)Ano cap of 0 or less, on either map (TED-05’s full-map cap included)difftest/ted18-cap-zero
TED-31no voxel left for the fit (the adaptive mask removed every voxel of the mask): fit_loglinear’s count-1 step indexes column 0 of an array of no column, IndexError: index 0 is out of bounds for axis 1 with size 0, after the adaptive mask image and the echo-time check (both fit types and modes; found by the clean room’s black-box probe, 2026-11-11)Ba clear error (the text both ports share); the replica completed a log-linear run of no fit voxel and stopped a curve fit with nilearn’s empty-mask text until 2026-11-12, in fix and in reproducefail/11-ted-31 and 2 more
TED-46a voxel whose log-linear observations are all equal (saturated samples: no decay) gets the batched lstsq’s rounding noise as its slope: T2* 1 ms, or 1e15-1e18 ms, never 500 ms (found by the clean room’s random testing, 2026-11-11; spec §7.2, §13.4)Athe exact solution (y, 0) in a rank-2 group, the batch’s other columns kept (tedana’s bits); reproduce keeps lstsq’s noise (since 2026-11-12)fail/14-ted-46 and 2 more
NP-03numpy’s float32 log is numpy’s own kernel on x86-64 CPUs with AVX2/FMA3 or AVX-512 (at most 4 ulp from the correctly rounded value), the C library’s elsewhere: tedana’s log-linear fit depends on the CPUAthe correctly rounded logf; named where a written value changessyn/syn-loglin and 49 more
GLIBC-07numpy’s float64 exp and log (glibc 2.35’s where numpy uses no AVX-512 kernel) and SciPy’s powers are not correctly roundedAcorrectly rounded functions; named where a written value changessyn/syn-ts-curve and 9 more
NP-04numpy’s float64 exp and log are Intel’s SVML kernels on CPUs with AVX-512, glibc’s elsewhereCthe path without them is the reference (decision D2); measured in tedana’s own variability–

Each decided by rules A to D (A: corrected by default; B: a clear error; C: kept and documented; D: measured first), each with upstream_defects='reproduce' (LARMORX_UPSTREAM_DEFECTS=reproduce) giving tedana’s behaviour, its errors and messages included; the findings: docs/findings/tedana-t2smap.md. The record’s cases: those whose fix named the entry (the record’s list of what fix changed), else the entry’s own case. fMRIPrep’s calls meet TED-05 (every call), TED-08 (three of the four curve fits) and the maths entries; the others need options or inputs fMRIPrep does not give.

Speed

The clean room is built for speed, and on fMRIPrep’s calls it is limited by reading and writing the compressed images: on ds003592’s call at one thread its log-linear fit and its curve fit take 0.10 and 0.09 s, and reading the echoes 1.6 s (the clean room’s findings).

Its benchmark (report, “The clean-room original”) ran tedana, the replica’s fix and the clean room in both modes on the same calls, the best of two runs at one and eight threads, on a machine shared with other work:

  • on the curve-fit calls 2.7 to 3.7 s at one thread, 10 to 15 times tedana’s speed, and 1.3 to 2.0 s at eight threads, 19 to 30 times;
  • on the log-linear call 3.0 s and 1.5 s, where tedana takes 6.2 s;
  • its files the same at one and eight threads in both modes.

The replica’s benchmark runs tedana 26.0.3 and the replica on the record’s five fMRIPrep calls (fMRIPrep’s four and ds003592’s with the log-linear fit). It times the whole command (reading, the masks, the fits, the maps, the combination, writing) and checks every reproduce run’s files against tedana’s.

  • At 14 threads the replica takes 3.0 to 4.2 times less time than tedana on the curve-fit calls.
  • At one thread it takes about as long (0.8 to 1.1 times tedana’s speed), because both modes run up to four passes to name the maths entries exactly.
  • On the log-linear call, which tedana finishes in 6 s, those passes make it slower.

The chart’s factor is the replica’s at 14 threads; the clean room’s factors are in the table below it.

Speed, both ports against the original: tedana t2smap 26.0.3

Seconds per call (wall time) on a log scale; the factor is tedana’s time divided by the replica’s at 14 threads.

  • tedana t2smap 26.0.3, 1 thread
  • replica, reproduce, 1 thread
  • replica, reproduce, 14 threads
  • replica, fix, 1 thread
  • replica, fix, 14 threads
  • clean room, fix, 1 thread
  • clean room, fix, 8 threads
  1. ds000210 cuedSGT, curvefittedana t2smap 26.0.3, 1 thread 29.2 sreplica, reproduce, 1 thread 31 sreplica, reproduce, 14 threads 8.91 sreplica, fix, 1 thread 32.6 sreplica, fix, 14 threads 9.51 sclean room, fix, 1 thread 2.72 sclean room, fix, 8 threads 1.43 s3.3×
  2. ds000210 rest, curvefittedana t2smap 26.0.3, 1 thread 46.2 sreplica, reproduce, 1 thread 44.7 sreplica, reproduce, 14 threads 13.3 sreplica, fix, 1 thread 44.9 sreplica, fix, 14 threads 13.1 sclean room, fix, 1 thread 3.67 sclean room, fix, 8 threads 1.95 s3.5×
  3. ds000258 rest, curvefittedana t2smap 26.0.3, 1 thread 32.2 sreplica, reproduce, 1 thread 42.2 sreplica, reproduce, 14 threads 10.8 sreplica, fix, 1 thread 41.2 sreplica, fix, 14 threads 10.4 sclean room, fix, 1 thread 2.92 sclean room, fix, 8 threads 1.33 s3×
  4. ds003592 rest, curvefittedana t2smap 26.0.3, 1 thread 44.3 sreplica, reproduce, 1 thread 42.1 sreplica, reproduce, 14 threads 10.6 sreplica, fix, 1 thread 42.2 sreplica, fix, 14 threads 10.6 sclean room, fix, 1 thread 3.05 sclean room, fix, 8 threads 1.56 s4.2×
  5. ds003592 rest, loglintedana t2smap 26.0.3, 1 thread 6.07 sreplica, reproduce, 1 thread 9.99 sreplica, reproduce, 14 threads 7.94 sreplica, fix, 1 thread 9.81 sreplica, fix, 14 threads 7.95 sclean room, fix, 1 thread 3.04 sclean room, fix, 8 threads 1.52 s0.76×
1 s10 s100 s

From the benchmark report docs/benchmarks/tedana-t2smap.json (32 logical CPUs; the best of 2 runs for the replica, tedana once; a machine shared with other work). Both of the replica’s modes run the stages from the mask to the combination again with the other mode’s maths functions, writing nothing, to name NP-03 and GLIBC-07 exactly where they change a value (up to four passes on these calls): one pass of a curve-fit call takes about a quarter of these times. The clean room, the port built for speed, from its own benchmark (docs/benchmarks/tedana-t2smap-original.json, 2026-11-11: the best of two runs at 1 and 8 threads, 8 the threads it was given, on a machine shared with other work; tedana, timed again there, took 6.24 to 49.04 s on these calls).

Speed, both ports against the original: tedana t2smap 26.0.3 (wall time of fMRIPrep’s calls in seconds, the best of two runs)

Calltedanareplica, fix, 1 threadreplica, fix, 8clean room, reproduce, 1clean room, fix, 1clean room, fix, 8× tedana's speed, 1× tedana's speed, 8
ds000210-sub-02-cuedSGT-curvefit27.530.210.52.732.721.4310.119.3
ds000210-sub-02-rest-curvefit4944.714.83.663.671.9513.425.1
ds000258-sub-21262-rest-curvefit3444.3133.012.921.3311.725.6
ds003592-sub-280-rest-curvefit47.143.412.63.163.051.5615.430.2
ds003592-sub-280-rest-loglin6.2410.18.43.053.041.522.054.11

Wall times (s) of fMRIPrep’s multi-echo t2smap calls: tedana 26.0.3 once (fresh), the replica’s default (larmorx-gpl 0.1.0rc1 from PyPI) and the clean room in fix and in reproduce, best of two runs at each thread count; validation/investigations/t2smap-original/bench.py. 32 logical CPUs, a machine shared with other work; the verdict: the clean room’s files the same at 1 and 8 threads in fix and in reproduce. The report: docs/benchmarks/tedana-t2smap.md.

Memory

The memory rule allows LARMORX at most 10 % above the original’s peak plus 8 MiB for the same call. Here the replica needs 44 % of tedana’s peak, and the clean room a quarter at one thread (at eight threads it decodes the three echoes at once: 44 %).

Peak memory, the replica against the original: tedana t2smap 26.0.3 (MiB, one thread)

Calltedana 26.0.3replica, reproducereplica, fixreplica ÷ tedana
fMRIPrep’s ds003592 call (curvefit)1,2465485470.439
the same call with --fittype loglin1,2235465460.446

Each call in a process of its own, its peak resident memory from /usr/bin/time -v (tedana’s program in the newest environment; the replica as the larmorx-gpl program), one run each; tests/parity/test_memory_peak.py checks the same calls through python -m the replica tedana t2smap. The rule: at most 10 % above the original’s peak plus 8 MiB. tedana holds the echoes in float64 and float32 copies at once; the replica holds the float32 data once, one pass’s arrays at a time.

Peak memory, the clean room against the original: tedana t2smap 26.0.3 (MiB, fMRIPrep’s ds003592 curve-fit call)

Runthe clean roomtedanaallowed
the clean room, reproduce, 1 thread(s)3071.24e31.38e3
the clean room, fix, 1 thread(s)3071.24e31.38e3
the clean room, fix, 8 thread(s)5461.24e31.38e3

tests/parity/test_memory_peak.py (the clean room’s test, its log): at most 10 % above tedana’s peak plus the interpreter’s 8 MiB, tedana at one thread through the cache of the original’s runs.

Random testing

Random testing runs random valid and invalid multi-echo inputs and option combinations through tedana and LARMORX and compares every output (how random testing works). A difference is shrunk to the smallest input and fewest options that still show it. It is then fixed and kept as a comparison case, registered as a defect of tedana, or reported.

