Evidence · FSL

FLIRT

flirt finds the affine matrix that aligns one image with another, and fMRIPrep uses its boundary-based registration to align every BOLD run with its T1w. Both ports also run the ten other programs FSL builds with it, from convert_xfm to applyxfm4D, and its scripts pairreg, extracttxt and aff2rigid. The replica (larmorx-nc) gives FSL 6.0.7.23’s matrices, images and printed results byte for byte in all 211 of its compared cases, out of 305, fMRIPrep’s own calls among them. The clean room (larmorx) is within its record’s thresholds in 239 of its 240 compared cases, out of 339, and gives FSL’s bytes in 209 of them. On fMRIPrep’s three calls its matrices are within 0.0113 mm of FSL’s at the 95th percentile of the brain, and on 14 threads it runs fMRIPrep’s call 12 to 14 times as fast as FSL.

Ports
  • replica larmorx-nc FSL Licence: non-commercial use only: translated from FSL 6.0.7.23's source (fsl-flirt 2111.4); it runs as a separate program
  • clean room larmorx Apache-2.0: crate larmorx-fsl, on the registration engine it shares with head-motion correction, written from specs/flirt.md, the FLIRT and BBR papers and runs of FSL 6.0.7.23, without FSL's source
Status
replica experimental; clean room experimental
Records
the replica's: standard comparison (FSL 6.0.7.23) · the clean room's: standard comparison (FSL 6.0.7.23) · benchmark fsl-flirt
Reference
flirt, convert_xfm, avscale, img2imgcoord, std2imgcoord, img2stdcoord, rmsdiff, makerot, midtrans, pointflirt, applyxfm4D, pairreg, extracttxt, aff2rigid · API page fsl-flirt

What is compared

FLIRT finds the affine matrix that best aligns an input image with a reference. A cost function compares the two images under a candidate matrix: the correlation ratio by default; mutual information, normalised correlation, least squares and others on request. A search over rotations on a coarse grid finds the basin. Powell’s method then refines it over a pyramid of blurred and subsampled copies, from 8 mm to the images’ own resolution, by the steps of a schedule. The boundary-based cost (-cost bbr) instead samples the input’s intensity on both sides of a white-matter surface, taken from a segmentation of the reference (-wmseg), and rewards a contrast across it. fMRIPrep relies on it to put each BOLD run on its T1w. -applyxfm applies a given matrix; convert_xfm inverts, concatenates and decomposes matrices.

fMRIPrep calls it once per BOLD run, in its boundary-based registration (bold_reg_wf/fsl_bbr_wf/flt_bbr), after mri_coreg’s initial alignment: flirt -in <boldref> -ref <T1w brain> -wmseg <white matter> -init <initial matrix> -schedule $FSLDIR/etc/flirtsch/bbr.sch -dof 6 -searchcost bbr -basescale 1.

The two ports run beside FSL 6.0.7.23’s programs on the same files with the same arguments. fMRIPrep’s own output is the file it wrote in its run, with FSL in fMRIPrep 25.2.6’s image.

bash
larmorx fsl flirt -in boldref.nii.gz -ref T1w_brain.nii.gz -dof 6 -cost bbr \
    -wmseg wm.nii.gz -init init.mat -schedule bbr.sch -omat bold2t1.mat -out bold_in_t1
larmorx fsl convert_xfm -omat t12bold.mat -inverse bold2t1.mat
larmorx fsl avscale --allparams bold2t1.mat boldref.nii.gz
larmorx fsl rmsdiff bold2t1.mat other.mat boldref.nii.gz

avscale prints the matrix’s rotations, scales and skews; rmsdiff the RMS deviation between two matrices, in mm.

python
import larmorx as lx

r = lx.fsl.flirt("boldref.nii.gz", "T1w_brain.nii.gz", init="init.mat", cost="bbr",
                 search_cost="bbr", dof=6, base_scale=1, schedule="bbr.sch",
                 wm_seg="wm_mask.nii.gz")
r.matrix          # the 4 × 4 matrix, as -omat writes it
r.image           # the BOLD reference on the T1w's grid

The calls above leave implementation at "auto". The API page describes every program and option. How to read what follows: how to read the evidence.

Its two ports

FSL comes under the FSL Licence, which allows non-commercial use only. So LARMORX has two ports of its FSL programs. The replica is translated from FSL’s source and lives in its own package, larmorx-nc (pip install larmorx-nc). The clean room is in larmorx itself. It is written without FSL’s code, from a description of what the programs do (specs/flirt.md), the papers (Jenkinson and Smith 2001; Jenkinson et al. 2002; Greve and Fischl 2009) and runs of FSL’s programs as black boxes.

implementation="auto" | "replica" | "clean-room" chooses between them ("original" in 0.1.0rc1). On the command line, larmorx --implementation clean-room fsl flirt ... does the same from the next release; in 0.1.0rc1 it is the environment variable LARMORX_IMPLEMENTATION. auto takes the replica when larmorx-nc is installed, else the clean room.

The replica follows FSL 6.0.7.23. It runs none of FSL’s programs and needs no FSL installation. Its FSL libraries are the ones the replicas of mcflirt and BET in larmorx-nc use too: newimage, its cost functions, and the matrix arithmetic of Armadillo 14.2.1 on OpenBLAS’s LAPACK.

The clean room follows FSL 6.0.7.23 too. It has every program, option, message and exit status of the replica. It computes FLIRT’s costs, Powell’s optimiser and its matrix algebra on the engine that LARMORX’s head-motion correction uses (lx.mri.hmc), so the two share one piece of code for the correlation ratio and Woods’ cost. It is built for speed. The boundary points, the correlation ratio’s bins and the search’s grid run in parallel and are summed in a fixed order, so the result is the same at any thread count. Its memory peak stays within 10 % of FSL’s, plus the interpreter’s 8 MiB, on every measured call.

The C library’s last digits (since 2026-11-04). FLIRT’s arithmetic calls the C library for its rotations’ sines and cosines (sinf, cosf), the angles of a matrix (atan2f), the Gaussian blurs (exp), the windowed sinc’s kernel (sinf, cos) and the mutual information’s logarithms (logf). FSL takes them from glibc, the C library of Linux, whose results are now and then one unit in the last place away from the exact value rounded once. LARMORX computes them correctly rounded, the same on every platform, and until 2026-11-04 the clean room did so in fix and in reproduce. Since then its reproduce computes them exactly as glibc 2.35 does, with LARMORX’s versions of glibc’s functions (larmorx_core::math), and what moved is large:

  • 41 cases of its record that were within the thresholds without FSL’s bytes now give FSL’s files byte for byte: FLIRT’s default registration and those with 6, 7 and 9 degrees of freedom, mutual information and normmi, the windowed sinc, the schedules, the verbose levels up to 3;
  • -cost normmi and weighted normmi pass, 10 mm off before;
  • fMRIPrep’s three boundary-based calls came closer to FSL’s matrices: at the 95th percentile of the brain from 0.0101, 0.0011 and 0.006 mm to 0.0065, 0.0002 and under 0.00001 mm, each then within FSL’s own variability (ds005040’s is 0.0113 mm since the clean-room catch-up of 2026-11-05, when the clean room took in the specification’s answers of 2026-11-03: the verdicts, History);
  • a registration is an iterative fit, so a few other cases moved the other way, by the size of the fit’s own sensitivity to those last bits and within their thresholds: bbr/default from 0.020 to 0.069 mm of its 0.1 mm, real/searchr from 0.033 to 0.135 mm and real/weights from 0.21 to 0.31 mm of their 0.5 mm. The clean room’s own two modes part by as much (findings).

Its default, fix, still computes these functions correctly rounded. That is a deliberate correction, GLIBC-07 in the register, and a registration names it nowhere: whether a registration moved with such last bits cannot be told without running it twice (a decision of the project of 2026-11-04, recorded with GLIBC-07). avscale, makerot and convert_xfm -fixscaleskew name it exactly where their printed or written values differ.

The verdicts at a glance

The replica is a translation of FSL’s source, so the question for it is whether it gives FSL’s bytes: every file FSL writes (the matrix as text, the registered image’s header and data), its exit status and its messages.

Agreement, replica against the original: FSL 6.0.7.23’s FLIRT programs (standard comparison, 305 cases of the record), by category and verdict
fMRIPrep's calls3
real registrations61
boundary-based registration11
schedules29
verbose levels8
edge cases and errors35
convert_xfm16
FLIRT's other programs132
random testing10

Select a segment to list its cases below.

The replica’s record runs 305 cases against FSL 6.0.7.23 (the record). 211 give FSL’s files byte for byte. In 87 FSL refuses the input and the replica refuses it alike, with the same message and exit status (rejected by both). In 7 FSL crashes or never finishes, where the replica goes on, or stops with the reason, as its register entry says (expected divergences: FSL-71, FSL-72, FSL-75, FSL-93, FSL-204). The cases cover:

  • fMRIPrep’s boundary-based calls on three fMRIPrep runs (a young adult, the data panel’s 82-year-old and its three-year-old), whose matrices are also fMRIPrep’s own, as text;
  • every cost function with and without weights, the degrees of freedom, the search options, every interpolation and output data type;
  • every schedule FSL ships, and written ones that use every command of the schedule language, the principal-axes alignment alignpaxes and parameter subsets of any weights among them;
  • -applyxfm and -applyisoxfm on 3D and 4D images, the boundary-based options (its types, slope, given coordinates and normals, a field map), the 2D mode, quaternion parameters (-anglerep quaternion), the debug volumes (-debug) and every error of the command line;
  • convert_xfm with every option;
  • each path of FLIRT’s other programs (avscale, img2imgcoord, std2imgcoord, img2stdcoord, rmsdiff, makerot, midtrans, pointflirt, applyxfm4D in every data type and interpolation) and of its scripts (pairreg, extracttxt, aff2rigid): their usage, every option, their printed results and messages.

Each case also runs the replica’s default, fix, where LARMORX corrects FSL’s deterministic defects and names them; on fMRIPrep’s calls it changes nothing.

Four random-testing runs of the replica:

  • Its first run, 3,945 random calls on small synthetic images, found two kinds of difference, both fixed and kept as cases of the record. The rest were settled next (findings): FSL corrupting its heap on small images, traced to its background value (FSL-75, below), a base-scale quotient FLIRT takes in single precision (fixed), and two defects LARMORX’s default had corrected without naming them (named now).
  • Two shorter runs of the new paths, 12,376 and 3,274 random calls (quaternions and the other programs), found one missing message, the background value’s order of NaN voxels and makerot’s notices, all fixed.
  • A run of half an hour on part 3’s paths, 209,657 random calls (midtrans, pointflirt, applyxfm4D, the schedule’s alignpaxes and parameter subsets, extracttxt), found applyxfm4D’s double-precision arithmetic, the order of NaN voxels in its background value, a singular matrix in a cost evaluation and a volume too large to allocate (FLIRT’s exceptions), and extracttxt’s messages for a range before the first line: all fixed and kept as cases. The sign of NaN results from images with NaN or infinite voxels in reproduce is left open (findings).

Every other finding that fits a run’s memory budget agrees, in its shrunk and drawn forms.

The clean room is written without FSL’s source, so its record holds it to thresholds instead (specs/flirt.md §13):

  • exit statuses and messages FSL’s, the numbers in them read as numbers;
  • every file FSL writes identical;
  • a registration’s matrix within a displacement of FSL’s over the reference’s brain: its 95th percentile within 0.1 mm for the boundary-based and 6-dof registrations, 0.5 mm for the others, or within FSL’s own variability where the registration is poorly conditioned;
  • its image then within a correlation of FSL’s.

A case is bit-identical when every value it compares is FSL’s. It is within the original’s own variability when the matrix is beyond the fixed threshold but within FSL’s own on the same call with its input perturbed by noise of half an intensity unit (for the worked examples, within that variability measured for them). It is open when a file only meets its threshold, or misses it.

Agreement, clean room against the original: FSL 6.0.7.23’s FLIRT programs (standard comparison, 339 cases of the record), by category and verdict
fMRIPrep's calls3
real registrations62
boundary-based registration11
schedules30
verbose levels21
edge cases and errors35
convert_xfm16
FLIRT's other programs140
-debug4
-out's header1
random testing16

Select a segment to list its cases below.

The clean room’s record runs the replica’s standard cases and its own against FSL 6.0.7.23 (the record): the verbose levels and -debug, and since the clean-room catch-up of 2026-11-05 also the specification’s answers and random testing’s permanent cases. 239 of its 240 compared cases are within the thresholds, since the record was made again on 2026-11-07 with the weighted correlation’s count (below). 89 are rejected by both alike. 10 are expected divergences: FLIRT crashes or never finishes (FSL-71, FSL-72, FSL-75, FSL-93, and pointflirt without points: FSL-204), or midtrans writes rows of NaN with exit status 0 (FSL-241, found by the catch-up’s runs of FSL; the clean room refuses such transforms in fix and in reproduce).

  • Bit-identical, 209 cases. Since 2026-11-04 most registrations: FLIRT’s default and 6, 7 and 9 degrees of freedom, the correlation ratio, mutual information and normmi, the schedules, -nosearch, -init, the output’s sinc interpolation. The verbose levels up to 3, with -debug. -applyxfm with nearest neighbour, trilinear and windowed-sinc interpolation, and convert_xfm. FLIRT’s other programs’ printed results and files: avscale, the coordinate programs, rmsdiff, makerot, midtrans, pointflirt, applyxfm4D but for its double-precision sinc and its spline kernels’ last bits, pairreg’s messages, extracttxt, aff2rigid. The schedules’ and the command line’s messages.
  • Within the thresholds. The verbose levels 5 to 7 and level 2 with -out: FSL’s lines, their numbers those of the clean room’s own path; at levels 5 and above each run of cost evaluations is counted as one, their number following the optimiser’s path. A registration that is not bit-identical ends a little elsewhere: its costs follow FSL’s arithmetic within their last bits, and the optimiser’s path then parts (the boundary-based registrations, -searchrx, -bins 64, the weighted costs).
  • fMRIPrep’s three boundary-based calls are within 0.0113 mm of FSL’s matrices at the 95th percentile of the brain. Two are within FSL’s own variability. ds005040’s 0.0113 mm is just beyond FSL’s own 0.0086 mm (open; 0.0065 mm before the catch-up): its boundary-based path, whose costs are not FSL’s to the last bit, parts elsewhere since a perturbed row keeps its scales (specs/flirt.md §8.1 step 9). The default 12-dof registration is FSL’s, byte for byte.
  • The one case it misses is the woods warning’s place in the stream: an ill-posed registration, whose matrix is FSL’s text.
  • Cases that pass since 2026-11-04. Weighted normcorr (0.508 mm of its 0.5 mm before) passes since 2026-11-07, its matrix and -cost normcorr’s FSL’s as text: the weighted correlation’s count takes each point’s weight as the point is visited (specs/flirt.md §6.4, the specification’s answer of 2026-11-06), where the clean room had added each row’s total at the row’s end, which rounds otherwise where the weights are fractional. FSL-91’s and FSL-92’s reproduced runs and a one-voxel input, where every cost is the same, pass since 2026-11-05: FLIRT’s search keeps its starting points of equal cost in the order FSL’s sort gives them (libstdc++'s std::sort, which is not stable), where the clean room had kept the order it found them in. -cost leastsq passes since the catch-up (its three-level sums), -cost normmi and weighted normmi since 2026-11-04. pointflirt without points is refused in fix and in reproduce, naming FSL-204.

The record’s findings explain each (findings).

Random testing’s run of every program, 9,854 realistic random calls in 24 minutes, found 2,712 differences. The kinds that were the clean room’s mistakes were fixed, each checked on the run’s cases of its kind: the headers of -applyisoxfm and of a reference without coordinates, the zstd and bzip2 outputs of integer images, the forms of an unset sform or qform, the messages of BBR without a boundary point, the -dof 3 messages, the coordinate programs’ and avscale’s last digits, convert_xfm -inverse. One is a defect of FSL’s own (FSL-111, below). The registrations whose paths part on random, often ill-posed pairs stay open, with the sinc and spline kernels’ last bits.

In fMRIPrep: the boundary-based registration of a BOLD reference

fMRIPrep’s own call on one of its runs, given to FSL 6.0.7.23, to the replica and to the clean room, each in reproduce and in fix, and at one thread: both ports’ parallel steps give the same values at any thread count. Choose the panels: the T1w (the reference), the white-matter mask drawn over the BOLD reference at the initial alignment and over FSL’s result (the boundary the cost aligns the BOLD contrast to), FSL’s registered image, the replica’s and the clean room’s. FSL’s result against its initial alignment shows what the registration changed; the replica’s differences from FSL are zero.

The clean room’s matrix is 0.0113 mm from FSL’s at the 95th percentile of the brain: within its record’s 0.1 mm, and just beyond FSL’s own 0.0086 mm on the same BOLD reference perturbed by noise of half an intensity unit. Its registered image then differs from FSL’s in 42.2 % of the voxels (1 − NCC 2.6 × 10⁻⁶), the largest where a voxel at the edge of the BOLD reference’s field of view falls inside it for one matrix and outside for the other. Below the viewer, the matrices as the files hold them, and the displacement each pair gives over the brain.

Worked example

fMRIPrep's boundary-based registration of a BOLD reference

What this shows: On fMRIPrep’s own call, the replica gives FSL’s registered image, every value, and FSL’s matrix byte for byte, the same bytes as fMRIPrep’s own. The clean room’s matrix is 0.0113 mm from FSL’s at the 95th percentile of the brain: within its record’s 0.1 mm, but beyond FSL’s own variability on this input (0.00855 mm), so the comparison stays open.

The BBR node fMRIPrep 25.2.6 ran on a young adult’s BOLD reference (ds005040, slice timing, field maps): flirt -in <boldref> -ref <T1w brain> -wmseg <white matter> -init <mri_coreg> -schedule bbr.sch -dof 6 -searchcost bbr -basescale 1, given to FSL 6.0.7.23 and to the replica (larmorx-nc) in reproduce and in fix. The white-matter mask is drawn over the BOLD reference at the initial alignment and over FSL’s result: the boundary the cost aligns the BOLD contrast to. The matrices follow as text, with fMRIPrep’s own, and the displacement each pair gives over the brain. LARMORX’s clean room (Apache-2.0, written from the specification without FSL’s source) runs the same call in reproduce and in fix and at one thread; FSL runs it once more on the BOLD reference perturbed by noise of half an intensity unit, which measures FSL’s own variability, the yardstick of a registration that is not bit-identical.