The replica’s random-testing run (2026-11-08) was 30 minutes long, and a check of its final build 10 minutes:

Random testing, the replica against the original: tedana t2smap 26.0.3 (the run of 2026-11-08 and the check of the final build)

Cases
the run: random cases, each compared with tedana 26.0.3 in both environments3,886
the same (every file, message and exit status; fix as the rule of the corrections expects)3,816
differing (triaged below: fixed in the replica, or TED-16 to TED-18 registered)59
a notice the default gave where the rule expected none (triaged below)11
echoes: 2, 3, 4, 5, 6594 / 1,922 / 794 / 392 / 184
fits: loglin, curvefit, default; modes all, ts982 / 1,979 / 925 / 3,137 / 749
combinations: t2s, paid472 / 861
files: .nii.gz, .nii, .hdr pairs, NIfTI-22,557 / 845 / 222 / 262
without a mask577
distinct options and settings drawn159
length (minutes; 12 jobs at one thread each)30.1
the check of the final build: random cases, each compared with tedana 26.0.3 in both environments1,268
the same (every file, message and exit status; fix as the rule of the corrections expects)1,267
differing (triaged below: fixed in the replica, or TED-16 to TED-18 registered)0
a notice the default gave where the rule expected none (triaged below)1
echoes: 2, 3, 4, 5, 6197 / 613 / 241 / 134 / 83
fits: loglin, curvefit, default; modes all, ts344 / 617 / 307 / 998 / 270
combinations: t2s, paid144 / 277
files: .nii.gz, .nii, .hdr pairs, NIfTI-2847 / 289 / 62 / 70
without a mask171
distinct options and settings drawn114
length (minutes; 12 jobs at one thread each)10.3

Differential random testing (docs/validation/difftest.md): random multi-echo series that share one S0 and one T2* map (a head phantom, rising and dropout voxels, temporal noise, NaN, infinite, zero, negative and −0 samples), two to six echoes in seconds or ms, 1 to 12 volumes, drawn geometries, float32, float64, scaled int16, uint16 and uint8 data, little- or big-endian; masks of every kind or none; every option, abbreviations and = forms, runs twice with and without --overwrite. tedana runs as its program in both environments, the replica in reproduce compared file by file, its fix checked by the rule of the corrections by default. Every case in a memory box of 1.5 GiB. The run (seed t2smap-trial, 30 minutes) found the differences triaged below; every one of its 70 differing or noticed cases agrees with the final build, and the 10-minute check of the final build found one notice (TED-15 in a run made twice), fixed with its permanent case.

The replica’s differences from the original in the run of 2026-11-08, triaged

StatusWhat it wasRegister
fixedvalidation/difftest/cases/tedana-t2smap/qform-signed-zero.jsonHeader bytes of every output (probe 2 and others): for an image with only a qform, nibabel’s img.affine is get_qform in numpy’s longdouble (x87 extended precision; np.dot from +0), so its zeros are +0.0 where larmorx-io’s float64 product gives -0.0, and oblique entries can differ in the last bit; the replica reads it as nibabel does (larmorx-io’s NiftiHeader::best_affine_nibabel, check_affine.py: 4,000 headers ident…–
fixedvalidation/difftest/cases/tedana-t2smap/fresh-header-base-affine.json2 cases (1078, 1386): an affine equal to the fresh header’s own base affine kept both codes 0 in the replica; nibabel’s Nifti1Image(data, affine) sets the sform (code 2) and qform (code 0) unconditionally (_affine2header), the allclose test is only the update at writing–
fixedvalidation/difftest/cases/tedana-t2smap/nifti2-computed-mask.jsonNIfTI-2 echoes without --mask (407 and 8 others): _convert_to_nifti1 returns any Nifti1Image as it is and a Nifti2Image is one, so the reference stays NIfTI-2 and nilearn’s mask and every output with it (the replica wrote NIfTI-1); a user mask is always made a Nifti1Image–
fixedvalidation/difftest/cases/tedana-t2smap/dummy-scans-every-volume.jsonNo volume left (6, 54, 183, 276, 868, 1177 and others; reproduce only where every volume is a dummy scan, TED-13): numpy’s lstsq of no rows returns zeros and T2* = 1/0 (its divide warning), then pandas’ ‘Shape of passed values is (0, 3), indices imply (0, 9)’ in the confounds table (nanpercentile of an empty array returns nanmean’s (0,)); a curve fit’s first bound, np.min of an empty column, raises outside curve_fit…–
registervalidation/difftest/cases/tedana-t2smap/ted16-exclude-every-volume.json--exclude removing every volume: tedana fits nothing (T2* 500 ms, S0 1 everywhere) and exits 0 with a curve fit’s or the dropout method’s unrelated errors otherwise; registered TED-16 (rule B): a clear error by default, reproduce follows tedanaTED-16
registervalidation/difftest/cases/tedana-t2smap/ted17-single-voxel.jsonA single voxel to fit (44, 2286, 2671): fit_loglinear’s np.squeeze of the one-voxel mask column is 0-dimensional and indexing it raises IndexError; registered TED-17 (rule A): fitted by default, reproduce raises tedana’s errorTED-17
registervalidation/difftest/cases/tedana-t2smap/ted18-cap-zero.jsonA limited T2* map whose 99.5th percentile is 0 (1500: more than 99.5 % of its values from voxels of one good echo): the cap sets every limited T2* to 0 and the RMSE divides by it (decay.py:83’s divide warning, missing in the replica); registered TED-18 (rule A): no cap of 0 or less by default (TED-05’s full-map cap followed it), reproduce caps as tedanaTED-18
fixedvalidation/difftest/cases/tedana-t2smap/rmse-infinite-percentiles.jsonnumpy’s floating-point warnings the replica did not print (182, 464, 940, 1137, 1184, 1476, 1533, 1636, 1889 and others): ‘Mean of empty slice’ over an empty axis whatever the result’s size; the scalar 0/0 of --n-independent-echos with no good voxel (utils.py:213, 230); PAID’s mean and std of infinite samples (_methods.py: reduce, subtract; lines per release); nanstd’s subtraction and nanpercentile’s interpolation o…–
fixedvalidation/difftest/cases/tedana-t2smap/shared-percentile-float32.jsonThe shared environment only (3459): numpy 2.3.5’s percentile divides q by a.dtype.type(100), so for float32 means the dropout rank is ceil((n - 1) * float32(0.33)) in float32 (301 means: rank 100; numpy 2.5.4: 99), another exemplar voxel and thresholds–
fixedvalidation/difftest/cases/tedana-t2smap/maths-notices-joint.jsonThe default’s maths notices (981, 1299, 2410; and the notices 917, 1160 the other way): one family of functions alone changed values the other’s choice took back, or changed values only together with it (numpy’s float32 log moves a curve fit’s start, on which path glibc’s exp differs); the program now runs the other mode’s functions altogether first, names nothing when they change nothing, else each family that chan…–
fixedvalidation/difftest/cases/tedana-t2smap/ted07-equal-variances.jsonNotices without a visible change: TED-07 where both variance maps hold the same float32 values (now named only where the swap changes one, outside the one-echo voxels: TED-08’s); TED-04 for a mask of another shape, for which tedana’s own path raises nilearn’s error (now only for a 3D mask of the data’s shape); GLIBC-07 as the maths notices above–
harnessProbes 39 and 60: tedana’s own log-linear T2* differs between the releases in a noise voxel (up to 93 % relative): numpy 2.5.4’s LAPACK 3.12.1 (dgelsd’s QR through dlarf1f) moves the slope’s last bits, which T2* = 1/slope magnifies near zero; the replica follows each release, and the suite now reports the step with that reason instead of requiring a log-linear fit identical–
harnessvalidation/difftest/cases/tedana-t2smap/ted15-first-of-two-runs.json1 case (t2smap-check 1240: a one-volume PAID series run twice without --overwrite): TED-15 was named by the first run, whose error (the default’s clear error, tedana’s ZeroDivisionError) the comparison did not see, the second run failing alike on the existing files; the suite’s fix pass now compares the first of two runs too–
harnessThe check’s harness process ended with SIGSEGV after writing its report (every case judged): the part’s Python extension had been replaced in place (cp) while the night ran; the part’s wheel script now replaces it by an atomic rename–

validation/difftest/triage.toml (block tedana-t2smap, 2026-11-08); each difference has a permanent case (validation/difftest/cases/tedana-t2smap/, 18 in all), run by the record.

The clean room’s random-testing run (2026-11-11) was 30 minutes long. Each random case was run by tedana, by the clean room in both modes and by the replica’s fix: (A) the clean room’s reproduce against tedana, and (B) its fix against the replica’s, both by the clean room’s comparison suite and the specification’s thresholds. Every kind of difference, with what it was:

Random testing, the clean room against the original: tedana t2smap 26.0.3 (A) and against the replica (B) (1,742 random calls, 2026-11-11)

Runagreediffertriage
1,742 random cases (30 minutes, seed t2smap-original-trial)every difference triaged and replayed through the part’s fixes1,692506 triage entries
(A) values (§17.3)The run 2026-11-11-t2smap-original-2 (1,742 cases, 50 differ): 23 runs that both programs stop with the same error before the combination (e.g. PAID’s zero weights in reproduce, findings 46, 175, 278), whose partial outputs were compared exactly; now file by file within 1e-5 relative, the confounds table number by number, a curve-fit run’s maps reported (§17.3 judges them only with the model’s full…––harness
(B) messages, (B) notices, (B) values (§17.3)The run 2026-11-11-t2smap-original-2: 13 cases with equal log-linear observations (uint8 saturated, e.g. findings 11, 348, 548, 999): TED-46’s correction in the clean room’s default moves the cap (TED-05’s count), the curve fit’s failures and messages, while the replica’s default does not take TED-46 yet; explained by the suite (2026-11-11: the same defects named, their counts aside), until…––explained
(B) noticesThe run 2026-11-11-t2smap-original-2: 4 cases whose TED-15 notice counted the first distinct echo count where the replica counts the first of the fit’s groups, count 1 with count 2 (findings 341: 16 against 98 = 16 + 82; 812, 860, 919); fixed (2026-11-11), permanent case tests/run.rs::ted15_counts_the_first_fit_group.––fixed
(B) messages, (B) noticesThe run 2026-11-11-t2smap-original-2: 3 cases with an infinite first-echo sample (findings 455, 588, 761): the default dropped every non-finite mean from the dropout percentile and named TED-01, where tedana (and the replica’s default) sort and match an infinite mean; TED-01 now drops NaN means only (2026-11-11), permanent case tests/run.rs::an_infinite_first_echo_mean_is_no_ted01.––fixed
(B) messages, (B) values (§17.3)The run 2026-11-11-t2smap-original-2: 7 cases whose (B) failures combined the replica’s known message gaps (its zero-padded thresholds, its single traceback) with a stopped run’s values or the two causes above (e.g. findings 278, 701, 879, 1060); with those fixed, the gaps are explained as in every other case (2026-11-11).––harness
(B) values (§17.3)The run 2026-11-11-t2smap-original-2: 1 case outside check.py’s model (TED-01’s NaN mean stops the model as it stops tedana; finding 1449) whose confounds table, compared as text, differed in its last digits; now number by number within 1e-5 (2026-11-11).––harness

Each case one command line on multi-echo files drawn from a recipe (two to six echoes, every image form and data type, every option); (A) the clean room’s reproduce against tedana, (B) its fix against the replica’s, both by the clean room’s comparison suite and specs/t2smap.md §17.3; every difference triaged in validation/difftest/triage.toml.

Every case

Each case of the two records, with what it tests, its command line, its verdict and the reason. Search, filter (or select a segment of a chart above), sort; #case=<name> links to one (the clean room’s cases are named original/<case>).

The clean room’s record:

125 cases

CaseCategoryWhat it testsVerdictChecks
original/helpthe specification's probe-h
command linet2smap -h
Bit-identical

the same exit status, output and messages (no file to compare)

9/9
original/versionthe specification's probe-v
command linet2smap -v
Bit-identical

the same exit status, output and messages (no file to compare)

9/9
original/no-argumentsthe specification's probe(no arguments)
command linet2smap (no arguments)
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 2, the same messages and error); the default against the replica’s the same

9/9
original/bad-choicethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --fittype exp
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --fittype exp
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 2, the same messages and error); the default against the replica’s the same

9/9
original/missing-filethe specification's probe-d <I>/none.nii.gz -e 0.012
command linet2smap -d <I>/none.nii.gz -e 0.012
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 2, the same messages and error); the default against the replica’s the same

9/9
original/bad-intthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --dummy-scans 1.5
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --dummy-scans 1.5
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 2, the same messages and error); the default against the replica’s the same

9/9
original/loglinthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/loglin-msthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 12 28 44 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 12 28 44 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/te-mixedthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 28 44 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 28 44 --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/te-negativethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e -0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e -0.012 0.028 0.044 --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/te-infthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 12 28 inf --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 12 28 inf --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/te-countthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/curvefitthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 11515/11526 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/curvefit-verbose-orig-prefixthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --convention orig --prefix sub-01
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --convention orig --prefix sub-01
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 13363/13812 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/ts-loglinthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/ts-curvefit-verbosethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts --fittype curvefit --verbose
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts --fittype curvefit --verbose
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 44306/46892 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/paidthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/paid-tsthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid --fitmode ts
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid --fitmode ts
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/no-maskthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/masktype-decaythe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/masktype-nonethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/masktype-boththe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype dropout decay
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype dropout decay
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/dummy-scansthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --dummy-scans 2
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --dummy-scans 2
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/dummy-scans-allthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --dummy-scans 6
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --dummy-scans 6
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/excludethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0,2:4
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0,2:4
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/exclude-dummythe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0,1,4 --dummy-scans 2
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0,1,4 --dummy-scans 2
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/exclude-dummy-allthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0,1 --dummy-scans 2
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0,1 --dummy-scans 2
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/exclude-stepthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0:4:2
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0:4:2
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

9/9
original/exclude-openthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 2:
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 2:
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

9/9
original/exclude-negativethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude -1
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude -1
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

9/9
original/exclude-negative-rangethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude -2:1
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude -2:1
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 2, the same messages and error); the default against the replica’s the same

9/9
original/exclude-textthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude a
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude a
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

9/9
original/exclude-largethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 7
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 7
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/exclude-equalthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 6
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 6
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/exclude-allthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0:6
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 0:6
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/exclude-tsthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 1 --fitmode ts
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --exclude 1 --fitmode ts
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/zcatthe specification's probe-d <I>/zcat.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/zcat.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/int16the specification's probe-d <I>/i16-e1.nii.gz <I>/i16-e2.nii.gz <I>/i16-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/i16-e1.nii.gz <I>/i16-e2.nii.gz <I>/i16-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/n-independent-3the specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos 3
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos 3
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/n-independent-4the specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos 4
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos 4
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/n-independent-0the specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos 0
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos 0
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/debug-curvefitthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --debug
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --debug
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 11515/11526 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/quietthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --quiet
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --quiet
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/existsthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

26/26
original/overwritethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz -f
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz -f
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

26/26
original/nan-samplethe specification's probe-d <I>/nan-e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/nan-e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/nan-sample-decaythe specification's probe-d <I>/nan-e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay
command linet2smap -d <I>/nan-e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/inf-samplethe specification's probe-d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/one-echothe specification's probe-d <I>/e1.nii.gz -e 0.012 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz -e 0.012 --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/two-echoesthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/three-dthe specification's probe-d <I>/e1-3d.nii.gz <I>/e1-3d.nii.gz <I>/e1-3d.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1-3d.nii.gz <I>/e1-3d.nii.gz <I>/e1-3d.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/mask-shiftedthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-shifted.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-shifted.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/abbreviationsthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask=<I>/mask.nii.gz --fitt curvefit --comb=paid
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask=<I>/mask.nii.gz --fitt curvefit --comb=paid
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 11515/11526 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/threadsthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --n-threads 2
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --n-threads 2
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 11515/11526 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-nan-echo2the specification's probe-d <I>/e1.nii.gz <I>/nan-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/nan-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-inf-echo1the specification's probe-d <I>/inf-e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/inf-e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-cap-zerothe specification's probe-d <I>/e1.nii.gz <I>/neg-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/neg-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-one-volumethe specification's probe-d <I>/one-e1.nii.gz <I>/one-e2.nii.gz <I>/one-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/one-e1.nii.gz <I>/one-e2.nii.gz <I>/one-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-paid-one-volumethe specification's probe-d <I>/one-e1.nii.gz <I>/one-e2.nii.gz <I>/one-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
command linet2smap -d <I>/one-e1.nii.gz <I>/one-e2.nii.gz <I>/one-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-paid-constantthe specification's probe-d <I>/const-e1.nii.gz <I>/const-e2.nii.gz <I>/const-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
command linet2smap -d <I>/const-e1.nii.gz <I>/const-e2.nii.gz <I>/const-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-single-voxelthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-one.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-one.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-mask-4dthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-4d.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-4d.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-mask-gridthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-grid.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask-grid.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-n-independent-negativethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos -1
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --n-independent-echos -1
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-dummy-negativethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --dummy-scans -1 --exclude 0
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --dummy-scans -1 --exclude 0
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-te-single-infthe specification's probe-d <I>/e1.nii.gz -e inf --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz -e inf --mask <I>/mask.nii.gz
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-debug-tsthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts --debug
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts --debug
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-help-40the specification's probe-h
command linet2smap -h
Bit-identical

the same exit status, output and messages (no file to compare)

9/9
original/x-curvefit-decaythe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --fittype curvefit --debug --verbose
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --fittype curvefit --debug --verbose
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 13390/14036 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-ts-curvefit-decaythe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --fittype curvefit --fitmode ts --debug --verbose
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --fittype curvefit --fitmode ts --debug --verbose
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 44169/48236 values identical; the curve fit’s (b)/(c) share 0.088 %; exit status, files and messages tedana’s

15/15
original/x-one-volume-curvefitthe specification's probe-d <I>/one-e1.nii.gz <I>/one-e2.nii.gz <I>/one-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --debug
command linet2smap -d <I>/one-e1.nii.gz <I>/one-e2.nii.gz <I>/one-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --debug
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 9531/9967 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-inf-sample-curvefitthe specification's probe-d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --debug --verbose
command linet2smap -d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --debug --verbose
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 13757/13812 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-two-echoes-ts-curvefitthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --verbose
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --verbose
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

15/15
original/x-curvefit-used-volumesthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --dummy-scans 1 --exclude 2 --debug
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --dummy-scans 1 --exclude 2 --debug
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 10751/10757 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-curvefit-no-maskthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --fittype curvefit --debug
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --fittype curvefit --debug
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 6643/6654 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-curvefit-one-echo-voxelsthe specification's probe-d <I>/e1.nii.gz <I>/mixed4-e2.nii.gz <I>/mixed4-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype dropout decay --fittype curvefit --verbose --debug
command linet2smap -d <I>/e1.nii.gz <I>/mixed4-e2.nii.gz <I>/mixed4-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype dropout decay --fittype curvefit --verbose --debug
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 13424/13812 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-ts-curvefit-one-echo-voxelsthe specification's probe-d <I>/e1.nii.gz <I>/mixed4-e2.nii.gz <I>/mixed4-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype dropout decay --fittype curvefit --fitmode ts --verbose --debug
command linet2smap -d <I>/e1.nii.gz <I>/mixed4-e2.nii.gz <I>/mixed4-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype dropout decay --fittype curvefit --fitmode ts --verbose --debug
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 44567/46892 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-cap-zero-curvefitthe specification's probe-d <I>/e1.nii.gz <I>/neg-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --debug
command linet2smap -d <I>/e1.nii.gz <I>/neg-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --debug
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