Data
OpenNeuro ds005040, sub-002 (CC0-1.0): the BOLD reference and the T1w fMRIPrep computed; shown inside the T1w’s brain mask, dilated by 2 voxels
Node
fMRIPrep 25.2.6, work/fmriprep_25_2_wf/sub_002_wf/bold_task_clip_run_1_wf/bold_fit_wf/bold_reg_wf/fsl_bbr_wf/flt_bbr
Command
flirt -in sub-002_task-clip_run-1_desc-coreg_boldref.nii.gz -ref sub-002_desc-preproc_T1w_masked.nii.gz -out sub-002_task-clip_run-1_desc-coreg_boldref_flirt.nii.gz -omat sub-002_task-clip_run-1_desc-coreg_boldref_flirt.mat -basescale 1 -searchcost bbr -dof 6 -init out.mat -schedule bbr.sch -wmseg sub-002_dseg_mask.nii.gz
Programs
FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels) (its run served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 301fab7dfc56; LARMORX 0.0.1 (crate larmorx-fsl: the clean room, Apache-2.0, written from specs/flirt.md), from LARMORX at commit 80cc208fa4a3
Matrix
the replica’s file byte for byte FSL’s; fMRIPrep’s own the same bytes; BBR moved the brain by 1.79 mm on average from the initial alignment (at most 3.36 mm)
Clean room
the clean room’s matrix 0.0113 mm from FSL’s at the 95th percentile of the brain (at most 0.0136 mm); FSL’s own on a perturbed input 0.00855 mm (at most 0.00997 mm); at 1 thread the same values as at 4; in fix, its default, another matrix (the comparisons below)
Grid
191 × 288 × 288 voxels of 0.85 × 0.833 × 0.833 mm; shown: 168 × 215 × 168 around the brain
Inputds005040-sub-002 T1w (the reference)fMRIPrep's preprocessed T1w × its brain mask (-ref)
InputWhite matter over the initial alignmentfMRIPrep's white-matter mask (-wmseg) drawn over the BOLD reference at -init: the boundary BBR fits
InputWhite matter over FSL's resultthe same mask drawn over FSL 6.0.7.23's registered BOLD reference
OriginalFSL flirt 6.0.7.23the conda installation, as built (the record's run, from the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce, 4 threads · against FSL flirt 6.0.7.23
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 4 threads (the record's run) · against FSL flirt 6.0.7.23
the original against its starting point: what the registration changedFSL flirt 6.0.7.23 − initial alignment (-init)scale ±9.5e4

Bit-identical15,842,304 of 15,842,304 voxels equal

the replica against the originalreplica − FSL flirt 6.0.7.23
the clean room against the originalclean room − FSL flirt 6.0.7.23scale ±9.5e4
the clean room against itself: fix and reproduceclean room, fix − reproducescale ±9.5e4

Axial slice 79 of 168; the crosshair where FSL's result differs most from its initial alignment. 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 its starting point
FSL flirt 6.0.7.23 − initial alignment (-init)referenceNot a comparison of ports: FSL’s registered image against the same input resampled at the call’s initial matrix (fMRIPrep’s mri_coreg, converted by lta_to_fsl), the change the boundary-based optimisation made.9.7e46,805,707 of 15,842,304 (43 %)3.04e30.8030.0107
The replica against the original
replica − FSL flirt 6.0.7.23Bit-identicalThe same values, every one, on this call; the matrices too, byte for byte as text.identical: 0 of 15,842,304 values differ
The replica against itself: thread counts, fix and reproduce
replica, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: FLIRT’s parallel steps (the boundary points, the blurs) give the same values at any thread count.identical: 0 of 15,842,304 values differ
replica, fix − reproduceBit-identicalNo correction of the register applies to this call: fix and reproduce give the same values.identical: 0 of 15,842,304 values differ
The clean room against the original and the replica
clean room − FSL flirt 6.0.7.23Open differenceWritten without FSL’s source, the clean room is not bit-identical by design: its matrix is 0.0113 mm from FSL’s at the 95th percentile of the brain (at most 0.0136 mm), against FSL’s own 0.00855 mm when its input changes by noise of half an intensity unit (at most 0.00997 mm); the images differ in 42.2% of their values, by at most 0.57 of the range (1 − NCC 2.6e-06; the largest where a voxel at the edge of the input’s field of view is inside it for one matrix and outside, at the background value, for the other), the input interpolated at slightly other positions. Within the record’s threshold (0.1 mm), above FSL’s own variability on this input: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md). record7.06e46,691,126 of 15,842,304 (42.2 %)47.20.5662.6e−6
clean room − replicaOpen differenceWritten without FSL’s source, the clean room is not bit-identical by design: its matrix is 0.0113 mm from the replica’s (FSL’s bytes) at the 95th percentile of the brain (at most 0.0136 mm), against FSL’s own 0.00855 mm when its input changes by noise of half an intensity unit (at most 0.00997 mm); the images differ in 42.2% of their values, by at most 0.57 of the range (1 − NCC 2.6e-06; the largest where a voxel at the edge of the input’s field of view is inside it for one matrix and outside, at the background value, for the other), the input interpolated at slightly other positions. Within the record’s threshold (0.1 mm), above FSL’s own variability on this input: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md). record7.06e46,691,126 of 15,842,304 (42.2 %)47.20.5662.6e−6
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel steps (the boundary points’ terms, the correlation ratio’s bins, the search’s grid points) are summed in a fixed order, so the values are the same at any thread count. recordidentical: 0 of 15,842,304 values differ
clean room, fix − reproduceDeliberate correctionGLIBC-07, a correction by default: the default computes sinf, cosf, atan2f, cos, exp and logf correctly rounded on every platform, where reproduce computes them as glibc 2.35 does, as FSL’s build (LARMORX’s glibc functions, since 2026-11-04); FLIRT names it nowhere: an iterative fit cannot tell whether its result moved with those last bits (as decided on 2026-11-04). Here the two modes’ matrices part by 0.00251 mm at the 95th percentile of the brain (at most 0.00252 mm), against FSL’s own 0.00855 mm on a perturbed input: the fit’s sensitivity to the C library’s last bits. record GLIBC-072.11e46,689,553 of 15,842,304 (42.2 %)8.610.1698.65e−8

Δ is A − B, computed in float64 over all 15,842,304 voxels of the outputs, not only those shown. The records’ thresholds: every file FSL writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages (the replica); FSL’s exit status and messages, the numbers in them read as numbers, the matrix FSL’s text or within 0.1 mm of it at the 95th percentile of the reference’s brain (0.5 mm for registrations other than the boundary-based and 6-dof ones), and the image from its own matrix with a correlation of at least 0.999 with FSL’s inside the reference’s mask, bit-identical from the same matrix (the clean room, specs/flirt.md §13).

The matrices as the files hold them, both ports against the original: FSL flirt 6.0.7.23 (input millimetres to reference millimetres, FLIRT’s convention)

FSL flirt 6.0.7.23replica, reproducereplica, fixclean room, reproduceclean room, fixfMRIPrep's output (FSL in its image)the initial matrix (-init)
row 10.9997946565 0.009039069054 -0.01813961096 -14.930365920.9997946565 0.009039069054 -0.01813961096 -14.930365920.9997946565 0.009039069054 -0.01813961096 -14.930365920.9997960418 0.008978942795 -0.01809395406 -14.928694790.9997961128 0.008979152222 -0.01809328011 -14.928762860.9997946565 0.009039069054 -0.01813961096 -14.930365920.99973863 0.00760034 -0.02156423 -14.54669762
row 2-0.008074701193 0.9985856646 0.05255029316 14.28776341-0.008074701193 0.9985856646 0.05255029316 14.28776341-0.008074701193 0.9985856646 0.05255029316 14.28776341-0.00801558061 0.9985819946 0.0526289076 14.2819025-0.008015822649 0.9985820188 0.0526290324 14.27941674-0.008074701193 0.9985856646 0.05255029316 14.28776341-0.00601620 0.99734318 0.07259799 12.04777431
row 30.01858896419 -0.05239302043 0.9984534561 81.720782730.01858896419 -0.05239302043 0.9984534561 81.720782730.01858896419 -0.05239302043 0.9984534561 81.720782730.01854085161 -0.05247313056 0.9984501524 81.729396160.01854018945 -0.05247325387 0.9984501581 81.729425770.01858896419 -0.05239302043 0.9984534561 81.720782730.02205871 -0.07244927 0.99712807 81.83596802
row 40 0 0 10 0 0 10 0 0 10 0 0 10 0 0 10 0 0 10.00000000 0.00000000 0.00000000 1.00000000
the file, byte for bytethe referenceidenticalidenticaldiffersdiffersidentical–

Each matrix file as written (-omat), its numbers as text; fMRIPrep’s own is the node’s output in the run’s working directory (FSL in fMRIPrep 25.2.6’s image), the initial matrix the call’s -init (mri_coreg’s, converted by lta_to_fsl). identical: the same bytes as FSL 6.0.7.23’s file. The clean room’s matrices are its own result, written as FSL writes them.

The displacement between two matrices over the reference’s brain, both ports against the original: FSL flirt 6.0.7.23 (mm)

Matriceslargestmean95th percentileFSL's RMS deviation (80 mm)
replica − FSL flirt 6.0.7.230000
replica, fix − FSL flirt 6.0.7.230000
clean room − FSL flirt 6.0.7.230.01360.006370.01130.0157
clean room, fix − FSL flirt 6.0.7.230.01280.005390.01010.0177
FSL flirt 6.0.7.23 on a perturbed input − FSL flirt 6.0.7.23 (its own variability)0.009970.005340.008550.00852
fMRIPrep’s output − FSL flirt 6.0.7.230000
FSL flirt 6.0.7.23 − the initial matrix (what the registration changed)3.361.792.834.08

For every one of the 2,446,665 voxels of the reference’s brain mask (its voxels above zero), the distance in millimetres between the two points of the input the matrices map onto it, computed in float64 from the matrices as the files hold them; the last column is FSL’s own measure (rmsdiff’s: the RMS deviation over a sphere of radius 80 mm about the origin).

Provenance of this example
  • Data: OpenNeuro ds005040, sub-002, CC0-1.0
  • Inputs: -in (sha256 db30ecbef0f5769a…); -ref (sha256 ade613b412e42a59…); -wmseg (sha256 5731cb930d625adb…); -init (sha256 a0b1a1fbd5ee6f32…)
  • Runs: figures cache entry fsl-flirt/fmriprep-ds005040-sub-002-noreconall, computed 2026-11-01T16:10:32+00:00 with larmorx 301fab7dfc56; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Inputs: -in (sha256 db30ecbef0f5769a…); -ref (sha256 ade613b412e42a59…); -wmseg (sha256 5731cb930d625adb…); -init (sha256 a0b1a1fbd5ee6f32…)
  • Runs: figures cache entry fsl-flirt/fmriprep-ds005040-sub-002-noreconall-original, computed 2026-11-02T21:04:08+00:00; 3 of its 4 runs made again since, last on 2026-11-07T02:05:04+00:00, with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); “larmorx 0.0.1 (crate larmorx-fsl: the clean-room original, Apache-2.0, written from specs/flirt.md)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-flirt.json, exported by python -m larmorx_validation evidence fsl-flirt.

A default registration

FLIRT without options beyond its inputs: 12 degrees of freedom, the correlation ratio, the search over rotations and the default schedule, registering the same BOLD reference to the T1w at half resolution (the record’s case real/default). Since 2026-11-04 the clean room’s matrix and registered image are FSL’s, byte for byte. Its default, fix, still computes the C library’s functions correctly rounded, and its matrix is 0.033 mm from FSL’s at the 95th percentile (the comparison clean room, fix − reproduce); FSL’s own on a perturbed input is 0.005 mm.

Worked example

A default registration: 12 degrees of freedom, the correlation ratio

What this shows: On FLIRT’s default registration, the replica and the clean room both give FSL’s matrix and image, every value. In fix, its default, the clean room computes the C library’s functions correctly rounded (GLIBC-07), and its matrix moves by 0.0331 mm at the 95th percentile of the brain, more than FSL’s own 0.0048 mm on a perturbed input.

The record’s case real/default: FLIRT’s default (12 degrees of freedom, the correlation ratio, its search over rotations and its multi-resolution schedule) registering ds005040’s BOLD reference to the subject’s T1w brain at half resolution, by FSL 6.0.7.23 and the replica in reproduce and in fix; the white-matter mask drawn over FSL’s result shows the fit. LARMORX’s clean room runs the same call in fix and in reproduce and at one thread, and FSL once more on the BOLD reference perturbed by noise of half an intensity unit (its own variability).

Data
OpenNeuro ds005040, sub-002 (CC0-1.0): the record’s inputs derived from fMRIPrep’s run (oracles/inputs/fsl-flirt-replica/); shown inside the T1w’s brain mask, dilated by 2 voxels
Case
real/default of the record (fsl-flirt-replica)
Command
flirt -in epi.nii -ref t1.nii -out out -omat out.mat
Programs
FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels) (its run served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 301fab7dfc56; LARMORX 0.0.1 (crate larmorx-fsl: the clean room, Apache-2.0, written from specs/flirt.md), from LARMORX at commit 80cc208fa4a3
Clean room
the clean room’s matrix 0 mm from FSL’s at the 95th percentile of the brain (at most 0 mm); FSL’s own on a perturbed input 0.0048 mm (at most 0.00606 mm)
Grid
96 × 144 × 144 voxels of 1.7 × 1.67 × 1.67 mm; shown: 88 × 112 × 88 around the brain
Inputds005040 sub-002 T1w (the reference)fMRIPrep's T1w × its brain mask, every second voxel (1.7 mm; -ref)
InputWhite matter over FSL's resultfMRIPrep's white-matter mask (every second voxel) drawn over FSL 6.0.7.23's registered BOLD reference: a check of the fit
OriginalFSL flirt 6.0.7.23the conda installation, as built (the record's run, from the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce, 4 threads · against FSL flirt 6.0.7.23
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 4 threads (the record's run) · against FSL flirt 6.0.7.23

Bit-identical1,990,656 of 1,990,656 voxels equal

the replica against the originalreplica − FSL flirt 6.0.7.23

Bit-identical1,990,656 of 1,990,656 voxels equal

the replica against itself: fix and reproducereplica, fix − reproduce

Bit-identical1,990,656 of 1,990,656 voxels equal

the clean room against the originalclean room − FSL flirt 6.0.7.23
the clean room against itself: fix and reproduceclean room, fix − reproducescale ±5.09e4

Axial slice 43 of 88; the crosshair at the centre of the brain. 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 replica against the original
replica − FSL flirt 6.0.7.23Bit-identicalThe same values, every one, on this call; the matrices too, byte for byte as text.identical: 0 of 1,990,656 values differ
The replica against itself: thread counts, fix and reproduce
replica, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: FLIRT’s parallel steps (the boundary points, the blurs) give the same values at any thread count.identical: 0 of 1,990,656 values differ
replica, fix − reproduceBit-identicalNo correction of the register applies to this call: fix and reproduce give the same values.identical: 0 of 1,990,656 values differ
The clean room against the original and the replica
clean room − FSL flirt 6.0.7.23Bit-identicalThe clean room gives FSL’s values, every one. recordidentical: 0 of 1,990,656 values differ
clean room − replicaBit-identicalThe clean room gives the replica’s (FSL’s bytes) values, every one. recordidentical: 0 of 1,990,656 values differ
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel steps (the boundary points’ terms, the correlation ratio’s bins, the search’s grid points) are summed in a fixed order, so the values are the same at any thread count. recordidentical: 0 of 1,990,656 values differ
clean room, fix − reproduceDeliberate correctionGLIBC-07, a correction by default: the default computes sinf, cosf, atan2f, cos, exp and logf correctly rounded on every platform, where reproduce computes them as glibc 2.35 does, as FSL’s build (LARMORX’s glibc functions, since 2026-11-04); FLIRT names it nowhere: an iterative fit cannot tell whether its result moved with those last bits (as decided on 2026-11-04). Here the two modes’ matrices part by 0.0331 mm at the 95th percentile of the brain (at most 0.042 mm), against FSL’s own 0.0048 mm on a perturbed input: the fit’s sensitivity to the C library’s last bits. record GLIBC-075.09e4866,816 of 1,990,656 (43.5 %)47.90.4312.59e−6

Δ is A − B, computed in float64 over all 1,990,656 voxels of the outputs, not only those shown. The records’ thresholds: every file FSL writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages (the replica); FSL’s exit status and messages, the numbers in them read as numbers, the matrix FSL’s text or within 0.1 mm of it at the 95th percentile of the reference’s brain (0.5 mm for registrations other than the boundary-based and 6-dof ones), and the image from its own matrix with a correlation of at least 0.999 with FSL’s inside the reference’s mask, bit-identical from the same matrix (the clean room, specs/flirt.md §13).

The matrices as the files hold them, both ports against the original: FSL flirt 6.0.7.23 (input millimetres to reference millimetres, FLIRT’s convention)

FSL flirt 6.0.7.23replica, reproducereplica, fixclean room, reproduceclean room, fix
row 11.016059013 0.01501063208 -0.0301453908 -16.246784331.016059013 0.01501063208 -0.0301453908 -16.246784331.016059013 0.01501063208 -0.0301453908 -16.246784331.016059013 0.01501063208 -0.0301453908 -16.246784331.016061684 0.01472990717 -0.03061194368 -16.19458306
row 2-0.01835764209 1.002918871 0.08292936251 11.46911575-0.01835764209 1.002918871 0.08292936251 11.46911575-0.01835764209 1.002918871 0.08292936251 11.46911575-0.01835764209 1.002918871 0.08292936251 11.46911575-0.01834992389 1.002943648 0.08271480958 11.47912664
row 30.03345222159 -0.08911393648 1.019501435 79.669905390.03345222159 -0.08911393648 1.019501435 79.669905390.03345222159 -0.08911393648 1.019501435 79.669905390.03345222159 -0.08911393648 1.019501435 79.669905390.03347213586 -0.0891252173 1.019498175 79.66912412
row 40 0 0 10 0 0 10 0 0 10 0 0 10 0 0 1
the file, byte for bytethe referenceidenticalidenticalidenticaldiffers

Each matrix file as written (-omat), its numbers as text. identical: the same bytes as FSL 6.0.7.23’s file. The clean room’s matrices are its own result, written as FSL writes them.