15/15
original/x-ts-curvefit-paidthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --combmode paid
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --combmode paid
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 33156/33206 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-exclude-all-nonethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --exclude 0:6
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --exclude 0:6
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-exclude-all-none-curvefitthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --exclude 0:6 --fittype curvefit
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --exclude 0:6 --fittype curvefit
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-exclude-all-decaythe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --exclude 0:6
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --exclude 0:6
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-dummy-all-nonethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-dummy-all-none-curvefitthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6 --fittype curvefit
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6 --fittype curvefit
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-dummy-all-none-tsthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6 --fitmode ts
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6 --fitmode ts
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-epi-boardthe specification's probe-d <I>/board-e1.nii.gz <I>/board-e2.nii.gz <I>/board-e3.nii.gz -e 0.012 0.028 0.044
command linet2smap -d <I>/board-e1.nii.gz <I>/board-e2.nii.gz <I>/board-e3.nii.gz -e 0.012 0.028 0.044
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-epi-emptythe specification's probe-d <I>/slab-e1.nii.gz <I>/slab-e2.nii.gz <I>/slab-e3.nii.gz -e 0.012 0.028 0.044
command linet2smap -d <I>/slab-e1.nii.gz <I>/slab-e2.nii.gz <I>/slab-e3.nii.gz -e 0.012 0.028 0.044
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-dummy-all-none-2ethe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-underflow-debugthe specification's probe-d <I>/e1.nii.gz <I>/fast-e2.nii.gz <I>/fast-e3.nii.gz -e 0.012 0.01205 0.0121 --mask <I>/mask.nii.gz --masktype none --debug
command linet2smap -d <I>/e1.nii.gz <I>/fast-e2.nii.gz <I>/fast-e3.nii.gz -e 0.012 0.01205 0.0121 --mask <I>/mask.nii.gz --masktype none --debug
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-underflow-debug-curvefitthe specification's probe-d <I>/e1.nii.gz <I>/fast-e2.nii.gz <I>/fast-e3.nii.gz -e 0.012 0.01205 0.0121 --mask <I>/mask.nii.gz --masktype none --debug --fittype curvefit
command linet2smap -d <I>/e1.nii.gz <I>/fast-e2.nii.gz <I>/fast-e3.nii.gz -e 0.012 0.01205 0.0121 --mask <I>/mask.nii.gz --masktype none --debug --fittype curvefit
Expected divergence

SciPy’s own floating-point warnings inside its trust-region and finite-difference steps (overflow in a scalar multiply and a dot product, invalid subtractions) are not reproduced: the clean room’s solver is its own (§12.3 lists the curve fit’s OptimizeWarning o…

13/13
original/x-inf-both-paidthe specification's probe-d <I>/e1.nii.gz <I>/both-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
command linet2smap -d <I>/e1.nii.gz <I>/both-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-inf-echo3-decay-paidthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/inf-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --combmode paid
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/inf-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --masktype decay --combmode paid
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-epi-tinythe specification's probe-d <I>/tiny-e1.nii.gz <I>/tiny-e2.nii.gz <I>/tiny-e3.nii.gz -e 0.012 0.028 0.044
command linet2smap -d <I>/tiny-e1.nii.gz <I>/tiny-e2.nii.gz <I>/tiny-e3.nii.gz -e 0.012 0.028 0.044
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-cap-zero-onethe specification's probe-d <I>/e1.nii.gz <I>/negmost-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/e1.nii.gz <I>/negmost-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-inf-sample-tsthe specification's probe-d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts
command linet2smap -d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-inf-sample-paidthe specification's probe-d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
command linet2smap -d <I>/e1.nii.gz <I>/inf-e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --combmode paid
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-te-count-independentthe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --n-independent-echos 3
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --n-independent-echos 3
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/x-pair-i16the specification's probe-d <I>/p16-e1.hdr <I>/p16-e2.hdr <I>/p16-e3.hdr -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/p16-e1.hdr <I>/p16-e2.hdr <I>/p16-e3.hdr -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-pair-i16-imgthe specification's probe-d <I>/p16-e1.img <I>/p16-e2.img <I>/p16-e3.img -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/p16-e1.img <I>/p16-e2.img <I>/p16-e3.img -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-analyzethe specification's probe-d <I>/a32-e1.hdr <I>/a32-e2.hdr <I>/a32-e3.hdr -e 0.012 0.028 0.044 --mask <I>/a32-mask.hdr
command linet2smap -d <I>/a32-e1.hdr <I>/a32-e2.hdr <I>/a32-e3.hdr -e 0.012 0.028 0.044 --mask <I>/a32-mask.hdr
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-analyze-nomaskthe specification's probe-d <I>/a32-e1.hdr <I>/a32-e2.hdr <I>/a32-e3.hdr -e 0.012 0.028 0.044
command linet2smap -d <I>/a32-e1.hdr <I>/a32-e2.hdr <I>/a32-e3.hdr -e 0.012 0.028 0.044
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-analyze-mask-4dthe specification's probe-d <I>/a32-e1.img <I>/a32-e2.img <I>/a32-e3.img -e 0.012 0.028 0.044 --mask <I>/a32-mask4d.hdr
command linet2smap -d <I>/a32-e1.img <I>/a32-e2.img <I>/a32-e3.img -e 0.012 0.028 0.044 --mask <I>/a32-mask4d.hdr
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-analyze-bethe specification's probe-d <I>/b32-e1.hdr <I>/b32-e2.hdr <I>/b32-e3.hdr -e 0.012 0.028 0.044 --mask <I>/a32-mask.hdr
command linet2smap -d <I>/b32-e1.hdr <I>/b32-e2.hdr <I>/b32-e3.hdr -e 0.012 0.028 0.044 --mask <I>/a32-mask.hdr
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-analyze-spmthe specification's probe-d <I>/s16-e1.hdr <I>/s16-e2.hdr <I>/s16-e3.hdr -e 0.012 0.028 0.044 --mask <I>/s16-mask.hdr
command linet2smap -d <I>/s16-e1.hdr <I>/s16-e2.hdr <I>/s16-e3.hdr -e 0.012 0.028 0.044 --mask <I>/s16-mask.hdr
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-analyze-calthe specification's probe-d <I>/c16-e1.hdr <I>/c16-e2.hdr <I>/c16-e3.hdr -e 0.012 0.028 0.044 --mask <I>/a32-mask.hdr
command linet2smap -d <I>/c16-e1.hdr <I>/c16-e2.hdr <I>/c16-e3.hdr -e 0.012 0.028 0.044 --mask <I>/a32-mask.hdr
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-analyze-matthe specification's probe-d <I>/mat-e1.img <I>/mat-e2.img <I>/mat-e3.img -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/mat-e1.img <I>/mat-e2.img <I>/mat-e3.img -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-mghthe specification's probe-d <I>/m-e1.mgz <I>/m-e2.mgz <I>/m-e3.mgz -e 0.012 0.028 0.044 --mask <I>/m-mask.mgz
command linet2smap -d <I>/m-e1.mgz <I>/m-e2.mgz <I>/m-e3.mgz -e 0.012 0.028 0.044 --mask <I>/m-mask.mgz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-mgh-nomaskthe specification's probe-d <I>/m-e1.mgz <I>/m-e2.mgz <I>/m-e3.mgz -e 0.012 0.028 0.044
command linet2smap -d <I>/m-e1.mgz <I>/m-e2.mgz <I>/m-e3.mgz -e 0.012 0.028 0.044
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-mgh-oblique-i16the specification's probe-d <I>/mo-e1.mgh <I>/mo-e2.mgh <I>/mo-e3.mgh -e 0.012 0.028 0.044 --mask <I>/mo-mask.mgz
command linet2smap -d <I>/mo-e1.mgh <I>/mo-e2.mgh <I>/mo-e3.mgh -e 0.012 0.028 0.044 --mask <I>/mo-mask.mgz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-mgh-oblique-nomaskthe specification's probe-d <I>/mo-e1.mgh <I>/mo-e2.mgh <I>/mo-e3.mgh -e 0.012 0.028 0.044
command linet2smap -d <I>/mo-e1.mgh <I>/mo-e2.mgh <I>/mo-e3.mgh -e 0.012 0.028 0.044
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-mgh-mask-on-niftithe specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mn-mask.mgz
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mn-mask.mgz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-ted46-loglinthe specification's probe-d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
command linet2smap -d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-ted46-loglin-tsthe specification's probe-d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts
command linet2smap -d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fitmode ts
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

13/13
original/x-ted46-curvefitthe specification's probe-d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --debug
command linet2smap -d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --verbose --debug
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 13387/13812 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-ted46-curvefit-tsthe specification's probe-d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --verbose
command linet2smap -d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz <I>/sat-e3.nii.gz -e 0.012 0.028 0.044 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --verbose
Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 44258/46892 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/x-ted46-curvefit-ts-2ethe specification's probe-d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --verbose --debug
command linet2smap -d <I>/sat-e1.nii.gz <I>/sat-e2.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --fittype curvefit --fitmode ts --verbose --debug
Bit-identical

tedana’s values bit for bit, its messages, files and exit status; the default against the replica’s the same by §17.3

15/15
original/x-dummy-all-none-ts-te2the specification's probe-d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6 --fitmode ts
command linet2smap -d <I>/e1.nii.gz <I>/e2.nii.gz <I>/e3.nii.gz -e 0.012 0.028 --mask <I>/mask.nii.gz --masktype none --dummy-scans 6 --fitmode ts
Rejected by both

tedana and the clean room’s reproduce stop the same way (exit status 1, the same messages and error); the default against the replica’s the same

13/13
original/threads-loglinthread countsthe default at 1, 4 and 8 threadsBit-identical

every output the same at 1, 4 and 8 threads (the clean room against itself)

5/5
original/threads-curvefit-tsthread countsthe default at 1, 4 and 8 threadsBit-identical

every output the same at 1, 4 and 8 threads (the clean room against itself)

5/5
original/ds003592-loglinfMRIPrep's callsfMRIPrep 25.2.6's ds003592 call ['--fittype', 'loglin']Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 11299199/11299200 values identical; exit status, files and messages tedana’s

13/13
original/ds003592-curvefit worked examplefMRIPrep's callsfMRIPrep 25.2.6's ds003592 call ['--fittype', 'curvefit']Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 11381033/11406278 values identical; the curve fit’s (b)/(c) share 0.305 %; exit status, files and messages tedana’s

15/15
original/ds003592-curvefit-verbosefMRIPrep's callsfMRIPrep 25.2.6's ds003592 call ['--fittype', 'curvefit', '--verbose']Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 11486292/11569034 values identical; the curve fit’s (b)/(c) share 0.305 %; exit status, files and messages tedana’s

15/15
original/ds000210-cuedSGT-curvefitfMRIPrep's callsfMRIPrep 25.2.6's ds000210-cuedSGT call ['--fittype', 'curvefit']Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 9430950/9436299 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/ds000210-rest-curvefitfMRIPrep's callsfMRIPrep 25.2.6's ds000210-rest call ['--fittype', 'curvefit']Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 13935546/13946060 values identical; the curve fit’s (b)/(c) share 0.0 %; exit status, files and messages tedana’s

15/15
original/ds000258-curvefitfMRIPrep's callsfMRIPrep 25.2.6's ds000258 call ['--fittype', 'curvefit']Within the original's own variability

specs/t2smap.md §17.3 by its check.py: 7311090/7342822 values identical; the curve fit’s (b)/(c) share 0.124 %; exit status, files and messages tedana’s

15/15

The replica’s record:

102 cases

CaseCategoryWhat it testsVerdictChecks
fmriprep/ds000210-sub-02-cuedSGT-curvefitfMRIPrep's callsfMRIPrep 25.2.6's t2smap call
command linet2smap -d '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_cuedSGT_run_01_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-02_task-cuedSGT_run-01_echo-1_bold_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_cuedSGT_run_01_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-02_task-cuedSGT_run-01_echo-2_bold_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_cuedSGT_run_01_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-02_task-cuedSGT_run-01_echo-3_bold_tshift_xformresampled.nii.gz' -e 13.0 27.0 43.0 --mask '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_cuedSGT_run_01_echo_1_wf/bold_native_wf/bold_t2smap_wf/dilate_mask/dilated_mask.nii.gz' --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (10 checks). The default, fix, changes 7,338 values, each change named (the rule of the corrections by default).

10/10
fmriprep/ds000210-sub-02-rest-curvefitfMRIPrep's callsfMRIPrep 25.2.6's t2smap call
command linet2smap -d '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_rest_run_01_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-02_task-rest_run-01_echo-1_bold_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_rest_run_01_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-02_task-rest_run-01_echo-2_bold_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_rest_run_01_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-02_task-rest_run-01_echo-3_bold_tshift_xformresampled.nii.gz' -e 13.700000000000001 30.0 47.0 --mask '<fMRIPrep run>/ds000210-sub-02-noreconall/work/fmriprep_25_2_wf/sub_02_wf/bold_task_rest_run_01_echo_1_wf/bold_native_wf/bold_t2smap_wf/dilate_mask/dilated_mask.nii.gz' --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (10 checks). The default, fix, changes 14,378 values, each change named (the rule of the corrections by default).

10/10
fmriprep/ds000258-sub-21262-rest-curvefitfMRIPrep's callsfMRIPrep 25.2.6's t2smap call
command linet2smap -d '<fMRIPrep run>/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-21262_task-rest_echo-1_bold_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-21262_task-rest_echo-2_bold_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-21262_task-rest_echo-3_bold_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/_echoidx_3/boldref_bold/sub-21262_task-rest_echo-4_bold_tshift_xformresampled.nii.gz' -e 12.0 28.0 44.0 60.0 --mask '<fMRIPrep run>/ds000258-sub-21262-noreconall/work/fmriprep_25_2_wf/sub_21262_wf/bold_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/dilate_mask/dilated_mask.nii.gz' --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (10 checks). The default, fix, changes 36,842 values, each change named (the rule of the corrections by default).

10/10
fmriprep/ds003592-sub-280-rest-curvefitfMRIPrep's callsfMRIPrep 25.2.6's t2smap call
command linet2smap -d '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-280_ses-1_task-rest_echo-1_bold_valid_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-280_ses-1_task-rest_echo-2_bold_valid_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-280_ses-1_task-rest_echo-3_bold_valid_tshift_xformresampled.nii.gz' -e 14.0 29.96 45.92 --mask '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/dilate_mask/dilated_mask.nii.gz' --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (10 checks). The default, fix, changes 39,865 values, each change named (the rule of the corrections by default).

10/10
fmriprep/ds003592-sub-280-rest-loglinfMRIPrep's callsfMRIPrep 25.2.6's t2smap call with --fittype loglin
command linet2smap -d '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-280_ses-1_task-rest_echo-1_bold_valid_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-280_ses-1_task-rest_echo-2_bold_valid_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-280_ses-1_task-rest_echo-3_bold_valid_tshift_xformresampled.nii.gz' -e 14.0 29.96 45.92 --mask '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/dilate_mask/dilated_mask.nii.gz' --fittype loglin
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 164,720 values, each change named (the rule of the corrections by default).

8/8
syn/syn-loglinsynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 f32-e4 -e 0.012 0.028 0.044 0.06 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e4.nii.gz -e 0.012 0.028 0.044 0.06 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 871 values, each change named (the rule of the corrections by default).

8/8
syn/syn-curvefitsynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 f32-e4 -e 0.012 0.028 0.044 0.06 --mask mask --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e4.nii.gz -e 0.012 0.028 0.044 0.06 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 38 values, each change named (the rule of the corrections by default).

8/8
syn/syn-i16-mssynthetic inputst2smap -d i16-e1 i16-e2 i16-e3 -e 12 28 44 --mask mask --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/i16-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/i16-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/i16-e3.nii.gz -e 12 28 44 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 13 values, each change named (the rule of the corrections by default).

8/8
syn/syn-ts-loglinsynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 -e 0.012 0.028 0.044 --mask mask --fitmode ts
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz -e 0.012 0.028 0.044 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --fitmode ts
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 3,960 values, each change named (the rule of the corrections by default).

8/8
syn/syn-ts-curvesynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 -e 0.012 0.028 0.044 --mask mask --fitmode ts --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz -e 0.012 0.028 0.044 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --fitmode ts --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 24 values, each change named (the rule of the corrections by default).

8/8
syn/syn-decaysynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 f32-e4 -e 0.012 0.028 0.044 0.06 --mask mask --masktype decay
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e4.nii.gz -e 0.012 0.028 0.044 0.06 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --masktype decay
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 887 values, each change named (the rule of the corrections by default).

8/8
syn/syn-bothsynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 f32-e4 -e 0.012 0.028 0.044 0.06 --mask mask --masktype dropout decay
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e4.nii.gz -e 0.012 0.028 0.044 0.06 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --masktype dropout decay
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 887 values, each change named (the rule of the corrections by default).

8/8
syn/syn-nonesynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 f32-e4 -e 0.012 0.028 0.044 0.06 --mask mask --masktype none --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e4.nii.gz -e 0.012 0.028 0.044 0.06 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --masktype none --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 38 values, each change named (the rule of the corrections by default).

8/8
syn/syn-excludesynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 f32-e4 -e 0.012 0.028 0.044 0.06 --mask mask --exclude 0,3:5,20 --dummy-scans 2 --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e4.nii.gz -e 0.012 0.028 0.044 0.06 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --exclude 0,3:5,20 --dummy-scans 2 --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 38 values, each change named (the rule of the corrections by default).

8/8
syn/syn-nindsynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 f32-e4 -e 0.012 0.028 0.044 0.06 --mask mask --n-independent-echos 5 --debug
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e4.nii.gz -e 0.012 0.028 0.044 0.06 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --n-independent-echos 5 --debug
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 871 values, each change named (the rule of the corrections by default).

8/8
syn/syn-zcatsynthetic inputst2smap -d zcat -e 0.012 0.028 0.044 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/zcat.nii.gz -e 0.012 0.028 0.044 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 982 values, each change named (the rule of the corrections by default).

8/8
syn/syn-mask4dsynthetic inputst2smap -d f32-e1 f32-e2 f32-e3 -e 0.012 0.028 0.044 --mask mask4d
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz -e 0.012 0.028 0.044 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask4d.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 982 values, each change named (the rule of the corrections by default).

8/8
syn/syn-infsynthetic inputst2smap -d inf-e1 nan-e2 f32-e3 -e 0.012 0.028 0.044 --mask mask --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/inf-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/nan-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz -e 0.012 0.028 0.044 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 10,441 values, each change named (the rule of the corrections by default).

8/8
syn/x-nomask-f32synthetic inputst2smap -d head-e1 head-e2 head-e3 -e 0.013 0.029 0.045
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e3.nii.gz -e 0.013 0.029 0.045
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 1,905 values, each change named (the rule of the corrections by default).

8/8
syn/x-nomask-i16-curvefitsynthetic inputst2smap -d head-i16-e1 head-i16-e2 head-i16-e3 -e 0.013 0.029 0.045 --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-i16-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-i16-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-i16-e3.nii.gz -e 0.013 0.029 0.045 --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 11 values, each change named (the rule of the corrections by default).

8/8
syn/x-nomask-zcatsynthetic inputst2smap -d head-zcat -e 0.013 0.029 0.045
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-zcat.nii.gz -e 0.013 0.029 0.045
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 1,905 values, each change named (the rule of the corrections by default).