The displacement between two matrices over the reference’s brain, both ports against the original: FSL flirt 6.0.7.23 (mm)

Matriceslargestmean95th percentileFSL's RMS deviation (80 mm)
replica − FSL flirt 6.0.7.230000
replica, fix − FSL flirt 6.0.7.230000
clean room − FSL flirt 6.0.7.230000
clean room, fix − FSL flirt 6.0.7.230.0420.01620.03310.0966
FSL flirt 6.0.7.23 on a perturbed input − FSL flirt 6.0.7.23 (its own variability)0.006060.002430.00480.0124

For every one of the 305,805 voxels of the reference’s brain mask (its voxels above zero), the distance in millimetres between the two points of the input the matrices map onto it, computed in float64 from the matrices as the files hold them; the last column is FSL’s own measure (rmsdiff’s: the RMS deviation over a sphere of radius 80 mm about the origin).

Provenance of this example
  • Data: OpenNeuro ds005040, sub-002, CC0-1.0
  • Inputs: epi.nii (sha256 fefefce4f108f492…); t1.nii (sha256 5a3725fbb21ceccc…)
  • Runs: figures cache entry fsl-flirt/real-default, computed 2026-11-01T16:13:50+00:00 with larmorx 301fab7dfc56; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Inputs: epi.nii (sha256 fefefce4f108f492…); t1.nii (sha256 5a3725fbb21ceccc…)
  • Runs: figures cache entry fsl-flirt/real-default-original, computed 2026-11-02T21:06:24+00:00; 3 of its 4 runs made again since, last on 2026-11-07T02:06:24+00:00, with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); “larmorx 0.0.1 (crate larmorx-fsl: the clean-room original, Apache-2.0, written from specs/flirt.md)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-flirt.json, exported by python -m larmorx_validation evidence fsl-flirt.

-applyxfm on a raw BOLD series

A given matrix applied to four volumes of the raw BOLD series with windowed-sinc interpolation, onto the T1w’s grid (the record’s case real/applyxfm-4d): the resampling part of FLIRT, with its blur of the input where the output grid is coarser. From the same matrix the clean room’s values are FSL’s, every one, since 2026-11-04. What the specification had left open was the kernel’s sinf and cos, which its reproduce now computes as glibc 2.35 does. Its default, fix, computes them correctly rounded and differs in 1.6 % of the values, by at most 2.6 × 10⁻⁷ of the image’s range, as the clean room did in both modes before.

Worked example

-applyxfm on a raw BOLD series, sinc interpolation

What this shows: From the same matrix, the replica and the clean room both give FSL’s image, every value. In fix, its default, the clean room computes the C library’s functions correctly rounded (GLIBC-07), and 1.6% of the values change.

The record’s case real/applyxfm-4d: four volumes of ds005040’s raw BOLD series resampled onto the T1w’s grid (every second voxel) through fMRIPrep’s initial matrix, with -applyxfm -interp sinc, by FSL 6.0.7.23 and the replica in reproduce and in fix; choose the volume below the panels. LARMORX’s clean room applies the same matrix in fix and in reproduce and at one thread.

Data
OpenNeuro ds005040, sub-002 (CC0-1.0): the record’s inputs derived from fMRIPrep’s run (oracles/inputs/fsl-flirt-replica/); shown inside the T1w’s brain mask, dilated by 2 voxels
Case
real/applyxfm-4d of the record (fsl-flirt-replica)
Command
flirt -in bold4.nii -ref t1.nii -init init.mat -applyxfm -interp sinc -out out
Programs
FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels) (its run served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 301fab7dfc56; LARMORX 0.0.1 (crate larmorx-fsl: the clean room, Apache-2.0, written from specs/flirt.md), from LARMORX at commit 80cc208fa4a3
Grid
96 × 144 × 144 voxels of 1.7 × 1.67 × 1.67 mm; shown: 88 × 112 × 88 around the brain
Inputds005040 sub-002 T1w (the reference)fMRIPrep's T1w × its brain mask, every second voxel (1.7 mm; -ref)
OriginalFSL flirt 6.0.7.23the conda installation, as built (the record's run, from the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce, 4 threads · against FSL flirt 6.0.7.23
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 4 threads (the record's run) · against FSL flirt 6.0.7.23

Bit-identical7,962,624 of 7,962,624 voxels equal

the replica against the originalreplica − FSL flirt 6.0.7.23

Bit-identical7,962,624 of 7,962,624 voxels equal

the replica against itself: fix and reproducereplica, fix − reproduce

Bit-identical7,962,624 of 7,962,624 voxels equal

the clean room against the originalclean room − FSL flirt 6.0.7.23
the clean room against itself: fix and reproduceclean room, fix − reproducescale ±0.0469

Axial slice 43 of 88; the crosshair at the centre of the brain. 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 replica against the original
replica − FSL flirt 6.0.7.23Bit-identicalThe same values, every one, on this call; the matrices too, byte for byte as text.identical: 0 of 7,962,624 values differ
The replica against itself: thread counts, fix and reproduce
replica, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: FLIRT’s parallel steps (the boundary points, the blurs) give the same values at any thread count.identical: 0 of 7,962,624 values differ
replica, fix − reproduceBit-identicalNo correction of the register applies to this call: fix and reproduce give the same values.identical: 0 of 7,962,624 values differ
The clean room against the original and the replica
clean room − FSL flirt 6.0.7.23Bit-identicalThe clean room gives FSL’s values, every one. recordidentical: 0 of 7,962,624 values differ
clean room − replicaBit-identicalThe clean room gives the replica’s (FSL’s bytes) values, every one. recordidentical: 0 of 7,962,624 values differ
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel steps (the boundary points’ terms, the correlation ratio’s bins, the search’s grid points) are summed in a fixed order, so the values are the same at any thread count. recordidentical: 0 of 7,962,624 values differ
clean room, fix − reproduceDeliberate correctionGLIBC-07, a correction by default: the default computes sinf, cosf, atan2f, cos, exp and logf correctly rounded on every platform, where reproduce computes them as glibc 2.35 does, as FSL’s build (LARMORX’s glibc functions, since 2026-11-04); FLIRT names it nowhere: an iterative fit cannot tell whether its result moved with those last bits (as decided on 2026-11-04). Here the two modes’ images differ in 1.6% of their values. record GLIBC-070.0469128,674 of 7,962,624 (1.62 %)0.000572.59e−73.17e−16

Δ is A − B, computed in float64 over all 7,962,624 voxels of the outputs, not only those shown. The records’ thresholds: every file FSL writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages (the replica); FSL’s exit status and messages, the numbers in them read as numbers, the matrix FSL’s text or within 0.1 mm of it at the 95th percentile of the reference’s brain (0.5 mm for registrations other than the boundary-based and 6-dof ones), and the image from its own matrix with a correlation of at least 0.999 with FSL’s inside the reference’s mask, bit-identical from the same matrix (the clean room, specs/flirt.md §13).

Provenance of this example
  • Data: OpenNeuro ds005040, sub-002, CC0-1.0
  • Inputs: bold4.nii (sha256 92acf4858e24217c…); t1.nii (sha256 5a3725fbb21ceccc…); init.mat (sha256 a0b1a1fbd5ee6f32…)
  • Runs: figures cache entry fsl-flirt/real-applyxfm-4d, computed 2026-11-01T16:14:02+00:00 with larmorx 301fab7dfc56; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Inputs: bold4.nii (sha256 92acf4858e24217c…); t1.nii (sha256 5a3725fbb21ceccc…); init.mat (sha256 a0b1a1fbd5ee6f32…)
  • Runs: figures cache entry fsl-flirt/real-applyxfm-4d-original, computed 2026-11-02T21:06:27+00:00; its runs made again since, last on 2026-11-07T02:06:27+00:00, with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); “larmorx 0.0.1 (crate larmorx-fsl: the clean-room original, Apache-2.0, written from specs/flirt.md)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-flirt.json, exported by python -m larmorx_validation evidence fsl-flirt.

On the data panel: an 82-year-old and a three-year-old

fMRIPrep’s boundary-based call on two runs of the data panel: the first echo of an 82-year-old’s multi-echo BOLD run (ds003592) and a three-year-old’s BOLD reference (ds000228, a small head and an oblique field of view). As data-panel cases beside the examples above, they are stored as three full-resolution slices, and the crosshair moves through them. On both, the clean room’s matrix is within FSL’s own variability: 0.0002 mm against 0.0031 mm, and under 0.00001 mm against 0.091 mm.

Worked example

Data panel: an 82-year-old's BOLD reference

What this shows: On the data panel’s 82-year-old, the replica gives FSL’s registered image, every value, and FSL’s matrix byte for byte, the same bytes as fMRIPrep’s own. The clean room’s matrix is 0.000212 mm from FSL’s at the 95th percentile of the brain, within FSL’s own variability on this input (0.00314 mm).

The BBR node fMRIPrep 25.2.6 ran on an 82-year-old’s first echo (ds003592, multi-echo): flirt -in <boldref> -ref <T1w brain> -wmseg <white matter> -init <mri_coreg> -schedule bbr.sch -dof 6 -searchcost bbr -basescale 1, given to FSL 6.0.7.23 and to the replica (larmorx-nc) in reproduce and in fix. The white-matter mask is drawn over the BOLD reference at the initial alignment and over FSL’s result: the boundary the cost aligns the BOLD contrast to. The matrices follow as text, with fMRIPrep’s own, and the displacement each pair gives over the brain. LARMORX’s clean room (Apache-2.0, written from the specification without FSL’s source) runs the same call in reproduce and in fix and at one thread; FSL runs it once more on the BOLD reference perturbed by noise of half an intensity unit, which measures FSL’s own variability, the yardstick of a registration that is not bit-identical. As a data-panel case beside the first example, it is stored as three full-resolution slices.

Data
OpenNeuro ds003592, sub-280 (CC0-1.0): the BOLD reference and the T1w fMRIPrep computed; shown inside the T1w’s brain mask, dilated by 2 voxels
Node
fMRIPrep 25.2.6, work/fmriprep_25_2_wf/sub_280_ses_1_wf/bold_ses_1_task_rest_echo_1_wf/bold_fit_wf/bold_reg_wf/fsl_bbr_wf/flt_bbr
Command
flirt -in sub-280_ses-1_task-rest_desc-coreg_boldref.nii.gz -ref sub-280_ses-1_desc-preproc_T1w_masked.nii.gz -out sub-280_ses-1_task-rest_desc-coreg_boldref_flirt.nii.gz -omat sub-280_ses-1_task-rest_desc-coreg_boldref_flirt.mat -basescale 1 -searchcost bbr -dof 6 -init out.mat -schedule bbr.sch -wmseg sub-280_ses-1_dseg_mask.nii.gz
Programs
FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels) (its run served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 301fab7dfc56; LARMORX 0.0.1 (crate larmorx-fsl: the clean room, Apache-2.0, written from specs/flirt.md), from LARMORX at commit 80cc208fa4a3
Matrix
the replica’s file byte for byte FSL’s; fMRIPrep’s own the same bytes; BBR moved the brain by 1.32 mm on average from the initial alignment (at most 1.91 mm)
Clean room
the clean room’s matrix 0.000212 mm from FSL’s at the 95th percentile of the brain (at most 0.000267 mm); FSL’s own on a perturbed input 0.00314 mm (at most 0.00328 mm); at 1 thread the same values as at 4; in fix, its default, another matrix (the comparisons below)
Grid
192 × 256 × 256 voxels of 1 × 1 × 1 mm; shown: 151 × 177 × 148 around the brain
Inputds003592-sub-280 T1w (the reference)fMRIPrep's preprocessed T1w × its brain mask (-ref)
InputWhite matter over the initial alignmentfMRIPrep's white-matter mask (-wmseg) drawn over the BOLD reference at -init: the boundary BBR fits
InputWhite matter over FSL's resultthe same mask drawn over FSL 6.0.7.23's registered BOLD reference
OriginalFSL flirt 6.0.7.23the conda installation, as built (the record's run, from the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce, 4 threads · against FSL flirt 6.0.7.23
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 4 threads (the record's run) · against FSL flirt 6.0.7.23
the original against its starting point: what the registration changedFSL flirt 6.0.7.23 − initial alignment (-init)scale ±1.05e3

Bit-identical12,582,912 of 12,582,912 voxels equal

the replica against the originalreplica − FSL flirt 6.0.7.23
the clean room against the originalclean room − FSL flirt 6.0.7.23scale ±1.05e3
the clean room against itself: fix and reproduceclean room, fix − reproducescale ±1.05e3

Axial slice 13 of 148; the crosshair where FSL's result differs most from its initial alignment; the three slices through it are stored at full resolution. The interactive viewer needs JavaScript; the metrics below are complete without it.

Stored for this example: three full-resolution slices, one per plane, through display voxel (75, 83, 13): where the main comparison differs most, or the field's centre where nothing differs; every metric is computed over every voxel of the files.

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 its starting point
FSL flirt 6.0.7.23 − initial alignment (-init)referenceNot a comparison of ports: FSL’s registered image against the same input resampled at the call’s initial matrix (fMRIPrep’s mri_coreg, converted by lta_to_fsl), the change the boundary-based optimisation made.1.1e35,204,755 of 12,582,912 (41.4 %)26.40.8360.00589
The replica against the original
replica − FSL flirt 6.0.7.23Bit-identicalThe same values, every one, on this call; the matrices too, byte for byte as text.identical: 0 of 12,582,912 values differ
The replica against itself: thread counts, fix and reproduce
replica, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: FLIRT’s parallel steps (the boundary points, the blurs) give the same values at any thread count.identical: 0 of 12,582,912 values differ
replica, fix − reproduceBit-identicalNo correction of the register applies to this call: fix and reproduce give the same values.identical: 0 of 12,582,912 values differ
The clean room against the original and the replica
clean room − FSL flirt 6.0.7.23Within the original's own variabilityWritten without FSL’s source, the clean room is not bit-identical by design: its matrix is 0.000212 mm from FSL’s at the 95th percentile of the brain (at most 0.000267 mm), against FSL’s own 0.00314 mm when its input changes by noise of half an intensity unit (at most 0.00328 mm); the images differ in 40.7% of their values, by at most 0.22 of the range (1 − NCC 1.2e-07; the largest where a voxel at the edge of the input’s field of view is inside it for one matrix and outside, at the background value, for the other), the input interpolated at slightly other positions. The difference is within FSL’s own variability. record2895,127,418 of 12,582,912 (40.7 %)0.1180.2181.17e−7
clean room − replicaWithin the original's own variabilityWritten without FSL’s source, the clean room is not bit-identical by design: its matrix is 0.000212 mm from the replica’s (FSL’s bytes) at the 95th percentile of the brain (at most 0.000267 mm), against FSL’s own 0.00314 mm when its input changes by noise of half an intensity unit (at most 0.00328 mm); the images differ in 40.7% of their values, by at most 0.22 of the range (1 − NCC 1.2e-07; the largest where a voxel at the edge of the input’s field of view is inside it for one matrix and outside, at the background value, for the other), the input interpolated at slightly other positions. The difference is within FSL’s own variability. record2895,127,418 of 12,582,912 (40.7 %)0.1180.2181.17e−7
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel steps (the boundary points’ terms, the correlation ratio’s bins, the search’s grid points) are summed in a fixed order, so the values are the same at any thread count. recordidentical: 0 of 12,582,912 values differ
clean room, fix − reproduceDeliberate correctionGLIBC-07, a correction by default: the default computes sinf, cosf, atan2f, cos, exp and logf correctly rounded on every platform, where reproduce computes them as glibc 2.35 does, as FSL’s build (LARMORX’s glibc functions, since 2026-11-04); FLIRT names it nowhere: an iterative fit cannot tell whether its result moved with those last bits (as decided on 2026-11-04). Here the two modes’ matrices part by 0.00136 mm at the 95th percentile of the brain (at most 0.00174 mm), against FSL’s own 0.00314 mm on a perturbed input: the fit’s sensitivity to the C library’s last bits. record GLIBC-077415,140,499 of 12,582,912 (40.9 %)0.2490.565.23e−7

Δ is A − B, computed in float64 over all 12,582,912 voxels of the outputs, not only those shown. The records’ thresholds: every file FSL writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages (the replica); FSL’s exit status and messages, the numbers in them read as numbers, the matrix FSL’s text or within 0.1 mm of it at the 95th percentile of the reference’s brain (0.5 mm for registrations other than the boundary-based and 6-dof ones), and the image from its own matrix with a correlation of at least 0.999 with FSL’s inside the reference’s mask, bit-identical from the same matrix (the clean room, specs/flirt.md §13).