8/8
syn/x-paidsynthetic inputst2smap -d head-e1 head-e2 head-e3 -e 0.013 0.029 0.045 --mask head-mask --combmode paid
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e3.nii.gz -e 0.013 0.029 0.045 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-mask.nii.gz --combmode paid
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 1,697 values, each change named (the rule of the corrections by default).

8/8
syn/x-paid-tssynthetic inputst2smap -d head-e1 head-e2 head-e3 -e 0.013 0.029 0.045 --mask head-mask --combmode paid --fitmode ts
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e3.nii.gz -e 0.013 0.029 0.045 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-mask.nii.gz --combmode paid --fitmode ts
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 3,892 values, each change named (the rule of the corrections by default).

8/8
syn/x-verbose-orig-prefixsynthetic inputst2smap -d head-e1 head-e2 head-e3 -e 0.013 0.029 0.045 --mask head-mask --fittype curvefit --verbose --convention orig --prefix sub-01
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e3.nii.gz -e 0.013 0.029 0.045 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-mask.nii.gz --fittype curvefit --verbose --convention orig --prefix sub-01
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 2,031 values, each change named (the rule of the corrections by default).

8/8
syn/x-verbose-tssynthetic inputst2smap -d head-e1 head-e2 head-e3 -e 0.013 0.029 0.045 --mask head-mask --fittype curvefit --verbose --fitmode ts
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e3.nii.gz -e 0.013 0.029 0.045 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-mask.nii.gz --fittype curvefit --verbose --fitmode ts
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 18,229 values, each change named (the rule of the corrections by default).

8/8
syn/x-nine-tssynthetic inputst2smap -d nine-e1 nine-e2 nine-e3 nine-e4 nine-e5 nine-e6 nine-e7 nine-e8 nine-e9 -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --fitmode ts --debug
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e4.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e5.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e6.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e7.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e8.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e9.nii.gz -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --fitmode ts --debug
Rejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0

12/12
syn/x-nine-allsynthetic inputst2smap -d nine-e1 nine-e2 nine-e3 nine-e4 nine-e5 nine-e6 nine-e7 nine-e8 nine-e9 -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --debug
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e4.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e5.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e6.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e7.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e8.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e9.nii.gz -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --debug
Rejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0

12/12
syn/x-one-volumesynthetic inputst2smap -d one-e1 one-e2 one-e3 -e 0.013 0.029 0.045 --masktype none
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e3.nii.gz -e 0.013 0.029 0.045 --masktype none
Rejected by both

tedana exits 1; the same from the replica. ValueError: The mask is invalid as it is empty: it masks all data.

12/12
syn/x-one-volume-tssynthetic inputst2smap -d one-e1 one-e2 one-e3 -e 0.013 0.029 0.045 --masktype none --fitmode ts
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e3.nii.gz -e 0.013 0.029 0.045 --masktype none --fitmode ts
Rejected by both

tedana exits 1; the same from the replica. ValueError: The mask is invalid as it is empty: it masks all data.

12/12
syn/x-paid-constsynthetic inputst2smap -d const-e1 const-e2 const-e3 -e 0.013 0.029 0.045 --mask head-mask --combmode paid
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/const-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/const-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/const-e3.nii.gz -e 0.013 0.029 0.045 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-mask.nii.gz --combmode paid
Rejected by both

tedana exits 1; the same from the replica. ZeroDivisionError: Weights sum to zero, can’t be normalized

12/12
syn/x-small-nomasksynthetic inputst2smap -d small-e1 small-e2 small-e3 -e 0.013 0.029 0.045
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/small-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/small-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/small-e3.nii.gz -e 0.013 0.029 0.045
Rejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0

12/12
syn/x-existssynthetic inputst2smap -d head-e1 head-e2 head-e3 -e 0.013 0.029 0.045 --mask head-mask (run twice)
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e3.nii.gz -e 0.013 0.029 0.045 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-mask.nii.gz (run twice in one folder)
Rejected by both

tedana exits 1; the same from the replica. RuntimeError: File /out/desc-adaptiveGoodSignal_mask.nii.gz already exists. In order to allow overwrite please use the --overwrite option in the command line or the overwrite parameter in the Python API.

12/12
syn/x-overwritesynthetic inputst2smap -d head-e1 head-e2 head-e3 -e 0.013 0.029 0.045 --mask head-mask --overwrite (run twice)
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-e3.nii.gz -e 0.013 0.029 0.045 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/head-mask.nii.gz --overwrite (run twice in one folder)
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 2,033 values, each change named (the rule of the corrections by default).

8/8
syn/x-nine-ts-masksynthetic inputst2smap -d nine-e1 nine-e2 nine-e3 nine-e4 nine-e5 nine-e6 nine-e7 nine-e8 nine-e9 -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --fitmode ts --debug --mask ones-nine
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e4.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e5.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e6.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e7.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e8.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e9.nii.gz -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --fitmode ts --debug --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/ones-nine.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 687 values, each change named (the rule of the corrections by default).

8/8
syn/x-nine-all-masksynthetic inputst2smap -d nine-e1 nine-e2 nine-e3 nine-e4 nine-e5 nine-e6 nine-e7 nine-e8 nine-e9 -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --debug --mask ones-nine
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e3.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e4.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e5.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e6.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e7.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e8.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/nine-e9.nii.gz -e 0.006 0.012 0.018000000000000002 0.024 0.03 0.036000000000000004 0.042 0.048 0.054 --debug --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/ones-nine.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 289 values, each change named (the rule of the corrections by default).

8/8
syn/x-one-volume-masksynthetic inputst2smap -d one-e1 one-e2 one-e3 -e 0.013 0.029 0.045 --masktype none --mask ones-one
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e3.nii.gz -e 0.013 0.029 0.045 --masktype none --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/ones-one.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 51 values, each change named (the rule of the corrections by default).

8/8
syn/x-one-volume-ts-masksynthetic inputst2smap -d one-e1 one-e2 one-e3 -e 0.013 0.029 0.045 --masktype none --fitmode ts --mask ones-one
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/extra/one-e3.nii.gz -e 0.013 0.029 0.045 --masktype none --fitmode ts --mask <workspace>/oracles/inputs/tedana-t2smap-replica/extra/ones-one.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 51 values, each change named (the rule of the corrections by default).

8/8
fail/00-ted-01failures and refusalsTED-01: t2smap -d nan-e1 e2 e3 -e 0.014 0.030 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/nan-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. ValueError: attempt to get argmax of an empty sequence The default (TED-01, rule A): the percentile and the exemplars over the means that are not NaN (an infinite mean sorts and matches as in tedana: since 2026-11-12, it was dropped with the NaN before); none left: a clear error.

13/13
fail/01-ted-02failures and refusalsTED-02: t2smap -d e1 e2 e3 -e 0.014 0.030 0.046 --mask mask --exclude 5
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --exclude 5
Rejected by both

tedana exits 1; the same from the replica. IndexError: index 5 is out of bounds for axis 0 with size 5 The default (TED-02, rule B): a clear error, as for larger indices.

13/13
fail/02-refusedfailures and refusalsa refused input: t2smap -d e1 e2 e3 -e 0.014 0.030 0.046 --mask mask --exclude 6
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --exclude 6
Rejected by both

tedana exits 1; the same from the replica. ValueError: The maximum exclude index (6) is greater than the number of timepoints in the data (5).

12/12
fail/03-ted-10failures and refusalsTED-10: t2smap -d e1 e2 e3 -e inf inf inf --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e inf inf inf --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. numpy.linalg.LinAlgError: SVD did not converge in Linear Least Squares The default (TED-10, rule B): a clear error.

13/13
fail/04-refusedfailures and refusalsa refused input: t2smap -d e1 e2 -e 0.014 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz -e 0.014 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. ValueError: Second dimension of data (2) does not match number of echoes provided (tes; 1)

12/12
fail/05-ted-12failures and refusalsTED-12: t2smap -d e1 -e 0.014 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz -e 0.014 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. IndexError: invalid index to scalar variable. The default (TED-12, rule B): a clear error.

13/13
fail/06-ted-12failures and refusalsTED-12: t2smap -d e1 -e 0.014 --mask mask --masktype none
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz -e 0.014 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --masktype none
Rejected by both

tedana exits 1; the same from the replica. ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 0, the array at index 0 has size 2 and the array at index 1 has size 1 The default (TED-12, rule B): a clear error.

13/13
fail/07-refusedfailures and refusalsa refused input: t2smap -d e1 e2 e3 -e 0.014 0.030 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. ValueError: Second dimension of data (3) does not match number of echoes provided (tes; 2)

12/12
fail/08-ted-13failures and refusalsTED-13: t2smap -d e1 e2 e3 -e 0.014 0.030 0.046 --mask mask --dummy-scans 5
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --dummy-scans 5
Rejected by both

tedana exits 1; the same from the replica. ValueError: attempt to get argmax of an empty sequence The default (TED-13, rule B): a clear error.

13/13
fail/09-ted-11failures and refusalsTED-11: t2smap -d e1 e2 e3 -e 0.014 0.030 0.046 --mask mask --n-independent-echos 0
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --n-independent-echos 0
Rejected by both

tedana exits 1; the same from the replica. ZeroDivisionError: division by zero The default (TED-11, rule B): a clear error.

13/13
fail/10-ted-04failures and refusalsTED-04: t2smap -d e1 e2 e3 -e 0.014 0.030 0.046 --mask mask_aff
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask_aff.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (9 checks). The default, fix, gives the same values.

9/9
fail/11-ted-31failures and refusalsTED-31: t2smap -d dead-e1 e2 e3 -e 0.014 0.030 0.046 --mask mask --masktype none
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/dead-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --masktype none
Rejected by both

tedana exits 1; the same from the replica. IndexError: index 0 is out of bounds for axis 1 with size 0 The default (TED-31, rule B): a clear error (the text both ports share); the replica completed a log-linear run of no fit voxel and stopped a curve fit with nilearn’s empty-mask text until 2026-11-12, in fix and in reproduce.