The matrices as the files hold them, both ports against the original: FSL flirt 6.0.7.23 (input millimetres to reference millimetres, FLIRT’s convention)

FSL flirt 6.0.7.23replica, reproducereplica, fixclean room, reproduceclean room, fixfMRIPrep's output (FSL in its image)the initial matrix (-init)
row 10.9999789461 0.000344049261 0.006523709465 -13.690186240.9999789461 0.000344049261 0.006523709465 -13.690186240.9999789461 0.000344049261 0.006523709465 -13.690186240.9999788404 0.0003442933607 0.006523883862 -13.69021080.9999788391 0.0003420573105 0.006533222366 -13.690714660.9999789461 0.000344049261 0.006523709465 -13.690186240.99999636 0.00144488 0.00235537 -13.52341270
row 20.0009319920387 0.9808839998 -0.1945906627 38.290287250.0009319920387 0.9808839998 -0.1945906627 38.290287250.0009319920387 0.9808839998 -0.1945906627 38.290287250.000931801665 0.980883597 -0.1945929454 38.29048350.0009357211036 0.9808863349 -0.1945791354 38.288591960.0009319920387 0.9808839998 -0.1945906627 38.29028725-0.00098047 0.98246926 -0.18642190 36.97615433
row 3-0.006465954659 0.1945926059 0.9808630277 55.09992093-0.006465954659 0.1945926059 0.9808630277 55.09992093-0.006465954659 0.1945926059 0.9808630277 55.09992093-0.006466171845 0.1945948849 0.9808626106 55.09981791-0.006474909406 0.1945810904 0.9808652498 55.10196123-0.006465954659 0.1945926059 0.9808630277 55.09992093-0.00258344 0.18641889 0.98246717 54.38763809
row 40 0 0 10 0 0 10 0 0 10 0 0 10 0 0 10 0 0 10.00000000 0.00000000 0.00000000 1.00000000
the file, byte for bytethe referenceidenticalidenticaldiffersdiffersidentical–

Each matrix file as written (-omat), its numbers as text; fMRIPrep’s own is the node’s output in the run’s working directory (FSL in fMRIPrep 25.2.6’s image), the initial matrix the call’s -init (mri_coreg’s, converted by lta_to_fsl). identical: the same bytes as FSL 6.0.7.23’s file. The clean room’s matrices are its own result, written as FSL writes them.

The displacement between two matrices over the reference’s brain, both ports against the original: FSL flirt 6.0.7.23 (mm)

Matriceslargestmean95th percentileFSL's RMS deviation (80 mm)
replica − FSL flirt 6.0.7.230000
replica, fix − FSL flirt 6.0.7.230000
clean room − FSL flirt 6.0.7.230.0002670.0001110.0002120.000396
clean room, fix − FSL flirt 6.0.7.230.001620.0007090.001280.0035
FSL flirt 6.0.7.23 on a perturbed input − FSL flirt 6.0.7.23 (its own variability)0.003280.002680.003140.00406
fMRIPrep’s output − FSL flirt 6.0.7.230000
FSL flirt 6.0.7.23 − the initial matrix (what the registration changed)1.911.321.711.91

For every one of the 1,471,297 voxels of the reference’s brain mask (its voxels above zero), the distance in millimetres between the two points of the input the matrices map onto it, computed in float64 from the matrices as the files hold them; the last column is FSL’s own measure (rmsdiff’s: the RMS deviation over a sphere of radius 80 mm about the origin).

Provenance of this example
  • Data: OpenNeuro ds003592, sub-280, CC0-1.0
  • Inputs: -in (sha256 e53f3282faee1ee9…); -ref (sha256 701828f56fca12a5…); -wmseg (sha256 be9a746f9714eab6…); -init (sha256 36960edc109a5abc…)
  • Runs: figures cache entry fsl-flirt/fmriprep-ds003592-sub-280-noreconall, computed 2026-11-01T16:12:39+00:00 with larmorx 301fab7dfc56; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Inputs: -in (sha256 e53f3282faee1ee9…); -ref (sha256 701828f56fca12a5…); -wmseg (sha256 be9a746f9714eab6…); -init (sha256 36960edc109a5abc…)
  • Runs: figures cache entry fsl-flirt/fmriprep-ds003592-sub-280-noreconall-original, computed 2026-11-02T21:05:35+00:00; 3 of its 4 runs made again since, last on 2026-11-07T02:05:57+00:00, with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); “larmorx 0.0.1 (crate larmorx-fsl: the clean-room original, Apache-2.0, written from specs/flirt.md)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-flirt.json, exported by python -m larmorx_validation evidence fsl-flirt.

Worked example

Data panel: a three-year-old's BOLD reference

What this shows: On the data panel’s three-year-old, the replica gives FSL’s registered image, every value, and FSL’s matrix byte for byte, the same bytes as fMRIPrep’s own. The clean room’s matrix is 4.87 × 10⁻⁶ mm from FSL’s at the 95th percentile of the brain, within FSL’s own variability on this input (0.0907 mm).

The BBR node fMRIPrep 25.2.6 ran on a three-year-old’s BOLD reference (ds000228): flirt -in <boldref> -ref <T1w brain> -wmseg <white matter> -init <mri_coreg> -schedule bbr.sch -dof 6 -searchcost bbr -basescale 1, given to FSL 6.0.7.23 and to the replica (larmorx-nc) in reproduce and in fix. The white-matter mask is drawn over the BOLD reference at the initial alignment and over FSL’s result: the boundary the cost aligns the BOLD contrast to. The matrices follow as text, with fMRIPrep’s own, and the displacement each pair gives over the brain. LARMORX’s clean room (Apache-2.0, written from the specification without FSL’s source) runs the same call in reproduce and in fix and at one thread; FSL runs it once more on the BOLD reference perturbed by noise of half an intensity unit, which measures FSL’s own variability, the yardstick of a registration that is not bit-identical. As a data-panel case beside the first example, it is stored as three full-resolution slices.

Data
OpenNeuro ds000228, sub-pixar008 (CC0-1.0): the BOLD reference and the T1w fMRIPrep computed; shown inside the T1w’s brain mask, dilated by 2 voxels
Node
fMRIPrep 25.2.6, work/fmriprep_25_2_wf/sub_pixar008_wf/bold_task_pixar_wf/bold_fit_wf/bold_reg_wf/fsl_bbr_wf/flt_bbr
Command
flirt -in sub-pixar008_task-pixar_desc-coreg_boldref.nii.gz -ref sub-pixar008_desc-preproc_T1w_masked.nii.gz -out sub-pixar008_task-pixar_desc-coreg_boldref_flirt.nii.gz -omat sub-pixar008_task-pixar_desc-coreg_boldref_flirt.mat -basescale 1 -searchcost bbr -dof 6 -init out.mat -schedule bbr.sch -wmseg sub-pixar008_dseg_mask.nii.gz
Programs
FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels) (its run served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 301fab7dfc56; LARMORX 0.0.1 (crate larmorx-fsl: the clean room, Apache-2.0, written from specs/flirt.md), from LARMORX at commit 80cc208fa4a3
Matrix
the replica’s file byte for byte FSL’s; fMRIPrep’s own the same bytes; BBR moved the brain by 0.77 mm on average from the initial alignment (at most 1.06 mm)
Clean room
the clean room’s matrix 4.87e-06 mm from FSL’s at the 95th percentile of the brain (at most 5.97e-06 mm); FSL’s own on a perturbed input 0.0907 mm (at most 0.113 mm); at 1 thread the same values as at 4; in fix, its default, another matrix (the comparisons below)
Grid
176 × 192 × 192 voxels of 1 × 1 × 1 mm; shown: 140 × 166 × 124 around the brain
Inputds000228-sub-pixar008 T1w (the reference)fMRIPrep's preprocessed T1w × its brain mask (-ref)
InputWhite matter over the initial alignmentfMRIPrep's white-matter mask (-wmseg) drawn over the BOLD reference at -init: the boundary BBR fits
InputWhite matter over FSL's resultthe same mask drawn over FSL 6.0.7.23's registered BOLD reference
OriginalFSL flirt 6.0.7.23the conda installation, as built (the record's run, from the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce, 4 threads · against FSL flirt 6.0.7.23
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 4 threads (the record's run) · against FSL flirt 6.0.7.23
the original against its starting point: what the registration changedFSL flirt 6.0.7.23 − initial alignment (-init)scale ±2.44e3

Bit-identical6,488,064 of 6,488,064 voxels equal

the replica against the originalreplica − FSL flirt 6.0.7.23
the clean room against the originalclean room − FSL flirt 6.0.7.23scale ±2.44e3
the clean room against itself: fix and reproduceclean room, fix − reproducescale ±2.44e3

Axial slice 20 of 124; the crosshair where FSL's result differs most from its initial alignment; the three slices through it are stored at full resolution. The interactive viewer needs JavaScript; the metrics below are complete without it.

Stored for this example: three full-resolution slices, one per plane, through display voxel (59, 64, 20): where the main comparison differs most, or the field's centre where nothing differs; every metric is computed over every voxel of the files.

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 its starting point
FSL flirt 6.0.7.23 − initial alignment (-init)referenceNot a comparison of ports: FSL’s registered image against the same input resampled at the call’s initial matrix (fMRIPrep’s mri_coreg, converted by lta_to_fsl), the change the boundary-based optimisation made.2.44e33,287,621 of 6,488,064 (50.7 %)40.30.9970.00308
The replica against the original
replica − FSL flirt 6.0.7.23Bit-identicalThe same values, every one, on this call; the matrices too, byte for byte as text.identical: 0 of 6,488,064 values differ
The replica against itself: thread counts, fix and reproduce
replica, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: FLIRT’s parallel steps (the boundary points, the blurs) give the same values at any thread count.identical: 0 of 6,488,064 values differ
replica, fix − reproduceBit-identicalNo correction of the register applies to this call: fix and reproduce give the same values.identical: 0 of 6,488,064 values differ
The clean room against the original and the replica
clean room − FSL flirt 6.0.7.23Within the original's own variabilityWritten without FSL’s source, the clean room is not bit-identical by design: its matrix is 4.87e-06 mm from FSL’s at the 95th percentile of the brain (at most 5.97e-06 mm), against FSL’s own 0.0907 mm when its input changes by noise of half an intensity unit (at most 0.113 mm); the images differ in 44.4% of their values, by at most 4.1e-06 of the range (1 − NCC 2.9e-13), the input interpolated at slightly other positions. The difference is within FSL’s own variability. record0.009162,879,043 of 6,488,064 (44.4 %)0.0003944.09e−62.94e−13
clean room − replicaWithin the original's own variabilityWritten without FSL’s source, the clean room is not bit-identical by design: its matrix is 4.87e-06 mm from the replica’s (FSL’s bytes) at the 95th percentile of the brain (at most 5.97e-06 mm), against FSL’s own 0.0907 mm when its input changes by noise of half an intensity unit (at most 0.113 mm); the images differ in 44.4% of their values, by at most 4.1e-06 of the range (1 − NCC 2.9e-13), the input interpolated at slightly other positions. The difference is within FSL’s own variability. record0.009162,879,043 of 6,488,064 (44.4 %)0.0003944.09e−62.94e−13
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 4 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel steps (the boundary points’ terms, the correlation ratio’s bins, the search’s grid points) are summed in a fixed order, so the values are the same at any thread count. recordidentical: 0 of 6,488,064 values differ
clean room, fix − reproduceDeliberate correctionGLIBC-07, a correction by default: the default computes sinf, cosf, atan2f, cos, exp and logf correctly rounded on every platform, where reproduce computes them as glibc 2.35 does, as FSL’s build (LARMORX’s glibc functions, since 2026-11-04); FLIRT names it nowhere: an iterative fit cannot tell whether its result moved with those last bits (as decided on 2026-11-04). Here the two modes’ matrices part by 0.00604 mm at the 95th percentile of the brain (at most 0.00745 mm), against FSL’s own 0.0907 mm on a perturbed input: the fit’s sensitivity to the C library’s last bits. record GLIBC-071.29e33,265,579 of 6,488,064 (50.3 %)0.8790.5751.47e−6

Δ is A − B, computed in float64 over all 6,488,064 voxels of the outputs, not only those shown. The records’ thresholds: every file FSL writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages (the replica); FSL’s exit status and messages, the numbers in them read as numbers, the matrix FSL’s text or within 0.1 mm of it at the 95th percentile of the reference’s brain (0.5 mm for registrations other than the boundary-based and 6-dof ones), and the image from its own matrix with a correlation of at least 0.999 with FSL’s inside the reference’s mask, bit-identical from the same matrix (the clean room, specs/flirt.md §13).

The matrices as the files hold them, both ports against the original: FSL flirt 6.0.7.23 (input millimetres to reference millimetres, FLIRT’s convention)

FSL flirt 6.0.7.23replica, reproducereplica, fixclean room, reproduceclean room, fixfMRIPrep's output (FSL in its image)the initial matrix (-init)
row 10.9998815716 0.002805483187 -0.01513461164 -5.2551043510.9998815716 0.002805483187 -0.01513461164 -5.2551043510.9998815716 0.002805483187 -0.01513461164 -5.2551043510.999881572 0.002805477616 -0.01513458952 -5.2551030770.9998816372 0.002871674131 -0.0151178772 -5.2597899130.9998815716 0.002805483187 -0.01513461164 -5.2551043510.99978888 0.00323859 -0.02029305 -5.28424311
row 20.0009408139998 0.9702730696 0.2420112957 -15.087933180.0009408139998 0.9702730696 0.2420112957 -15.087933180.0009408139998 0.9702730696 0.2420112957 -15.087933180.0009408141138 0.9702731353 0.242011312 -15.087940010.0008725537645 0.9702728647 0.242012653 -15.081048960.0009408139998 0.9702730696 0.2420112957 -15.087933180.00181520 0.96972632 0.24418817 -15.75937462
row 30.01536364661 -0.2419968245 0.9701554835 103.10732530.01536364661 -0.2419968245 0.9701554835 103.10732530.01536364661 -0.2419968245 0.9701554835 103.10732530.0153636238 -0.2419968242 0.9701554841 103.10732750.01536342958 -0.241997145 0.970155457 103.10737950.01536364661 -0.2419968245 0.9701554835 103.10732530.02046951 -0.24417339 0.96951562 102.36650848
row 40 0 0 10 0 0 10 0 0 10 0 0 10 0 0 10 0 0 10.00000000 0.00000000 0.00000000 1.00000000
the file, byte for bytethe referenceidenticalidenticaldiffersdiffersidentical–

Each matrix file as written (-omat), its numbers as text; fMRIPrep’s own is the node’s output in the run’s working directory (FSL in fMRIPrep 25.2.6’s image), the initial matrix the call’s -init (mri_coreg’s, converted by lta_to_fsl). identical: the same bytes as FSL 6.0.7.23’s file. The clean room’s matrices are its own result, written as FSL writes them.

The displacement between two matrices over the reference’s brain, both ports against the original: FSL flirt 6.0.7.23 (mm)

Matriceslargestmean95th percentileFSL's RMS deviation (80 mm)
replica − FSL flirt 6.0.7.230000
replica, fix − FSL flirt 6.0.7.230000
clean room − FSL flirt 6.0.7.235.97e−63.07e−64.87e−66.8e−6
clean room, fix − FSL flirt 6.0.7.230.007450.003260.006040.00822
FSL flirt 6.0.7.23 on a perturbed input − FSL flirt 6.0.7.23 (its own variability)0.1130.05620.09070.236
fMRIPrep’s output − FSL flirt 6.0.7.230000
FSL flirt 6.0.7.23 − the initial matrix (what the registration changed)1.060.7680.9791.3

For every one of the 1,229,002 voxels of the reference’s brain mask (its voxels above zero), the distance in millimetres between the two points of the input the matrices map onto it, computed in float64 from the matrices as the files hold them; the last column is FSL’s own measure (rmsdiff’s: the RMS deviation over a sphere of radius 80 mm about the origin).

Provenance of this example
  • Data: OpenNeuro ds000228, sub-pixar008, CC0-1.0
  • Inputs: -in (sha256 8f6990655e926cfd…); -ref (sha256 90ec7743ccdfb7f1…); -wmseg (sha256 d6b7d92587c21922…); -init (sha256 a12603b405470263…)
  • Runs: figures cache entry fsl-flirt/fmriprep-ds000228-sub-pixar008-noreconall, computed 2026-11-01T16:13:46+00:00 with larmorx 301fab7dfc56; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Inputs: -in (sha256 8f6990655e926cfd…); -ref (sha256 90ec7743ccdfb7f1…); -wmseg (sha256 d6b7d92587c21922…); -init (sha256 a12603b405470263…)
  • Runs: figures cache entry fsl-flirt/fmriprep-ds000228-sub-pixar008-noreconall-original, computed 2026-11-02T21:06:20+00:00; 3 of its 4 runs made again since, last on 2026-11-07T02:06:22+00:00, with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-flirt 2111.4; ~/fsl-6.0.7.23, never on the PATH, OpenBLAS on its Haswell kernels); “larmorx 0.0.1 (crate larmorx-fsl: the clean-room original, Apache-2.0, written from specs/flirt.md)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-flirt.json, exported by python -m larmorx_validation evidence fsl-flirt.

FLIRT’s other programs

FSL builds these programs with FLIRT, each with FSL’s command line in larmorx fsl <program> and a Python function. avscale decomposes a matrix into rotations, translations, scales and skews (and the half transforms). img2imgcoord, std2imgcoord and img2stdcoord map coordinates between images through a matrix. rmsdiff gives the RMS deviation between two matrices over a sphere of 80 mm, or the largest and RMS displacement over a mask. makerot writes a rotation about an axis through a centre. What they print is FSL’s text, character for character, from both ports, in the record’s programs/ cases below and in the probe calls behind them (avscale on 400 random matrices).

python
import numpy as np

m = np.loadtxt("bold2t1.mat")
d = lx.fsl.avscale(m, "boldref.nii.gz", all_params=True)       # avscale --allparams
xyz = lx.fsl.img2imgcoord([[10, 20, 30]], "boldref.nii.gz", "T1w.nii.gz", xfm=m)
lx.fsl.rmsdiff(m, np.eye(4), "boldref.nii.gz")                   # mm
r = lx.fsl.makerot(30, axis=(1, 2, 3), centre=(4, 5, 6))         # 30 degrees

The rest of the programs FSL builds with FLIRT came with the replica’s part 3 (History). midtrans finds the transform to the “middle” of several transforms of images to one template, a Procrustes average about the template’s centre. pointflirt finds the affine matrix that maps one set of points onto another in the least-squares sense. applyxfm4D resamples every volume of a series through one matrix or one per volume, in the series’ own data type: char, short, int, float or double, with double precision where FSL keeps it. FLIRT’s scripts: pairreg registers two brains refined with their skulls, extracttxt prints the lines around a match in a text file, and aff2rigid approximates a standard-space matrix by a rigid one that keeps the anterior commissure, the AC-PC line and the mid-sagittal plane.

Their matrices and images are FSL’s bit for bit in the replica’s record (midtrans’ and pointflirt’s singular-value decompositions follow FSL’s own LAPACK word for word). The clean room prints the same text and writes the same files: its fits’ matrices within 10⁻⁹ of FSL’s (a unit of the printed results’ last digit), applyxfm4D’s nearest-neighbour and trilinear images bit for bit, its sinc and spline images within their last bits, and pairreg’s matrix within the record’s displacement threshold.

python
mid = lx.fsl.midtrans(["A2temp.mat", "B2temp.mat", "C2temp.mat"])    # 4 x 4
aff = lx.fsl.pointflirt(in_points, ref_points)                         # N x 3 each, in mm
lx.fsl.applyxfm4d("bold.nii.gz", "boldref.nii.gz", "out.nii.gz", "mats_dir", interp="spline")
rigid = lx.fsl.aff2rigid("T1w2MNI.mat")

Random testing after the catch-up

The clean-room catch-up’s random-testing run (2026-11-05) drew the earlier runs’ cases again and further, on the clean room with the specification’s answers: 13,907 random calls of every program in 31 minutes, each compared as the record compares. 12,334 agreed within the thresholds; of the 1,140 differences of the run of 2026-11-02 drawn again, 696 now agree. Each of the 1,560 that differed is in one category below. Most are registrations whose path parts from FSL’s on random, often ill-posed pairs, and messages whose number follows that path. In 38 pairreg runs FSL corrupts its own heap (FSL-75).

Four classes were the clean room’s own: FSL’s Cost minima line on a one-point grid, -out’s negative zeros of a completed sform, applyxfm4D’s int32 padding and, found afterwards through BET’s random testing, the order of the search’s starting points of equal cost, which FSL sorts with std::sort (873 of the run’s path findings: 755 registrations, 101 pairreg runs and 17 message counts). Each was fixed with a permanent case of the record, and each of their findings agrees on replay with the final build. FSL’s sub-threshold costs on a test volume without intensity range are a question for the specification.

Random testing, the clean room against the original: FSL 6.0.7.23’s FLIRT programs (13,907 random calls, 2026-11-05)

Cases
Random cases, each compared with FSL 6.0.7.2313,907
within the record’s thresholds (§13), every exact check passed12,334
differing (each one in a category below)1,560
a register entry named in one mode only (a notice)13
flirt, applyxfm4D, pairreg8,134 / 804 / 315
the other programs (convert_xfm, avscale, midtrans, …)4,967
each cost (corratio, mutualinfo, normmi, normcorr, leastsq, woods, labeldiff, bbr)865 / 401 / 434 / 462 / 333 / 68 / 96 / 744
3, 6, 7, 9, 12 dof167 / 1,039 / 372 / 323 / 348
-applyxfm, -applyisoxfm1,738 / 727
-debug, -verbose268 / 547
distinct options and settings drawn135
length (minutes; 14 jobs at one thread each)31.2

A random-testing run of the clean room, 31 minutes long, after the specification’s answers of 2026-11-03 (seed 20261103: the cases of the random-testing runs of 2026-11-02 drawn again and further, FSL’s runs served by the cache of the original’s runs where they had run), each compared as the record compares (specs/flirt.md §13: exit status and messages FSL’s, files identical, matrices within a displacement over the reference’s brain, images from the clean room’s own matrix by their correlation). The build: larmorx 5d35983b24db; four classes were fixed after it, from its findings. The notices: FSL-09 6, FSL-111 10, FSL-187 1, FSL-79 1, FSL-80 4 cases. Every case in a memory box of 3 GiB (peaks below 0.35 GiB).

The clean room’s differences from the original in that run, by category

CategoryCasesVerdictReproducer
Fixed after the run (the clean-room catch-up, its step 10), with permanent cases: FLIRT’s search kept its starting points of equal cost in the order it found them, where FSL sorts the list by cost with libstdc++'s std::sort, which is not stable, before it enters S and P. On these random pairs many costs tie (an input or a pair that carries no information at the coarse scales: every cost 1), so another starting point came first and the registration ended elsewhere, or its messages followed another path. Each of these agrees on replay with the final build and differed with the build before the fix (catchup_replay.py; trial_catchup_replay_sorted.txt, trial_catchup_replay_before_sort.txt): 755 registrations, 101 pairreg runs and 17 message counts873fixed: agrees on replay with the final build20261103-00035-original
A registration ends elsewhere than FSL’s (its matrix beyond the record’s threshold, or only the image written from it) on the run’s random, often ill-posed pairs: small synthetic heads and blobs, odd orientations and voxel sizes, where FSL’s own variability is often 0. The costs follow FSL’s arithmetic within their last bits and the optimiser’s path then parts; left for FLIRT’s next clean-room part and the final validation phase338Open difference20261103-00088-original
Messages whose number follows the optimiser’s path, printed a different number of times: mcflirt-style COG warnings, FSL-90’s ‘Parameters do not form a valid axis’, ‘Erroneous dof’, Woods’ warning, -verbose’s lines237Open difference20261103-00045-original
pairreg where FSL itself corrupts its heap (Segmentation fault, free(): invalid size or pointer, double free: FSL-75, undefined behaviour never reproduced) and the clean room registers and finishes38Expected divergence20261103-00203-original
pairreg’s three registrations: their paths part as a registration’s do23Open difference20261103-00185-original
Fixed after the run, with a permanent case: -out’s header where a reference with a qform only and qfac −1 has its sform completed from the qform, whose zeros times −1 are negative zeros that FSL writes (the clean room wrote zeros); each agrees on replay with the final build23fixed: agrees on replay with the final build20261103-00570-original
Fixed after the run, with a permanent case: a verbose search on a one-point grid (-nosearch) finds no local minimum, so FSL lists the overall minimum only on its Cost minima line, without a COST line; each agrees on replay with the final build14fixed: agrees on replay with the final build20261103-00413-original
A test volume with no intensity range at a scale (Testvol intensity clamped between 0 and 0): FSL prints ‘WARNING: Found 0 or less sub-threshold costs’ and makes every grid point a minimum, where the clean room evaluates the fine grid’s costs (corratio’s cost of no variance, 1): a question for the specification (specs/flirt.md §8.1 steps 6 and 7)9Open difference20261103-01093-original
-applyxfm and -applyisoxfm images from the same matrix: the windowed sinc’s and the nearest neighbour’s last bits on small images3Open difference20261103-01750-original
aff2rigid’s corrected matrix (FSL-81) against the replica’s: not yet looked at1Open difference20261103-00473-original
Fixed after the run, with a permanent case: applyxfm4D of an int32 image of 2147483647, whose padding FSL writes exactly in the type (and its interpolated voxels through single precision: FSL-09); it agrees on replay with the final build1fixed: agrees on replay with the final build20261103-03732-original

Every differing case of the run in one category by its program, its failed checks and their details (validation/investigations/cleanroom-flirt/trial_catchup_categories.py, its output validation/investigations/cleanroom-flirt/trial_catchup_triage.json), each category’s cause established by replaying findings (catchup_replay.py, with the final build and with the build before the search’s sort) and by probes (catchup_finding.py, catchup8_int_nn.py; docs/findings/fsl-flirt-clean-room.md); the triage’s entries: validation/difftest/triage.toml. The reproducer is a finding of the run (difftest --replay).

Every case

Each case of the two records, the replica’s and the clean room’s, both against FSL 6.0.7.23 (a column each), with what it tests, its command line, each port’s verdict and the reason. Search, filter by category or verdict (or select a segment of a chart above), sort by a column; #case=<name> links to one case.

349 cases

CaseCategoryWhat it testsreplicaclean roomChecks
fmriprep/ds000228-sub-pixar008-noreconall/bold_task_pixar_wf worked examplefMRIPrep's callsfMRIPrep's BBR call (ds000228-sub-pixar008)
command lineflirt -in sub-pixar008_task-pixar_desc-coreg_boldref.nii.gz -ref sub-pixar008_desc-preproc_T1w_masked.nii.gz -out sub-pixar008_task-pixar_desc-coreg_boldref_flirt.nii.gz -omat sub-pixar008_task-pixar_desc-coreg_boldref_flirt.mat -basescale 1 -searchcost bbr -dof 6 -init out.mat -schedule bbr.sch -wmseg sub-pixar008_dseg_mask.nii.gz
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages; fMRIPrep’s own matrix (its image’s FSL) the same as text

Within the original's own variability

stdout: the same lines (numbers aside): numbers differ; sub-pixar008_task-pixar_desc-coreg_boldref_flirt.nii.gz: correlation with FSL’s image inside the reference’s mask: 1.0; the matrix 0 mm from FSL’s (the record’s precision); within FSL’s own variability on this call, measured for the worked example: 0.0907 mm at the 95th percentile (its input perturbed by noise of half an intensity unit). In fix, its default, 3,265,591 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named, as decided…

7/7 · 8/8
fmriprep/ds003592-sub-280-noreconall/bold_ses_1_task_rest_echo_1_wf worked examplefMRIPrep's callsfMRIPrep's BBR call (ds003592-sub-280)
command lineflirt -in sub-280_ses-1_task-rest_desc-coreg_boldref.nii.gz -ref sub-280_ses-1_desc-preproc_T1w_masked.nii.gz -out sub-280_ses-1_task-rest_desc-coreg_boldref_flirt.nii.gz -omat sub-280_ses-1_task-rest_desc-coreg_boldref_flirt.mat -basescale 1 -searchcost bbr -dof 6 -init out.mat -schedule bbr.sch -wmseg sub-280_ses-1_dseg_mask.nii.gz
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages; fMRIPrep’s own matrix (its image’s FSL) the same as text

Within the original's own variability

stdout: the same lines (numbers aside): numbers differ; sub-280_ses-1_task-rest_desc-coreg_boldref_flirt.mat: 95th-percentile displacement over the reference’s brain: 0.0002; sub-280_ses-1_task-rest_desc-coreg_boldref_flirt.nii.gz: correlation with FSL’s image inside the reference’s mask: 1.0; within FSL’s own variability on this call, measured for the worked example: 0.00314 mm at the 95th percentile (its input perturbed by noise of half an intensity unit). In fix, its default, 5,140,511 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce compu…

7/7 · 8/8
fmriprep/ds005040-sub-002-noreconall/bold_task_clip_run_1_wf worked examplefMRIPrep's callsfMRIPrep's BBR call (ds005040-sub-002)
command lineflirt -in sub-002_task-clip_run-1_desc-coreg_boldref.nii.gz -ref sub-002_desc-preproc_T1w_masked.nii.gz -out sub-002_task-clip_run-1_desc-coreg_boldref_flirt.nii.gz -omat sub-002_task-clip_run-1_desc-coreg_boldref_flirt.mat -basescale 1 -searchcost bbr -dof 6 -init out.mat -schedule bbr.sch -wmseg sub-002_dseg_mask.nii.gz
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages; fMRIPrep’s own matrix (its image’s FSL) the same as text

Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ; sub-002_task-clip_run-1_desc-coreg_boldref_flirt.mat: 95th-percentile displacement over the reference’s brain: 0.0113; sub-002_task-clip_run-1_desc-coreg_boldref_flirt.nii.gz: correlation with FSL’s image inside the reference’s mask: 0.999978. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability on this call 0.00855 mm. In fix, its default, 6,689,565 values change, from sinf, cosf, atan2f, cos, exp and logf computed …

7/7 · 8/8
real/anglerep-quaternionreal registrations-anglerep quaternion: the rotations as a quaternion's last three components (12 dof)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -anglerep quaternion -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/anglerep-quaternion-dof6real registrations-anglerep quaternion at 6 dof, -out, verbose (the parameters printed)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -anglerep quaternion -dof 6 -out out -omat out.mat -v
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/default worked examplereal registrationsthe default (12 dof, corratio) with -out and -omat
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -out out -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 866,828 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/default-fullreal registrationsthe full-size T1w as reference
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1full.nii.gz: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/dof6real registrations-dof 6
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/dof7real registrations-dof 7
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 7 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/dof9real registrations-dof 9
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 9 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/cost-mutualinforeal registrations-cost mutualinfo (and -searchcost corratio)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost mutualinfo -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/cost-normmireal registrations-cost normmi (and -searchcost corratio)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost normmi -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/cost-normcorrreal registrations-cost normcorr (and -searchcost corratio)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost normcorr -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/cost-leastsqreal registrations-cost leastsq (and -searchcost corratio)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost leastsq -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 11 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/cost-woodsreal registrations-cost woods (and -searchcost corratio)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost woods -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/cost-labeldiffreal registrations-cost labeldiff (and -searchcost corratio)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost labeldiff -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Within the original's own variability

The matrix beyond the record’s fixed threshold (0.5 mm) but within FSL’s own variability, which the record measured on the same call with its input perturbed by noise of half an intensity unit (a poorly conditioned registration): out.mat: 95th-percentile displacement over the reference’s brain: 1.2259. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/searchcost-normmireal registrations-searchcost normmi
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -searchcost normmi -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/nosearchreal registrations-nosearch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -nosearch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/searchrreal registrations-searchrx/ry/rz with narrow ranges
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -searchrx -20 30 -searchry 15 -15 -searchrz -5 5 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.1349. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/coarse-finereal registrations-coarsesearch 30 -finesearch 10
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -coarsesearch 30 -finesearch 10 -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/initreal registrations-init (fMRIPrep's initial matrix)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/usesqformreal registrations-usesqform
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -usesqform -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/bins64real registrations-bins 64
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -bins 64 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/noresamplereal registrations-noresample
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -noresample -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/noclampreal registrations-noclamp
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -noclamp -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/minsamplingreal registrations-minsampling 4
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -minsampling 4 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/forcescalingreal registrations-forcescaling
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -forcescaling -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
real/weightsreal registrations-refweight and -inweight
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -inweight '<inweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/weights-normcorrreal registrationsweights with -cost normcorr
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost normcorr -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/weights-leastsqreal registrationsweights with -cost leastsq
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost leastsq -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/weights-normmireal registrationsweights with -cost normmi
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost normmi -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/weights-mutualinforeal registrationsweights with -cost mutualinfo
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost mutualinfo -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/weights-woodsreal registrationsweights with -cost woods
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost woods -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

stderr: the same lines (numbers aside): no (threshold the same text, numbers read as numbers)

6/6 · 4/5
real/weights-empty-refweightreal registrationsan empty -refweight (every voxel 0): refused by default (FSL-91)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<zero:t1.nii:zeroweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
real/weights-empty-inweightreal registrationsan empty -inweight (every voxel 0): refused by default (FSL-91)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -inweight '<zero:epi.nii:zeroinweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
real/applyxfm-nan-borderreal registrations-applyxfm of a BOLD reference with NaN voxels in its border and an infinite one: the padding's background value (FSL-22: NaN sorted last by default)
command lineflirt -in '<nanedge:epi.nii:epi_nanedge.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, the clean room corrects FSL-22 and names it: 1,149,794 values change.

6/6
real/labeldiff-dsegreal registrations-cost labeldiff on a label image
command lineflirt -in '<dseg_shift.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<dseg.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost labeldiff -searchcost labeldiff -dof 6 -interp nearestneighbour -out out -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.nii.gz: correlation with FSL’s image inside the reference’s mask: 1.0. In fix, its default, 4 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 7/7
real/known-motionreal registrationsa known rigid motion of the T1w (-dof 6)
command lineflirt -in '<t1_moved.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 10 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/interp-nearestneighbourreal registrations-interp nearestneighbour for -out
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -interp nearestneighbour -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/interp-sincreal registrations-interp sinc for -out
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -interp sinc -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 32,052 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/interp-splinereal registrations-interp spline for -out
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -interp spline -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/interp-trilinearreal registrations-interp trilinear for -out
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -interp trilinear -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/sinc-windowreal registrations-sincwidth 5 -sincwindow blackman
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -interp sinc -sincwidth 5 -sincwindow blackman -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 21,767 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/sinc-rectreal registrations-sincwindow rectangular
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -interp sinc -sincwindow rectangular -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 47,524 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/datatype-charreal registrations-datatype char
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -datatype char -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 756,121 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/datatype-shortreal registrations-datatype short
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -datatype short -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 560,026 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/datatype-intreal registrations-datatype int
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -datatype int -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 416,242 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/datatype-floatreal registrations-datatype float
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -datatype float -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/datatype-doublereal registrations-datatype double
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -datatype double -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/int16-inputreal registrationsan int16 input (its type kept)
command lineflirt -in '<epi_i16.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 416,491 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/paddingsizereal registrations-paddingsize 2
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -paddingsize 2 -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/setbackgroundreal registrations-setbackground 100
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -setbackground 100 -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/noresampblurreal registrations-noresampblur
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -noresampblur -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/applyisoxfmreal registrations-applyisoxfm 3
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyisoxfm 3 -out out -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/applyxfm-4d worked examplereal registrations-applyxfm on four volumes of the raw series, sinc
command lineflirt -in '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -interp sinc -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 128,674 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/applyisoxfm-4d-blurreal registrations-applyisoxfm 6 on four volumes of the raw series (blurred first)
command lineflirt -in '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyisoxfm 6 -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/applyisoxfm-4d-blur-setbackgroundreal registrations-applyisoxfm 6 with -setbackground (the blur's outside a constant)
command lineflirt -in '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyisoxfm 6 -setbackground 40 -interp spline -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/applyxfm-usesqformreal registrations-applyxfm -usesqform
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -usesqform -applyxfm -out out -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
real/out-onlyreal registrationsa registration with -out and no -omat (the matrix printed)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.nii.gz: correlation with FSL’s image inside the reference’s mask: 1.0. In fix, its default, 839,947 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
real/basescale-smallreal registrations0.6 mm voxels (the base scale from the voxel size)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1_small.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/basescale-small-initreal registrations0.6 mm voxels with -init (the initial matrix in the base scale's millimetres)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1_small.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -nosearch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/basescale-bigreal registrations14 mm voxels (the base scale from the voxel size)
command lineflirt -in '<epi_big.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
real/basescale-forcedreal registrations-basescale 2
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -basescale 2 -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
bbr/defaultboundary-based registrationBBR (FSL's bbr.sch) with -cost bbr
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0693; out.nii.gz: correlation with FSL’s image inside the reference’s mask: 0.999654; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 841,077 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not name…