13/13
fail/12-ted-31failures and refusalsTED-31: t2smap -d dead-e1 e2 e3 -e 0.014 0.030 0.046 --mask mask --masktype none --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/dead-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --masktype none --fittype curvefit
Rejected by both

tedana exits 1; the same from the replica. IndexError: index 0 is out of bounds for axis 1 with size 0 The default (TED-31, rule B): a clear error (the text both ports share); the replica completed a log-linear run of no fit voxel and stopped a curve fit with nilearn’s empty-mask text until 2026-11-12, in fix and in reproduce.

13/13
fail/13-ted-31failures and refusalsTED-31: t2smap -d dead-e1 e2 e3 -e 0.014 0.030 0.046 --mask mask --masktype none --fitmode ts
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/dead-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --masktype none --fitmode ts
Rejected by both

tedana exits 1; the same from the replica. IndexError: index 0 is out of bounds for axis 1 with size 0 The default (TED-31, rule B): a clear error (the text both ports share); the replica completed a log-linear run of no fit voxel and stopped a curve fit with nilearn’s empty-mask text until 2026-11-12, in fix and in reproduce.

13/13
fail/14-ted-46failures and refusalsTED-46: t2smap -d sat-e1 sat-e2 sat-e3 -e 0.014 0.030 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (9 checks). The default, fix, changes 97 values, each change named (the rule of the corrections by default).

9/9
fail/15-ted-46failures and refusalsTED-46: t2smap -d sat-e1 sat-e2 sat-e3 -e 0.014 0.030 0.046 --mask mask --fittype curvefit
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (9 checks). The default, fix, changes 169 values, each change named (the rule of the corrections by default).

9/9
fail/16-ted-46failures and refusalsTED-46: t2smap -d sat-e1 sat-e2 sat-e3 -e 0.014 0.030 0.046 --mask mask --fitmode ts --masktype none
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sat-e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --fitmode ts --masktype none
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (9 checks). The default, fix, changes 305 values, each change named (the rule of the corrections by default).

9/9
fail/17-refusedfailures and refusalsa refused input: t2smap -d e1-3d e1-3d e1-3d -e 0.014 0.030 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1-3d.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1-3d.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1-3d.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. ValueError: Expected 4D image

12/12
fail/18-refusedfailures and refusalsa refused input: t2smap -d e1 e2 e3 -e 0.014 0.030 0.046 --mask mask_grid
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask_grid.nii.gz
Rejected by both

tedana exits 1; the same from the replica. is different from img shape:(6, 6, 4)

12/12
fail/19-refusedfailures and refusalsa refused input: t2smap -d e1 e2 e3 -e 0.014 0.030 0.046 --mask mask_grid_empty
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask_grid_empty.nii.gz
Rejected by both

tedana exits 1; the same from the replica. ValueError: The mask is invalid as it is empty: it masks all data.

12/12
fail/20-ted-09failures and refusalsTED-09: t2smap -d e1 both-e2 e3 -e 0.014 0.030 0.046 --mask mask --combmode paid
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/both-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --combmode paid
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (9 checks). The default, fix, changes 113 values, each change named (the rule of the corrections by default).

9/9
fail/21-refusedfailures and refusalsa refused input: t2smap -d e1 big-e2 big-e3 -e 0.012 0.01205 0.0121 --mask mask --masktype none --debug
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/big-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/big-e3.nii.gz -e 0.012 0.01205 0.0121 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --masktype none --debug
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 89 values, each change named (the rule of the corrections by default).

8/8
fail/22-refusedfailures and refusalsa refused input: t2smap -d sci-e1 sci-e2 sci-e3 -e 0.014 0.030 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sci-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sci-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/sci-e3.nii.gz -e 0.014 0.030 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 42 values, each change named (the rule of the corrections by default).

8/8
fail/23-ted-13failures and refusalsTED-13: t2smap -d e1 e2 e3 -e 0.014 0.030 --mask mask --masktype none --fitmode ts --dummy-scans 5
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/failures/e3.nii.gz -e 0.014 0.030 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/failures/mask.nii.gz --masktype none --fitmode ts --dummy-scans 5
Rejected by both

tedana exits 1; the same from the replica. ValueError: Shape of passed values is (0, 3), indices imply (0, 9) The default (TED-13, rule B): a clear error.

13/13
formats/analyze-f32file formatst2smap -d a32-e1.hdr a32-e2.hdr a32-e3.hdr -e 0.014 0.03 0.046 --mask amask.hdr
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e2.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e3.hdr -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/amask.hdr
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 176 values, each change named (the rule of the corrections by default).

8/8
formats/analyze-i16-be-scaled-matfile formatst2smap -d a16be-e1.hdr a16be-e2.hdr a16be-e3.hdr -e 0.014 0.03 0.046 --mask amask.hdr --fittype curvefit --verbose
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a16be-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a16be-e2.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a16be-e3.hdr -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/amask.hdr --fittype curvefit --verbose
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 176 values, each change named (the rule of the corrections by default).

8/8
formats/analyze-head-nomaskfile formatst2smap -d ahead-e1.hdr ahead-e2.hdr ahead-e3.hdr -e 0.014 0.03 0.046
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/ahead-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/ahead-e2.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/ahead-e3.hdr -e 0.014 0.03 0.046
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 819 values, each change named (the rule of the corrections by default).

8/8
formats/analyze-u8-glcal-Mfile formatst2smap -d a8-e1.img a8-e2.img a8-e3.img -e 0.014 0.03 0.046 --mask mask --combmode paid
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a8-e1.img <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a8-e2.img <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a8-e3.img -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz --combmode paid
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values.

8/8
formats/analyze-gz-f64-pixdimfile formatst2smap -d a64-e1.hdr.gz a64-e2.hdr.gz a64-e3.hdr.gz -e 0.014 0.03 0.046 --mask amask64.hdr.gz --fitmode ts
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a64-e1.hdr.gz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a64-e2.hdr.gz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a64-e3.hdr.gz -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/amask64.hdr.gz --fitmode ts
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 247 values, each change named (the rule of the corrections by default).

8/8
formats/analyze-zcat-i32file formatst2smap -d azcat.hdr -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/azcat.hdr -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values.

8/8
formats/mgh-f32-obliquefile formatst2smap -d m32-e1.mgh m32-e2.mgh m32-e3.mgh -e 0.014 0.03 0.046 --mask mmask.mgz
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e1.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e2.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e3.mgh -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mmask.mgz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 176 values, each change named (the rule of the corrections by default).

8/8
formats/mgz-i16-nomask-curvefitfile formatst2smap -d mhead-e1.mgz mhead-e2.mgz mhead-e3.mgz -e 0.014 0.03 0.046 --fittype curvefit --masktype dropout decay
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mhead-e1.mgz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mhead-e2.mgz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mhead-e3.mgz -e 0.014 0.03 0.046 --fittype curvefit --masktype dropout decay
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 4 values, each change named (the rule of the corrections by default).

8/8
formats/mgz-u16-norasfile formatst2smap -d mu16-e1.mgz mu16-e2.mgz mu16-e3.mgz -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mu16-e1.mgz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mu16-e2.mgz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mu16-e3.mgz -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values.

8/8
formats/mgz-u8file formatst2smap -d m8-e1.mgz m8-e2.mgz m8-e3.mgz -e 0.014 0.03 0.046 --mask mask --dummy-scans 1
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m8-e1.mgz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m8-e2.mgz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m8-e3.mgz -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz --dummy-scans 1
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 51 values, each change named (the rule of the corrections by default).

8/8
formats/mgh-zcat-i32-nofooterfile formatst2smap -d mzcat.mgh -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mzcat.mgh -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 127 values, each change named (the rule of the corrections by default).

8/8
formats/mixed-formatsfile formatst2smap -d a32-e1.hdr m32-e2.mgh m16-e3.mgz -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e2.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m16-e3.mgz -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values.

8/8
formats/x-short-firstfile formatst2smap -d short-e1.hdr a32-e2.hdr a32-e3.hdr -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/short-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e2.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e3.hdr -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. - could the file be damaged?

12/12
formats/x-short-laterfile formatst2smap -d a32-e1.hdr short-e2.hdr a32-e3.hdr -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/short-e2.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e3.hdr -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. - could the file be damaged?

12/12
formats/x-analyze-code7file formatst2smap -d code7-e1.hdr a32-e2.hdr a32-e3.hdr -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/code7-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e2.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e3.hdr -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. nibabel.spatialimages.HeaderDataError: data code 7 not recognized

12/12
formats/x-mat-no-keyfile formatst2smap -d nokey-e1.hdr a32-e2.hdr a32-e3.hdr -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/nokey-e1.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e2.hdr <workspace>/oracles/inputs/tedana-t2smap-replica/formats/a32-e3.hdr -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. ValueError: mat file found but no “mat” or “M” in it

12/12
formats/x-mgh-dims0file formatst2smap -d dims0-e1.mgh m32-e2.mgh m32-e3.mgh -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/dims0-e1.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e2.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e3.mgh -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. nibabel.freesurfer.mghformat.MGHError: Dimensions of the data should be non-zero

12/12
formats/x-mgh-versionfile formatst2smap -d version2-e1.mgh m32-e2.mgh m32-e3.mgh -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/version2-e1.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e2.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e3.mgh -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. nibabel.spatialimages.HeaderDataError: Unknown MGH format version

12/12
formats/x-mgh-short-laterfile formatst2smap -d m32-e1.mgh mshort-e2.mgh m32-e3.mgh -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e1.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mshort-e2.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e3.mgh -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. - could the file be damaged?