6/6 · 7/7
bbr/fmriprep-formboundary-based registrationfMRIPrep's form: -searchcost bbr -basescale 1
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -searchcost bbr -basescale 1
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0693; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
bbr/type-global_absboundary-based registration-bbrtype global_abs
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -bbrtype global_abs
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0693; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
bbr/type-local_absboundary-based registration-bbrtype local_abs
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -bbrtype local_abs
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0268; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
bbr/debugboundary-based registration-debug with -cost bbr (BBR's debug volume `debug`, part 3)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -debug
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: debug.nii.gz: correlation with FSL’s image inside the reference’s mask: 0.999853; out.mat: 95th-percentile displacement over the reference’s brain: 0.0693; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 22,922 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not nam…

6/6 · 13/13
bbr/slopeboundary-based registration-bbrslope -0.25
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -bbrslope -0.25
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0063; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
bbr/coordsboundary-based registration-wmcoords and -wmnorms
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmcoords '<wmcoords.txt: derived from ds005040'"'"'s fMRIPrep run>' -wmnorms '<wmnorms.txt: derived from ds005040'"'"'s fMRIPrep run>' -cost bbr -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.004; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
bbr/fieldmapboundary-based registration-fieldmap -pedir 2 -echospacing (the BBR resampling for -out)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -fieldmap '<fmap.nii: derived from ds005040'"'"'s fMRIPrep run>' -pedir 2 -echospacing 0.0007 -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0797; out.nii.gz: correlation with FSL’s image inside the reference’s mask: 0.999949; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 985,605 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not name…

6/6 · 7/7
bbr/fieldmap-maskboundary-based registration-fieldmapmask -pedir -2
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -fieldmap '<fmap.nii: derived from ds005040'"'"'s fMRIPrep run>' -fieldmapmask '<fmapmask.nii: derived from ds005040'"'"'s fMRIPrep run>' -pedir -2
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0518; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
bbr/weightsboundary-based registrationBBR with -refweight
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>'
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.mat: 95th-percentile displacement over the reference’s brain: 0.0045; stdout: the same lines (numbers aside): numbers differ. The clean room’s registration: the cost’s last bits and the optimiser’s path (docs/findings/fsl-flirt-clean-room.md); FSL’s own variability is not measured for this case. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
bbr/bad-typeboundary-based registration-bbrtype unknown (FSL's exception)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/bbr.sch -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -dof 6 -omat out.mat -cost bbr -bbrtype nope
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/simple3D.schschedulesFSL's simple3D.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/simple3D.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
schedule/xyztrans.schschedulesFSL's xyztrans.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/xyztrans.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/ytransonly.schschedulesFSL's ytransonly.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/ytransonly.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/ztransonly.schschedulesFSL's ztransonly.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/ztransonly.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/sch3Dtrans_3dofschedulesFSL's sch3Dtrans_3dof
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/sch3Dtrans_3dof -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/measurecost1.schschedulesFSL's measurecost1.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/measurecost1.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/pairreg1.schschedulesFSL's pairreg1.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/pairreg1.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
schedule/pairreg2.schschedulesFSL's pairreg2.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/pairreg2.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
schedule/pairreg3.schschedulesFSL's pairreg3.sch
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -schedule $FSLDIR/etc/flirtsch/pairreg3.sch -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
schedule/sch2D_3dofschedulesFSL's sch2D_3dof on one slice
command lineflirt -in '<slice_in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<slice_ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -2D -schedule $FSLDIR/etc/flirtsch/sch2D_3dof -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/sch2D_4dofschedulesFSL's sch2D_4dof on one slice
command lineflirt -in '<slice_in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<slice_ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -2D -schedule $FSLDIR/etc/flirtsch/sch2D_4dof -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
schedule/sch2D_5dofschedulesFSL's sch2D_5dof on one slice
command lineflirt -in '<slice_in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<slice_ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -2D -schedule $FSLDIR/etc/flirtsch/sch2D_5dof -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/sch2D_6dofschedulesFSL's sch2D_6dof on one slice
command lineflirt -in '<slice_in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<slice_ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -2D -schedule $FSLDIR/etc/flirtsch/sch2D_6dof -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
real/2dreal registrations-2D's default schedule on one slice, with -out
command lineflirt -in '<slice_in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<slice_ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -2D -out out -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
verbose/vverbose levels-v
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -v -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
verbose/2verbose levels-verbose 2
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 2 -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
verbose/3verbose levels-verbose 3
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 3 -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
verbose/5verbose levels-verbose 5
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 5 -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ. Printed numbers differ in their last digits. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
verbose/printinitverbose levels-printinit
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -printinit -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
verbose/debugsaveverbose levels-debugsave -verbose 4 (the search's volumes saved)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -debugsave -verbose 4 -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
verbose/debugverbose levels-debug: the pyramid's volumes and weights saved (no weights: the one-voxel weight)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -debug -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 11/11
verbose/debug-weightsverbose levels-debug with -refweight and -inweight (the weights' levels saved)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -refweight '<refweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -inweight '<inweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -debug -dof 6 -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 15/15
schedule/every-commandschedulesa schedule using every command and option
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule all.sch -omat out.mat (with all.sch written first)
Rejected by both

flirt exits 255; the same from the replica. Wrong number of args to SETOPTION

Rejected by both

flirt exits 255; the same from the clean room. Wrong number of args to SETOPTION

5/5 · 4/4
schedule/args-copyschedulesa schedule: copy U
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Rejected by both

flirt exits 255; the same from the replica. Wrong number of args to COPY

Rejected by both

flirt exits 255; the same from the clean room. Wrong number of args to COPY

5/5 · 4/4
schedule/args-optimiseschedulesa schedule: optimise 6 U
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Rejected by both

flirt exits 255; the same from the replica. Wrong number of args to OPTIMISE

Rejected by both

flirt exits 255; the same from the clean room. Wrong number of args to OPTIMISE

5/5 · 4/4
schedule/bad-ifschedulesa schedule: if 1 =< 2
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Rejected by both

flirt exits 255; the same from the replica. Cannot recognise operator =< in IF statement

Rejected by both

flirt exits 255; the same from the clean room. Cannot recognise operator =< in IF statement

5/5 · 4/4
schedule/exitschedulesa schedule: setrow U 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/no-matrixschedulesa schedule: clear U
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/bad-costfunctionschedulesa schedule: setoption costfunction nope
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Rejected by both

flirt exits 255; the same from the replica. Unrecognised cost function: nope

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised cost function: nope

5/5 · 4/4
schedule/bad-matnameschedulesa schedule: clear Uq
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Rejected by both

flirt exits -11; the same from the replica. Can only accept UA to UZ, not Uq

Rejected by both

flirt exits -11; the same from the clean room. Can only accept UA to UZ, not Uq

5/5 · 4/4
schedule/alignpaxesschedulesalignpaxes on real images
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/alignpaxes-constantschedulesalignpaxes on constant images (zero moments, total 0)
command lineflirt -in '<const:12x10x8:2x2x2:c.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<const:12x10x8:2x2x2:c.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/alignpaxes-nanschedulesalignpaxes on an input with a NaN voxel (eig_sym fails: the exception)
command lineflirt -in '<nan:epi.nii:epi_nan.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/alignpaxes-2dschedulesalignpaxes in 2D mode (one slice: a zero third moment)
command lineflirt -in '<slice_in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<slice_ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -2D -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/alignpaxes-argsschedulesalignpaxes without its matrix
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Rejected by both

flirt exits 255; the same from the replica. Wrong number of args to ALIGNPAXES

Rejected by both

flirt exits 255; the same from the clean room. Wrong number of args to ALIGNPAXES

5/5 · 4/4
schedule/paramsubset-generalschedulesa parameter subset whose column has two ones in different rows of another (rz; tx+ty)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/paramsubset-weightedschedulesa parameter subset with weights other than 0 and 1 sharing rows (the SVD's QR path)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
schedule/paramsubset-singularschedulesa parameter subset reaching a singular matrix in a cost evaluation (the exception caught by FLIRT's main: FSL-68)
command lineflirt -in '<const:1x1x1:1x1x1:one.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<const:1x1x1:1x1x1:one.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-68 and names it

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/no-argsedge cases and errorsno arguments (the usage, exit 1)
command lineflirt
Rejected by both

flirt exits 1; the same from the replica.

Rejected by both

flirt exits 1; the same from the clean room.

5/5 · 4/4
edge/helpedge cases and errors-help
command lineflirt -help
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/versionedge cases and errors-version
command lineflirt -version
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/no-inedge cases and errorsno -in
command lineflirt -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>'
Rejected by both

flirt exits 2; the same from the replica. ERROR:: Input volume filename not found

Rejected by both

flirt exits 2; the same from the clean room. ERROR:: Input volume filename not found

5/5 · 4/4
edge/no-refedge cases and errorsno -ref
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>'
Rejected by both

flirt exits 2; the same from the replica. ERROR:: Reference volume filename not found

Rejected by both

flirt exits 2; the same from the clean room. ERROR:: Reference volume filename not found

5/5 · 4/4
edge/not-optionedge cases and errorsa word that is not an option
command lineflirt epi.nii
Rejected by both

flirt exits 255; the same from the replica. Unrecognised option e

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised option e

5/5 · 4/4
edge/unknown-1edge cases and errorsan unknown option with one word after it
command lineflirt -frob x
Rejected by both

flirt exits 255; the same from the replica. Lacking argument to option -frob

Rejected by both

flirt exits 255; the same from the clean room. Lacking argument to option -frob

5/5 · 4/4
edge/unknown-2edge cases and errorsan unknown option with two words after it
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -frob x y
Rejected by both

flirt exits 255; the same from the replica. Unrecognised option -frob

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised option -frob

5/5 · 4/4
edge/lackingedge cases and errorsan option lacking its argument
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref
Rejected by both

flirt exits 255; the same from the replica. Lacking argument to option -ref

Rejected by both

flirt exits 255; the same from the clean room. Lacking argument to option -ref

5/5 · 4/4
edge/lacking-2edge cases and errorsa two-argument option lacking one
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -searchrx 10
Rejected by both

flirt exits 255; the same from the replica. Lacking argument to option -searchrx

Rejected by both

flirt exits 255; the same from the clean room. Lacking argument to option -searchrx

5/5 · 4/4
edge/bad-costedge cases and errors-cost nope
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost nope
Rejected by both

flirt exits 255; the same from the replica. Unrecognised cost function type: nope

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised cost function type: nope

5/5 · 4/4
edge/bad-searchcostedge cases and errors-searchcost nope
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -searchcost nope
Rejected by both

flirt exits 255; the same from the replica. Unrecognised cost function type: nope

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised cost function type: nope

5/5 · 4/4
edge/bad-interpedge cases and errors-interp cubic
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -interp cubic
Rejected by both

flirt exits 255; the same from the replica. Unrecognised interpolation method: cubic

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised interpolation method: cubic

5/5 · 4/4
edge/bad-datatypeedge cases and errors-datatype long
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -datatype long
Rejected by both

flirt exits 255; the same from the replica. Unrecognised data type: long

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised data type: long

5/5 · 4/4
edge/bad-sincwindowedge cases and errors-sincwindow gauss
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -sincwindow gauss
Rejected by both

flirt exits 255; the same from the replica. Unrecognised sinc window: gauss

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised sinc window: gauss

5/5 · 4/4
edge/bad-anglerepedge cases and errors-anglerep rodrigues
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -anglerep rodrigues
Rejected by both

flirt exits 255; the same from the replica. Unrecognised angle representation: rodrigues

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised angle representation: rodrigues

5/5 · 4/4
edge/bad-pediredge cases and errors-pedir 0
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -pedir 0
Rejected by both

flirt exits 255; the same from the replica. Unrecognised argument to pedir (0) - it should be a number between -3 and 3 (not 0)

Rejected by both

flirt exits 255; the same from the clean room. Unrecognised argument to pedir (0) - it should be a number between -3 and 3 (not 0)

5/5 · 4/4
edge/missing-inedge cases and errorsan input that does not exist
command lineflirt -in nothere.nii.gz -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat o.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/missing-refedge cases and errorsa reference that does not exist
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref nothere -omat o.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/missing-initedge cases and errorsan -init file that does not exist
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init none.mat -omat o.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/missing-scheduleedge cases and errorsa -schedule file that does not exist
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule none.sch -omat o.mat
Rejected by both

flirt exits 255; the same from the replica. Could not open filenone.sch

Rejected by both

flirt exits 255; the same from the clean room. Could not open filenone.sch

5/5 · 4/4
edge/bad-initedge cases and errorsan -init file with too few numbers
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init bad.mat -omat o.mat (with bad.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/coords-mismatchedge cases and errors-wmcoords and -wmnorms of different lengths
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -wmcoords '<wmcoords.txt: derived from ds005040'"'"'s fMRIPrep run>' -wmnorms n.txt -cost bbr (with n.txt written first)
Rejected by both

flirt exits 1; the same from the replica. Coordinate and Normal matrices wmcoords.txt and n.txt are either zero size or different sizes

Rejected by both

flirt exits 1; the same from the clean room. Coordinate and Normal matrices wmcoords.txt and n.txt are either zero size or different sizes

5/5 · 4/4
edge/outtype-unsetedge cases and errorsFSLOUTPUTTYPE unset
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -out out
Rejected by both

flirt exits 1; the same from the replica. ERROR:: Environment variable FSLOUTPUTTYPE is not set! Please make sure that the appropriate configuration file is sourced by your shell (e.g. by putting it in .profile). e.g. bash or sh users add the

Rejected by both

flirt exits 1; the same from the clean room. ERROR:: Environment variable FSLOUTPUTTYPE is not set! Please make sure that the appropriate configuration file is sourced by your shell (e.g. by putting it in .profile). e.g. bash or sh users add the

5/5 · 4/4
edge/outtype-NIFTIedge cases and errorsFSLOUTPUTTYPE=NIFTI
command lineFSLOUTPUTTYPE=NIFTI flirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
edge/outtype-NIFTI_PAIRedge cases and errorsFSLOUTPUTTYPE=NIFTI_PAIR
command lineFSLOUTPUTTYPE=NIFTI_PAIR flirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
edge/outtype-NIFTI2_GZedge cases and errorsFSLOUTPUTTYPE=NIFTI2_GZ
command lineFSLOUTPUTTYPE=NIFTI2_GZ flirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -out out
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
edge/outtype-ANALYZEedge cases and errorsFSLOUTPUTTYPE=ANALYZE
command lineFSLOUTPUTTYPE=ANALYZE flirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -out out
Rejected by both

flirt exits 1; the same from the replica. ERROR:: Unrecognised value (ANALYZE) of environment variable FSLOUTPUTTYPE Legal values are: NIFTI, NIFTI_PAIR, NIFTI2, NIFTI2_PAIR, NIFTI_GZ, NIFTI_PAIR_GZ, NIFTI2_GZ, NIFTI2_PAIR_GZ, NIFTI_ZST, NIFT

Rejected by both

flirt exits 1; the same from the clean room. ERROR:: Unrecognised value (ANALYZE) of environment variable FSLOUTPUTTYPE Legal values are: NIFTI, NIFTI_PAIR, NIFTI2, NIFTI2_PAIR, NIFTI_GZ, NIFTI_PAIR_GZ, NIFTI2_GZ, NIFTI2_PAIR_GZ, NIFTI_ZST, NIFT

5/5 · 4/4
convert_xfm/no-argsconvert_xfmno arguments
command lineconvert_xfm
Rejected by both

convert_xfm exits 1; the same from the replica.

Rejected by both

convert_xfm exits 1; the same from the clean room.

5/5 · 4/4
convert_xfm/helpconvert_xfm-help
command lineconvert_xfm -help
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
convert_xfm/inverseconvert_xfm-omat -inverse
command lineconvert_xfm -omat inv.mat -inverse '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
convert_xfm/concatconvert_xfm-omat -concat
command lineconvert_xfm -omat cat.mat -concat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
convert_xfm/fixscaleskewconvert_xfm-fixscaleskew
command lineconvert_xfm -omat f.mat -fixscaleskew s.mat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' (with s.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
convert_xfm/verboseconvert_xfm-v with -concat and -inverse (the matrices printed)
command lineconvert_xfm -v -concat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -inverse '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
convert_xfm/verbose-1convert_xfm-verbose 1 (the result printed)
command lineconvert_xfm -verbose 1 '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
convert_xfm/copyconvert_xfm-omat only (the matrix written again)
command lineconvert_xfm -omat c.mat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
convert_xfm/missingconvert_xfman input matrix that does not exist
command lineconvert_xfm -omat o.mat none.mat
Rejected by both

convert_xfm exits 254; the same from the replica. Could not open matrix file none.mat Cannot read input-matrix

Rejected by both

convert_xfm exits 254; the same from the clean room. Could not open matrix file none.mat Cannot read input-matrix

5/5 · 4/4
convert_xfm/shortconvert_xfma 3-row input matrix
command lineconvert_xfm -omat o.mat s.mat (with s.mat written first)
Rejected by both

convert_xfm exits 254; the same from the replica. Cannot read input-matrix

Rejected by both

convert_xfm exits 254; the same from the clean room. Cannot read input-matrix

5/5 · 4/4
convert_xfm/no-inputconvert_xfmno input matrix
command lineconvert_xfm -omat o.mat
Rejected by both

convert_xfm exits 2; the same from the replica. Input matrix filename not found

Rejected by both

convert_xfm exits 2; the same from the clean room. Input matrix filename not found

5/5 · 4/4
convert_xfm/two-inputsconvert_xfmtwo input matrices
command lineconvert_xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' x.mat
Rejected by both

convert_xfm exits 2; the same from the replica. Unknown option x.mat

Rejected by both

convert_xfm exits 2; the same from the clean room. Unknown option x.mat

5/5 · 4/4
convert_xfm/unknownconvert_xfman unknown option
command lineconvert_xfm -frob x '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Rejected by both

convert_xfm exits 255; the same from the replica. Unrecognised option -frob

Rejected by both

convert_xfm exits 255; the same from the clean room. Unrecognised option -frob

5/5 · 4/4
convert_xfm/lackingconvert_xfm-omat lacking its argument
command lineconvert_xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -omat
Rejected by both

convert_xfm exits 2; the same from the replica. Lacking argument to option -omat

Rejected by both

convert_xfm exits 2; the same from the clean room. Lacking argument to option -omat

5/5 · 4/4
convert_xfm/singularconvert_xfm-inverse of a singular matrix
command lineconvert_xfm -inverse z.mat (with z.mat written first)
Rejected by both

convert_xfm exits -6; the same from the replica. terminate called after throwing an instance of ‘std::runtime_error’ what(): inv(): matrix is singular

Rejected by both

convert_xfm exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::runtime_error’ what(): inv(): matrix is singular

5/5 · 4/4
edge/bbrstep-0edge cases and errorssetoption bbrstep 0 with BBR: FLIRT's loop over the boundary points never ends (FSL-71)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost bbr -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first)
Expected divergence

FLIRT never finishes (FSL-71: undefined behaviour); the replica refuses in fix and in reproduce

Expected divergence

FSL never finishes (FSL-71: undefined behaviour); the clean room refuses in fix and in reproduce

3/3 · 2/2
edge/read-not-numbersedge cases and errorsa schedule's read of a file holding a word that is not a number: FLIRT reads for ever, its memory growing (FSL-72)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -schedule s.sch -omat out.mat (with s.sch written first) (with bad.txt written first)
Expected divergence

FLIRT never finishes (FSL-72: undefined behaviour); the replica refuses in fix and in reproduce

Expected divergence

FSL never finishes (FSL-72: undefined behaviour); the clean room refuses in fix and in reproduce

3/3 · 2/2
edge/background-overflowedge cases and errorsa 2 x 1 x 1 input and an 8 x 1 x 1 reference: FLIRT's pyramid computes the background of a 3-voxel-wide level, writing past its buffer (FSL-75; a 2-voxel input since FSL-92 refuses one voxel by default)
command lineflirt -in '<const:2x1x1:2.5x2.5x2.5:in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<const:8x1x1:1x1.5x0.5:ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Expected divergence

FSL crashes (FSL-75: undefined behaviour); the replica finishes in fix and in reproduce and names FSL-75

Expected divergence

FSL crashes (FSL-75: undefined behaviour); the clean room finishes naming it

3/3 · 2/2
edge/background-emptyedge cases and errors-applyxfm of a 1-voxel input: its background value reads an empty buffer (FSL-75)
command lineflirt -in '<const:1x1x1:2x2x2:in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<const:6x6x6:2x2x2:ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -applyxfm -init id.mat -out out (with id.mat written first)
Expected divergence

FSL crashes (FSL-75: undefined behaviour); the replica refuses in fix and in reproduce

Expected divergence

FSL crashes (FSL-75: undefined behaviour); the clean room refuses in fix and in reproduce

3/3 · 2/2
edge/one-voxel-inputedge cases and errorsa 1 x 1 x 1 input registered to the T1w: one value, FLIRT's matrix an artefact of its search; refused by default (FSL-92)
command lineflirt -in '<const:1x1x1:2x2x2:one.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
edge/zero-voxel-sizeedge cases and errorsa header voxel size of 0 along z: FLIRT's base scale 0, every voxel size infinite, it never finishes; counted as 1 by default, refused with reproduce (FSL-93)
command lineflirt -in '<zerovox:epi.nii:epi_zerovox.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat
Expected divergence

FLIRT never finishes (FSL-93: undefined behaviour); the replica corrects it by default and refuses with reproduce

Expected divergence

FSL never finishes (FSL-93: undefined behaviour); the clean room corrects it by default and refuses with reproduce

3/3 · 2/2
edge/basescale-0edge cases and errors-basescale 0: every voxel size infinite, FLIRT never finishes (FSL-93)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -basescale 0 -omat out.mat
Expected divergence

FLIRT never finishes (FSL-93: undefined behaviour); the replica refuses in fix and in reproduce

Expected divergence

FSL never finishes (FSL-93: undefined behaviour); the clean room refuses in fix and in reproduce

3/3 · 2/2
convert_xfm/missing-concatconvert_xfm-concat with a file that does not exist
command lineconvert_xfm -concat none.mat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Rejected by both

convert_xfm exits 253; the same from the replica. Could not open matrix file none.mat Cannot read concat-matrix

Rejected by both

convert_xfm exits 253; the same from the clean room. Could not open matrix file none.mat Cannot read concat-matrix

5/5 · 4/4
programs/applyxfm4D-usageFLIRT's other programsapplyxfm4D without arguments (its usage, exit 255)
command lineapplyxfm4D
Rejected by both

applyxfm4D exits 255; the same from the replica. Usage: ~/fsl-6.0.7.23/bin/applyxfm4D <transformation matrix file/dir> --interp, -interp <nearestneighbour (or nn), trilinear, spline, sinc (defau

Rejected by both

applyxfm4D exits 255; the same from the clean room. Usage: ~/fsl-6.0.7.23/bin/applyxfm4D <transformation matrix file/dir> --interp, -interp <nearestneighbour (or nn), trilinear, spline, sinc (defau

5/5 · 4/4
programs/applyxfm4D-defaultFLIRT's other programsa 4D int16 series, a matrix per volume (MAT_0000n), sinc
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz mats (with mats/MAT_00000 written first) (with mats/MAT_00001 written first) (with mats/MAT_00002 written first) (with mats/MAT_00003 written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 13,770 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
programs/applyxfm4D-single-trilinearFLIRT's other programs-singlematrix with trilinear interpolation, verbose
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -singlematrix -interp trilinear -v (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-fourdigit-nnFLIRT's other programs--fourdigit (MAT_000n) with nearest neighbour (nn)
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz m4 --fourdigit --interp nn (with m4/MAT_0000 written first) (with m4/MAT_0001 written first) (with m4/MAT_0002 written first) (with m4/MAT_0003 written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-userprefix-splineFLIRT's other programs-userprefix with spline on an integer series (FSL-79)
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz mu -userprefix vol -interp spline (with mu/vol0000 written first) (with mu/vol0001 written first) (with mu/vol0002 written first) (with mu/vol0003 written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-float-sincFLIRT's other programsa float input (one volume), -singlematrix, sinc
command lineapplyxfm4D '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -singlematrix (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 25,096 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
programs/applyxfm4D-3dFLIRT's other programs-3D (the rectangular sinc of width 9) on a float volume
command lineapplyxfm4D '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -3D (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 147,498 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
programs/applyxfm4D-3d-spline-floatFLIRT's other programs-3D with spline on a float volume
command lineapplyxfm4D '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -3d -interp spline (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-int8FLIRT's other programsan int8 input (FSL-09: written as uint8), trilinear
command lineapplyxfm4D '<int8:epi.nii:epi_i8.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -singlematrix -interp trilinear (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-09 and names it: 163,840 values change

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, the clean room corrects FSL-09 and names it: 163,840 values change.

6/6
programs/applyxfm4D-uint8FLIRT's other programsa uint8 input, trilinear (FSL-111: its values 128 to 255 read as signed bytes; read as stored by default)
command lineapplyxfm4D '<uint8:epi.nii:epi_u8.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -singlematrix -interp trilinear (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-111 and names it: 63,871 values change

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, the clean room corrects FSL-111 and names it: 63,871 values change.

6/6
programs/applyxfm4D-uint8-3d-splineFLIRT's other programs-3D spline on a uint8 volume (FSL-111 with FSL-79's coefficients in the type)
command lineapplyxfm4D '<uint8:epi.nii:epi_u8.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -3D -interp spline (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-111 and names it: 84,973 values change

Open difference

Within the record’s thresholds, not bit-identical: out.nii.gz: correlation with FSL’s values (spline): 1.0. From the same matrix: the sinc and spline kernels’ last float bits (an open question of the specification). In fix, its default, the clean room corrects FSL-111, FSL-79 and names it: 84,972 values change.

6/6 · 7/7
programs/applyxfm4D-double-sincFLIRT's other programsa double series (sinc sums in double precision; the background value kept in double precision where the mask sets it)
command lineapplyxfm4D '<double:bold4.nii:bold4_f64.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz mats (with mats/MAT_00000 written first) (with mats/MAT_00001 written first) (with mats/MAT_00002 written first) (with mats/MAT_00003 written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: out.nii.gz: correlation with FSL’s values (sinc): 1.0. From the same matrix: the sinc and spline kernels’ last float bits (an open question of the specification). In fix, its default, 13,588 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 7/7
programs/applyxfm4D-double-3d-splineFLIRT's other programs-3D spline on a double volume (its coefficients in double precision)
command lineapplyxfm4D '<double:epi.nii:epi_f64.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -3D -interp spline (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-double-nnFLIRT's other programsa double volume, nearest neighbour (values through float)
command lineapplyxfm4D '<double:epi.nii:epi_f64.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -singlematrix -interp nn (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-3d-spline-intFLIRT's other programs-3D spline on an int16 volume (FSL-79: coefficients and values in the type)
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -3D -interp spline (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-nan-borderFLIRT's other programsNaN voxels in the input's border and an infinite one: the background value's NaN order (FSL-22), spline coefficients through them
command lineapplyxfm4D '<nanedge:epi.nii:epi_nanedge.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -singlematrix -interp spline (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, the clean room corrects FSL-22 and names it: 16,342 values change.

6/6
programs/applyxfm4D-nan-border-3dFLIRT's other programsthe same with -3D and sinc
command lineapplyxfm4D '<nanedge:epi.nii:epi_nanedge.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -3D (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 163,429 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
programs/applyxfm4D-metresFLIRT's other programsan input whose header gives metres (a nonsense header): read as FSL reads it
command lineapplyxfm4D '<metres:epi.nii:epi_m.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -singlematrix -interp trilinear (with one.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/applyxfm4D-bad-interpFLIRT's other programsan unknown interpolation (its message, exit 1)
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat -interp cubic (with one.mat written first)
Rejected by both

applyxfm4D exits 1; the same from the replica. interpolation method: cubic is not a valid choice. Please choose from nearestneighbour, trilinear, sinc, or spline.

Rejected by both

applyxfm4D exits 1; the same from the clean room. interpolation method: cubic is not a valid choice. Please choose from nearestneighbour, trilinear, sinc, or spline.

5/5 · 4/4
programs/applyxfm4D-userprefix-lastFLIRT's other programs-userprefix as the last word (std::string of null)
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz mu -userprefix (with mu/vol0000 written first) (with mu/vol0001 written first) (with mu/vol0002 written first) (with mu/vol0003 written first)
Rejected by both

applyxfm4D exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): basic_string: construction from null is not valid

Rejected by both

applyxfm4D exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): basic_string: construction from null is not valid

5/5 · 4/4
programs/applyxfm4D-missing-matrixFLIRT's other programsa directory without the matrices (FSL aborts)
command lineapplyxfm4D '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz none
Rejected by both

applyxfm4D exits -6; the same from the replica. Could not open matrix file none/MAT_00000 terminate called after throwing an instance of ‘std::logic_error’ what(): matrix multiplication: incompatible matrix dimensions: 0x0 and 4x4

Rejected by both

applyxfm4D exits -6; the same from the clean room. Could not open matrix file none/MAT_00000 terminate called after throwing an instance of ‘std::logic_error’ what(): matrix multiplication: incompatible matrix dimensions: 0x0 and 4x4

5/5 · 4/4
programs/applyxfm4D-missing-inputFLIRT's other programsan input that does not exist (FSL aborts)
command lineapplyxfm4D none.nii.gz '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' out.nii.gz one.mat (with one.mat written first)
Rejected by both

applyxfm4D exits -6; the same from the replica. Image Exception : #63 :: No image files match: none terminate called after throwing an instance of ‘std::runtime_error’ what(): No image files match: none

Rejected by both

applyxfm4D exits -6; the same from the clean room. Image Exception : #63 :: No image files match: none terminate called after throwing an instance of ‘std::runtime_error’ what(): No image files match: none

5/5 · 4/4
programs/aff2rigid-usageFLIRT's other programsaff2rigid without arguments (its usage, exit 1)
command linePATH=$FSLDIR/bin:/usr/bin:/bin aff2rigid
Rejected by both

aff2rigid exits 1; the same from the replica.

Rejected by both

aff2rigid exits 1; the same from the clean room.

5/5 · 4/4
programs/aff2rigid-defaultFLIRT's other programsaff2rigid of an affine standard-space matrix (numpy's arithmetic)
command linePATH=$FSLDIR/bin:/usr/bin:/bin aff2rigid aff.mat rigid.mat (with aff.mat written first)
Rejected by both

aff2rigid exits 1; the same from the replica. Traceback (most recent call last): File “~/fsl-6.0.7.23/bin/aff2rigid”, line 105, in v41=mat(zeros((4,1))) ^^^ NameError: name ‘mat’ is not defined

Rejected by both

aff2rigid exits 1; the same from the clean room. Traceback (most recent call last): File “~/fsl-6.0.7.23/bin/aff2rigid”, line 105, in v41=mat(zeros((4,1))) ^^^ NameError: name ‘mat’ is not defined

5/5
programs/aff2rigid-fmriprepFLIRT's other programsaff2rigid of fMRIPrep's initial matrix
command linePATH=$FSLDIR/bin:/usr/bin:/bin aff2rigid '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' rigid.mat
Rejected by both

aff2rigid exits 1; the same from the replica. Traceback (most recent call last): File “~/fsl-6.0.7.23/bin/aff2rigid”, line 105, in v41=mat(zeros((4,1))) ^^^ NameError: name ‘mat’ is not defined

Rejected by both

aff2rigid exits 1; the same from the clean room. Traceback (most recent call last): File “~/fsl-6.0.7.23/bin/aff2rigid”, line 105, in v41=mat(zeros((4,1))) ^^^ NameError: name ‘mat’ is not defined

5/5
programs/extracttxt-usageFLIRT's other programsextracttxt with one argument (its usage, exit 255)
command lineextracttxt word
Rejected by both

extracttxt exits 2; the same from the replica. ~/fsl-6.0.7.23/bin/extracttxt: 68: exit: Illegal number: -1

Rejected by both

extracttxt exits 2; the same from the clean room. ~/fsl-6.0.7.23/bin/extracttxt: 68: exit: Illegal number: -1

5/5
programs/extracttxt-restFLIRT's other programsthe lines from the first match to the end
command lineextracttxt value t.txt (with t.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/extracttxt-countFLIRT's other programsthree lines from the first match
command lineextracttxt e.4 t.txt 3 (with t.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/extracttxt-regex-startFLIRT's other programsa regular expression, two lines, starting 2 lines after
command lineextracttxt ^line.[5-6] t.txt 2 2 (with t.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/extracttxt-beforeFLIRT's other programsa negative relative start (lines before the match)
command lineextracttxt end$ t.txt 4 -3 (with t.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/extracttxt-zero-countFLIRT's other programszero trailing lines (sed's range ends before it starts)
command lineextracttxt line.7 t.txt 0 (with t.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/extracttxt-line0FLIRT's other programsa relative start reaching line 0 (sed's error, exit 1)
command lineextracttxt ^line.1.v t.txt 2 -1 (with t.txt written first)
Rejected by both

extracttxt exits 1; the same from the replica. sed: -e expression #1, char 4: invalid usage of line address 0

Rejected by both

extracttxt exits 1; the same from the clean room. sed: -e expression #1, char 4: invalid usage of line address 0

5/5
programs/extracttxt-dc-negativeFLIRT's other programsa start before line 1: dc reads its own negative result as a subtraction (stack empty), sed the range as an option
command lineextracttxt e.2 t.txt 3 -4 (with t.txt written first)
Rejected by both

extracttxt exits 1; the same from the replica. dc: stack empty sed: invalid option – ‘2’ Usage: sed [OPTION]… {script-only-if-no-other-script} [input-file]… -n, --quiet, --silent suppress automatic printing of pattern space --debug annotate p

Rejected by both

extracttxt exits 1; the same from the clean room. dc: stack empty sed: invalid option – ‘2’ Usage: sed [OPTION]… {script-only-if-no-other-script} [input-file]… -n, --quiet, --silent suppress automatic printing of pattern space --debug annotate p

5/5
programs/extracttxt-sed-optionFLIRT's other programsa start on line 0 with no trailing line: the range -1,-1 (sed's unknown option)
command lineextracttxt e.4 t.txt 0 -4 (with t.txt written first)
Rejected by both

extracttxt exits 1; the same from the replica. sed: invalid option – ‘1’ Usage: sed [OPTION]… {script-only-if-no-other-script} [input-file]… -n, --quiet, --silent suppress automatic printing of pattern space --debug annotate program execution

Rejected by both

extracttxt exits 1; the same from the clean room. sed: invalid option – ‘1’ Usage: sed [OPTION]… {script-only-if-no-other-script} [input-file]… -n, --quiet, --silent suppress automatic printing of pattern space --debug annotate program execution

5/5
programs/extracttxt-signedFLIRT's other programssigned arguments (+2, -0): dc reads each sign as a command
command lineextracttxt e.4 t.txt -0 +2 (with t.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/extracttxt-nomatchFLIRT's other programsa word that never matches (nothing printed, exit 0)
command lineextracttxt nothing t.txt (with t.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/extracttxt-missingFLIRT's other programsa file that does not exist (grep's message, exit 0)
command lineextracttxt x none.txt
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/pairreg-usageFLIRT's other programspairreg with four arguments (its usage, exit 1)
command linepairreg a b c d
Rejected by both

pairreg exits 1; the same from the replica.

Rejected by both

pairreg exits 1; the same from the clean room.