12/12
formats/x-mgh-gz-namefile formatst2smap -d gzname-e1.mgh.gz m32-e2.mgh m32-e3.mgh -e 0.014 0.03 0.046 --mask mask
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/formats/gzname-e1.mgh.gz <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e2.mgh <workspace>/oracles/inputs/tedana-t2smap-replica/formats/m32-e3.mgh -e 0.014 0.03 0.046 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/formats/mask.nii.gz
Rejected by both

tedana exits 1; the same from the replica. nibabel.filebasedimages.ImageFileError: Cannot work out file type of “/oracles/inputs/tedana-t2smap-replica/formats/gzname-e1.mgh.gz”

12/12
spread/ds003592-sub-280-rest-curvefittedana's own variabilitytedana's own spread on fMRIPrep's ds003592 call
command linet2smap -d '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_0/boldref_bold/sub-280_ses-1_task-rest_echo-1_bold_valid_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_1/boldref_bold/sub-280_ses-1_task-rest_echo-2_bold_valid_tshift_xformresampled.nii.gz' '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/_echoidx_2/boldref_bold/sub-280_ses-1_task-rest_echo-3_bold_valid_tshift_xformresampled.nii.gz' -e 14.0 29.96 45.92 --mask '<fMRIPrep run>/ds003592-sub-280-noreconall/work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_native_wf/bold_t2smap_wf/dilate_mask/dilated_mask.nii.gz' --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in its reference setup (every file, message and exit status) and at 4 threads; tedana’s own spread, values that differ: --n-threads 4: 1; numpy’s AVX-512 kernels on: 22,899; tedana 25.0.1 (fMRIPrep 25.2.6’s image): 176,938; the shared releases (numpy 2.3.5 (OpenBLAS 0.3.30), SciPy 1.15.3): 28,808.

8/8
spread/syn-curvefittedana's own variabilitytedana's own spread on synthetic echoes (fMRIPrep's form, --fittype curvefit)
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz -e 12 28 44 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --fittype curvefit
Bit-identical

Bit-identical to tedana 26.0.3 in its reference setup (every file, message and exit status) and at 4 threads; tedana’s own spread, values that differ: --n-threads 4: 0; numpy’s AVX-512 kernels on: 8; tedana 25.0.1 (fMRIPrep 25.2.6’s image): 447; the shared releases (numpy 2.3.5 (OpenBLAS 0.3.30), SciPy 1.15.3): 5.

8/8
spread/syn-loglintedana's own variabilitytedana's own spread on synthetic echoes (fMRIPrep's form, --fittype loglin)
command linet2smap -d <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e1.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e2.nii.gz <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/f32-e3.nii.gz -e 12 28 44 --mask <workspace>/oracles/inputs/tedana-t2smap-replica/synthetic/mask.nii.gz --fittype loglin
Bit-identical

Bit-identical to tedana 26.0.3 in its reference setup (every file, message and exit status) and at 4 threads; tedana’s own spread, values that differ: --n-threads 4: 0; numpy’s AVX-512 kernels on: 0; tedana 25.0.1 (fMRIPrep 25.2.6’s image): 1,646; the shared releases (numpy 2.3.5 (OpenBLAS 0.3.30), SciPy 1.15.3): 0.

8/8
difftest/curvefit-no-volumerandom testingrandom testing (2026-11-08): a curve fit of no volume: the first voxel's np.min bound of an empty column raises outside curve_fit's tryRejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0

12/12
difftest/dummy-scans-every-volumerandom testingrandom testing (2026-11-08): every volume a dummy scan with the decay method: numpy's lstsq of no rows, then pandas' shape error in the confounds table (TED-13 refuses it by default)Rejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0

12/12
difftest/empty-mean-no-voxelrandom testingrandom testing (2026-11-08): no volume and no voxel: numpy's 'Mean of empty slice' over an empty axis whatever the result's sizeRejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0

12/12
difftest/fresh-header-base-affinerandom testingrandom testing (2026-11-08): an affine equal to the fresh header's own base affine: nibabel sets the sform and qform of every new imageBit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values. A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/independent-echoes-no-good-voxelrandom testingrandom testing (2026-11-08): --n-independent-echos with no good voxel: numpy's scalar 0/0 warningRejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0

12/12
difftest/maths-notices-jointrandom testingrandom testing (2026-11-08): a change of one voxel's RMSE that needs both of the other mode's function families: NP-03 and GLIBC-07 named togetherBit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 23 values, each change named (the rule of the corrections by default). A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/nifti2-computed-maskrandom testingrandom testing (2026-11-08): NIfTI-2 echoes without a mask: the reference stays a Nifti2Image, so nilearn's mask and every output are NIfTI-2Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 19 values, each change named (the rule of the corrections by default). A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/np03-taken-backrandom testingrandom testing (2026-11-08): numpy's float32 log alone changes values that its float64 exp then takes back: nothing namedBit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values. A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/paid-infinite-samplerandom testingrandom testing (2026-11-08): PAID with an infinite sample: numpy's subtract warning in the standard deviationBit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 266 values, each change named (the rule of the corrections by default). A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/qform-signed-zerorandom testingrandom testing (2026-11-08): a qform-only image: nibabel's get_qform in x87 extended precision sums from +0, so the outputs' srow hold +0.0 where a float64 product gives -0.0Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 22 values, each change named (the rule of the corrections by default). A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/rmse-infinite-percentilesrandom testingrandom testing (2026-11-08): an infinite RMSE: numpy's warnings of nanstd's subtraction and of nanpercentile's interpolationBit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 2,233 values, each change named (the rule of the corrections by default). A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/shared-percentile-float32random testingrandom testing (2026-11-08): numpy 2.3.5's percentile of float32 data divides q in float32: another dropout rank (the shared environment)Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values. A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/ted04-other-shaperandom testingrandom testing (2026-11-08): a mask of another shape: tedana's path raises nilearn's error itself, TED-04 not namedRejected by both

tedana exits 1; the same from the replica. [ 0. 0. 0. 1. ]]

12/12
difftest/ted07-equal-variancesrandom testingrandom testing (2026-11-08): --verbose where both variance maps hold the same values: TED-07 not namedBit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, gives the same values. A mistake of the replica’s found by random testing, fixed; its case.

8/8
difftest/ted15-first-of-two-runsrandom testingrandom testing (2026-11-08): a one-volume PAID series run twice without --overwrite: TED-15 is named by the first run, whose error differs from tedana' (the second run then fails alike); the fix pass compares the first runs tooRejected by both

tedana exits 1; the same from the replica. RuntimeError: File /out/sub-01_adaptive_mask.nii.gz already exists. In order to allow overwrite please use the --overwrite option in the command line or the overwrite parameter in the Python API.

12/12
difftest/ted16-exclude-every-volumerandom testingrandom testing (2026-11-08): --exclude removing every volume: tedana fits nothing and writes T2* = 500 ms, S0 = 1 (TED-16, refused by default)Rejected by both

tedana exits 1; the same from the replica. IndexError: index -1 is out of bounds for axis 0 with size 0 The default (TED-16, rule B): a clear error.

12/12
difftest/ted17-single-voxelrandom testingrandom testing (2026-11-08): a single voxel left for the fit: fit_loglinear's squeezed mask column is 0-dimensional, IndexError (TED-17, fitted by default)Rejected by both

tedana exits 1; the same from the replica. IndexError: too many indices for array: array is 0-dimensional, but 1 were indexed The default (TED-17, rule A): fitted.

12/12
difftest/ted18-cap-zerorandom testingrandom testing (2026-11-08): a limited T2* map whose 99.5th percentile is 0: the cap sets every limited value to 0 and the RMSE divides by it (TED-18, no cap of 0 by default)Bit-identical

Bit-identical to tedana 26.0.3 in both kept environments: every file, message and exit status (8 checks). The default, fix, changes 11 values, each change named (the rule of the corrections by default). TED-18’s case (random testing).

8/8

Known limits

  • The clean room’s reproduce holds tedana’s fitted numbers within the specification’s thresholds, not bit for bit; the replica is the exact one. SciPy’s own floating-point warnings inside its trust-region steps, which tedana prints in degenerate fits, are not part of that contract, and the clean room does not print them (the record’s one expected divergence).
  • Both ports read NIfTI-1 and -2 (.nii, .nii.gz, pairs), ANALYZE 7.5 and MGH. The other formats nibabel reads (AFNI, MINC, PAR/REC) are refused with a clear message, a documented gap.
  • The replica’s log lines are tedana’s. Python’s warnings name the replica’s own place instead of tedana’s installed file, its progress bars are its own, and dataset_description.json describes the replica and the library releases it reproduces where tedana describes its Python (decision D6).
  • Both reproduce tedana 26.0.3, not 25.0.1, fMRIPrep 25.2.6’s: the step between the two is measured above, not reproduced.
  • The records hold fMRIPrep’s calls on its three multi-echo fMRIPrep runs. The replica’s full comparison and the long random-testing runs come in the final validation phase.
  • The cross-platform checks (how they work) hold the other five platforms to LARMORX’s own Linux output.

History

  • 2026-11-08: the replica, translated from tedana 26.0.3 with numpy’s, SciPy’s, nibabel’s and nilearn’s arithmetic; tedana’s defects TED-01 to TED-18 found and the maths entries NP-03, NP-04 and GLIBC-07, each corrected by default, refused with a clear error or kept by rule; its random-testing run, 3,886 random cases in 30 minutes, and the 10-minute check of its final build (the roadmap).
  • 2026-11-09: the fMRIPrep runs of ds000210 and ds000258 made again with OpenBLAS’s Haswell kernels.
  • 2026-11-11 and 12: the clean room, written from specs/t2smap.md, tedana’s documentation, the papers it names and runs of tedana and of the replica; TED-31 and TED-46 found and corrected by both ports; its record, its random-testing run of 2026-11-11 (1,742 random cases in 30 minutes) and its benchmark; the replica’s record made again, ANALYZE and MGH images read (the roadmap).
  • 2026-11-13: the clean room shown in the three worked examples that showed tedana and the replica only (its runs of ds000258’s call made then; ds003592’s are those of its own example).

Data credits

The data this page shows or names: each dataset with its version, DOI and licence, from LARMORX's test-data catalog. Public-domain dedications (CC0, PDDL) ask nothing; the datasets are cited all the same.