5/5 · 4/4
programs/pairreg-defaultFLIRT's other programspairreg of the T1w and its moved copy (brains and skulls the same)
command linepairreg '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1_moved.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1_moved.nii: derived from ds005040'"'"'s fMRIPrep run>' pair.mat -dof 6
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
programs/pairreg-metresFLIRT's other programspairreg of images whose headers give metres (nonsense headers)
command linepairreg '<metres:epi.nii:epi_m.nii: derived from ds005040'"'"'s fMRIPrep run>' '<metres:epi.nii:epi_m.nii: derived from ds005040'"'"'s fMRIPrep run>' '<metres:epi.nii:epi_m.nii: derived from ds005040'"'"'s fMRIPrep run>' '<metres:epi.nii:epi_m.nii: derived from ds005040'"'"'s fMRIPrep run>' pair.mat -dof 6
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6 · 5/5
programs/midtrans-usageFLIRT's other programsmidtrans without transforms (its message and usage, exit 1)
command linemidtrans
Rejected by both

midtrans exits 1; the same from the replica. Must specify at least 2 transforms to find the mid-transform Part of FSL (ID: 2412.6-dirty) midtrans Copyright© 2010, University of Oxford (Mark Jenkinson) Usage: midtrans [options] transform1 trans

Rejected by both

midtrans exits 1; the same from the clean room. Must specify at least 2 transforms to find the mid-transform Part of FSL (ID: 2412.6-dirty) midtrans Copyright© 2010, University of Oxford (Mark Jenkinson) Usage: midtrans [options] transform1 trans

5/5 · 4/4
programs/midtrans-helpFLIRT's other programsmidtrans -h (its usage, exit 1)
command linemidtrans -h
Rejected by both

midtrans exits 1; the same from the replica. Part of FSL (ID: 2412.6-dirty) midtrans Copyright© 2010, University of Oxford (Mark Jenkinson) Usage: midtrans [options] transform1 transform2 … transformN e.g. midtrans -o temp2mid.mat A2temp.mat

Rejected by both

midtrans exits 1; the same from the clean room. Part of FSL (ID: 2412.6-dirty) midtrans Copyright© 2010, University of Oxford (Mark Jenkinson) Usage: midtrans [options] transform1 transform2 … transformN e.g. midtrans -o temp2mid.mat A2temp.mat

5/5 · 4/4
programs/midtrans-oneFLIRT's other programsone transform only (at least 2 needed, exit 1)
command linemidtrans a.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first)
Rejected by both

midtrans exits 1; the same from the replica. Must specify at least 2 transforms to find the mid-transform Part of FSL (ID: 2412.6-dirty) midtrans Copyright© 2010, University of Oxford (Mark Jenkinson) Usage: midtrans [options] transform1 trans

Rejected by both

midtrans exits 1; the same from the clean room. Must specify at least 2 transforms to find the mid-transform Part of FSL (ID: 2412.6-dirty) midtrans Copyright© 2010, University of Oxford (Mark Jenkinson) Usage: midtrans [options] transform1 trans

5/5 · 4/4
programs/midtrans-defaultFLIRT's other programsthree transforms, printed (the MNI origin assumed)
command linemidtrans a.mat b.mat c.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/midtrans-out-verboseFLIRT's other programs-o with -v (every stage printed, the matrix written)
command linemidtrans -v -o mid.mat a.mat b.mat id.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit-identical: mid.mat: within 1e-9 of FSL’s in every element: 1e-10.

6/6
programs/midtrans-template-separateFLIRT's other programs--template (its centre of gravity) and --separate
command linemidtrans --template=@epi.nii --separate=img2mid -o mid.mat a.mat b.mat c.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/midtrans-template-50mmFLIRT's other programs--template with voxels of 50 mm (a nonsense header: its centre of gravity far out)
command linemidtrans --template=@const:6x5x4:50x50x50:t50.nii a.mat b.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/midtrans-debugFLIRT's other programs--debug (the origin at 0 and the debug matrices)
command linemidtrans --debug a.mat c.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/midtrans-singularFLIRT's other programsa transform with a determinant below 1e-6 (its message, exit 1)
command linemidtrans a.mat sing.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first) (with sing.mat written first)
Rejected by both

midtrans exits 1; the same from the replica. Could not read matrix sing.mat

Rejected by both

midtrans exits 1; the same from the clean room. Could not read matrix sing.mat

5/5 · 4/4
programs/midtrans-tokenFLIRT's other programsan option after the transforms (read as a file: an empty matrix, the size error caught)
command linemidtrans a.mat b.mat -v (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/midtrans-not-squareFLIRT's other programsa 3 x 4 matrix (the determinant's error, caught)
command linemidtrans a.mat r.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first) (with r.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/midtrans-3x3FLIRT's other programsa 3 x 3 matrix (the product's size error, caught)
command linemidtrans a.mat r.mat (with a.mat written first) (with b.mat written first) (with c.mat written first) (with id.mat written first) (with r.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-usageFLIRT's other programspointflirt without arguments (the compulsory options, exit 1)
command linepointflirt
Rejected by both

pointflirt exits 1; the same from the replica. *************************************************** The following COMPULSORY options have not been set: -i filename of input volume coordinates -r filename of reference volume coordinates ************

Rejected by both

pointflirt exits 1; the same from the clean room. *************************************************** The following COMPULSORY options have not been set: -i filename of input volume coordinates -r filename of reference volume coordinates ************

5/5 · 4/4
programs/pointflirt-defaultFLIRT's other programseight points, the matrix printed
command linepointflirt -i in.txt -r ref.txt (with in.txt written first) (with ref.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-out-verboseFLIRT's other programs-v and -o (the coordinates printed, the matrix written)
command linepointflirt -v -i in.txt -r ref.txt -o aff.mat (with in.txt written first) (with ref.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-voxFLIRT's other programs--vox with --invol and --refvol (voxel coordinates scaled)
command linepointflirt -i in.txt -r ref.txt --vox --invol=@epi.nii --refvol=@t1.nii (with in.txt written first) (with ref.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-vox-oddFLIRT's other programs--vox with voxels of 0.01 mm and of 50 mm (nonsense headers)
command linepointflirt -i in.txt -r ref.txt --vox --invol=@const:4x4x4:0.01x0.01x0.01:v001.nii --refvol=@const:4x4x4:50x50x50:v50.nii (with in.txt written first) (with ref.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-vox-missingFLIRT's other programs--vox without the volumes (its error, exit 2)
command linepointflirt -i in.txt -r ref.txt --vox (with in.txt written first) (with ref.txt written first)
Rejected by both

pointflirt exits 2; the same from the replica. ERROR: need to specify --invol and --refvol to use --vox

Rejected by both

pointflirt exits 2; the same from the clean room. ERROR: need to specify --invol and --refvol to use --vox

5/5 · 4/4
programs/pointflirt-nullspaceFLIRT's other programsfour points (a nullspace: the warning, a non-unique solution)
command linepointflirt -i in4.txt -r ref4.txt (with in4.txt written first) (with ref4.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-countFLIRT's other programsfiles with different numbers of points (its error, exit 255)
command linepointflirt -i in.txt -r ref4.txt (with in.txt written first) (with ref.txt written first) (with ref4.txt written first)
Rejected by both

pointflirt exits 255; the same from the replica. ERROR: Different number of coordinates in the two files

Rejected by both

pointflirt exits 255; the same from the clean room. ERROR: Different number of coordinates in the two files

5/5 · 4/4
programs/pointflirt-manyFLIRT's other programstwenty points (x·xᵀ by BLAS dsyrk)
command linepointflirt -i in20.txt -r ref20.txt (with in20.txt written first) (with ref20.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-missingFLIRT's other programsa coordinate file that does not exist (FSL aborts)
command linepointflirt -i none.txt -r ref.txt (with in.txt written first) (with ref.txt written first)
Rejected by both

pointflirt exits 255; the same from the replica. Could not open matrix file none.txt ERROR: Different number of coordinates in the two files

Rejected by both

pointflirt exits 255; the same from the clean room. Could not open matrix file none.txt ERROR: Different number of coordinates in the two files

5/5 · 4/4
programs/pointflirt-no-pointsFLIRT's other programstwo coordinate files that do not exist (no points: FSL crashes, FSL-204; refused in fix and in reproduce)
command linepointflirt -i none1.txt -r none2.txt
Expected divergence

FSL crashes (FSL-204: undefined behaviour); the replica refuses in fix and in reproduce

Expected divergence

pointflirt given two coordinate files that do not exist (no points), where FSL crashes (FSL-204, registered on 2026-11-03): since 2026-11-05 the clean room refuses the call in fix and in reproduce, naming FSL-204 (outside the record before). FSL crashes (FSL-204: undefined behaviour); the clean room refuses in fix and in reproduce

3/3 · 2/2
programs/pointflirt-threeFLIRT's other programsthree points: a square file is read with points as columns
command linepointflirt -i in3.txt -r ref3.txt (with in3.txt written first) (with ref3.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6 · 5/5
programs/pointflirt-twoFLIRT's other programstwo points (two rows of three: FSL reads past its rows, aborts)
command linepointflirt -i in2.txt -r ref2.txt (with in2.txt written first) (with ref2.txt written first)
Rejected by both

pointflirt exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): copy into submatrix: incompatible matrix dimensions: 3x1 and 2x1

Rejected by both

pointflirt exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): copy into submatrix: incompatible matrix dimensions: 3x1 and 2x1

5/5 · 4/4
programs/pointflirt-four-columnsFLIRT's other programspoints of four coordinates (FSL's block sizes clash, aborts)
command linepointflirt -i in4c.txt -r ref4c.txt (with in4c.txt written first) (with ref4c.txt written first)
Rejected by both

pointflirt exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): copy into submatrix: incompatible matrix dimensions: 3x1 and 4x1

Rejected by both

pointflirt exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): copy into submatrix: incompatible matrix dimensions: 3x1 and 4x1

5/5 · 4/4
programs/avscale-usageFLIRT's other programsavscale without arguments (its usage, exit 255)
command lineavscale
Rejected by both

avscale exits 255; the same from the replica. Usage: ~/fsl-6.0.7.23/bin/avscale [–allparams/–inverteddies] matrixfile [non-reference-volume]

Rejected by both

avscale exits 255; the same from the clean room. Usage: ~/fsl-6.0.7.23/bin/avscale [–allparams/–inverteddies] matrixfile [non-reference-volume]

5/5
programs/avscale-defaultFLIRT's other programsavscale of fMRIPrep's initial matrix
command lineavscale '<init.mat: derived from ds005040'"'"'s fMRIPrep run>'
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/avscale-allparamsFLIRT's other programs--allparams, the translations about a volume's centre
command lineavscale --allparams aff.mat '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' (with aff.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/avscale-inverteddiesFLIRT's other programs--inverteddies (the check and the inverted eddy matrix)
command lineavscale --inverteddies aff.mat (with aff.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/avscale-flipFLIRT's other programsa matrix that swaps left and right (the half transforms of the swap)
command lineavscale flip.mat (with flip.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/avscale-singularFLIRT's other programsa singular matrix: Armadillo's solve throws, avscale aborts
command lineavscale sing.mat (with sing.mat written first)
Rejected by both

avscale exits -6; the same from the replica. matrix multiplication: problem with matrix inverse; suggest to use solve() instead terminate called after throwing an instance of ‘std::runtime_error’ what(): matrix multiplication: problem with matri

Rejected by both

avscale exits -6; the same from the clean room. matrix multiplication: problem with matrix inverse; suggest to use solve() instead terminate called after throwing an instance of ‘std::runtime_error’ what(): matrix multiplication: problem with matri

5/5
programs/avscale-missing-matrixFLIRT's other programsa matrix file that does not exist (exit 254)
command lineavscale none.mat
Rejected by both

avscale exits 254; the same from the replica. Could not open matrix file none.mat

Rejected by both

avscale exits 254; the same from the clean room. Could not open matrix file none.mat

5/5
programs/avscale-empty-volumeFLIRT's other programsan empty volume name: the centre of an empty volume
command lineavscale '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' ''
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/avscale-missing-volumeFLIRT's other programsa volume that does not exist: newimage's exception, abort
command lineavscale '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' none.nii.gz
Rejected by both

avscale exits -6; the same from the replica. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii.gz Error : No image files match: none Failed to read volume none.nii.gz Error : No image fil

Rejected by both

avscale exits -6; the same from the clean room. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii.gz Error : No image files match: none Failed to read volume none.nii.gz Error : No image fil

5/5
programs/img2imgcoord-usageFLIRT's other programsimg2imgcoord without arguments (its usage, exit 1)
command lineimg2imgcoord
Rejected by both

img2imgcoord exits 1; the same from the replica.

Rejected by both

img2imgcoord exits 1; the same from the clean room.

5/5
programs/img2imgcoord-xfmFLIRT's other programsimg2imgcoord through fMRIPrep's initial matrix
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2imgcoord-voxFLIRT's other programsimg2imgcoord -vox
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -vox coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2imgcoord-mmFLIRT's other programsimg2imgcoord -mm
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -mm coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2imgcoord-verboseFLIRT's other programsimg2imgcoord verbose (sizes, matrices; fixed notation after them)
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -v coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2imgcoord-missing-coordsFLIRT's other programsimg2imgcoord: a coordinate file that does not exist
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' none.txt
Rejected by both

img2imgcoord exits 255; the same from the replica. Could not open matrix file none.txt

Rejected by both

img2imgcoord exits 255; the same from the clean room. Could not open matrix file none.txt

5/5
programs/img2imgcoord-not-numbersFLIRT's other programsimg2imgcoord: coordinates that stop at a word
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' bad.txt (with bad.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2imgcoord-lackingFLIRT's other programsimg2imgcoord: an option lacking its argument (a message, the run goes on)
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt -xfm (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2imgcoord-unknownFLIRT's other programsimg2imgcoord: an unknown option (exit 255)
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -frob x coords.txt (with coords.txt written first)
Rejected by both

img2imgcoord exits 255; the same from the replica. Unrecognised option -frob

Rejected by both

img2imgcoord exits 255; the same from the clean room. Unrecognised option -frob

5/5
programs/img2imgcoord-missing-imageFLIRT's other programsimg2imgcoord: an image that does not exist (abort)
command lineimg2imgcoord -src none.nii -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Rejected by both

img2imgcoord exits -6; the same from the replica. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

Rejected by both

img2imgcoord exits -6; the same from the clean room. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

5/5
programs/std2imgcoord-usageFLIRT's other programsstd2imgcoord without arguments (its usage, exit 1)
command linestd2imgcoord
Rejected by both

std2imgcoord exits 1; the same from the replica.

Rejected by both

std2imgcoord exits 1; the same from the clean room.

5/5
programs/std2imgcoord-xfmFLIRT's other programsstd2imgcoord through fMRIPrep's initial matrix
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-voxFLIRT's other programsstd2imgcoord -vox
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -vox coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-mmFLIRT's other programsstd2imgcoord -mm
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -mm coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-verboseFLIRT's other programsstd2imgcoord verbose (sizes, matrices; fixed notation after them)
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -verbose coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-missing-coordsFLIRT's other programsstd2imgcoord: a coordinate file that does not exist
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' none.txt
Rejected by both

std2imgcoord exits 255; the same from the replica. Could not open matrix file none.txt

Rejected by both

std2imgcoord exits 255; the same from the clean room. Could not open matrix file none.txt

5/5
programs/std2imgcoord-not-numbersFLIRT's other programsstd2imgcoord: coordinates that stop at a word
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' bad.txt (with bad.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-lackingFLIRT's other programsstd2imgcoord: an option lacking its argument (a message, the run goes on)
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt -xfm (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-unknownFLIRT's other programsstd2imgcoord: an unknown option (exit 255)
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -frob x coords.txt (with coords.txt written first)
Rejected by both

std2imgcoord exits 255; the same from the replica. Unrecognised option -frob

Rejected by both

std2imgcoord exits 255; the same from the clean room. Unrecognised option -frob

5/5
programs/std2imgcoord-missing-imageFLIRT's other programsstd2imgcoord: an image that does not exist (abort)
command linestd2imgcoord -img none.nii -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Rejected by both

std2imgcoord exits -6; the same from the replica. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

Rejected by both

std2imgcoord exits -6; the same from the clean room. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

5/5
programs/img2stdcoord-usageFLIRT's other programsimg2stdcoord without arguments (its usage, exit 1)
command lineimg2stdcoord
Rejected by both

img2stdcoord exits 1; the same from the replica.

Rejected by both

img2stdcoord exits 1; the same from the clean room.

5/5
programs/img2stdcoord-xfmFLIRT's other programsimg2stdcoord through fMRIPrep's initial matrix
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2stdcoord-voxFLIRT's other programsimg2stdcoord -vox
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -vox coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2stdcoord-mmFLIRT's other programsimg2stdcoord -mm
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -mm coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2stdcoord-verboseFLIRT's other programsimg2stdcoord verbose (sizes, matrices; fixed notation after them)
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -verbose coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2stdcoord-missing-coordsFLIRT's other programsimg2stdcoord: a coordinate file that does not exist
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' none.txt
Rejected by both

img2stdcoord exits 255; the same from the replica. Could not open matrix file none.txt

Rejected by both

img2stdcoord exits 255; the same from the clean room. Could not open matrix file none.txt

5/5
programs/img2stdcoord-not-numbersFLIRT's other programsimg2stdcoord: coordinates that stop at a word
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' bad.txt (with bad.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2stdcoord-lackingFLIRT's other programsimg2stdcoord: an option lacking its argument (a message, the run goes on)
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt -xfm (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2stdcoord-unknownFLIRT's other programsimg2stdcoord: an unknown option (exit 255)
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -frob x coords.txt (with coords.txt written first)
Rejected by both

img2stdcoord exits 255; the same from the replica. Unrecognised option -frob

Rejected by both

img2stdcoord exits 255; the same from the clean room. Unrecognised option -frob

5/5
programs/img2stdcoord-missing-imageFLIRT's other programsimg2stdcoord: an image that does not exist (abort)
command lineimg2stdcoord -img none.nii -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Rejected by both

img2stdcoord exits -6; the same from the replica. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

Rejected by both

img2stdcoord exits -6; the same from the clean room. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

5/5
programs/img2imgcoord-flirt-prematFLIRT's other programsimg2imgcoord -flirt and -premat
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -flirt -premat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-standardFLIRT's other programsstd2imgcoord from FSL's MNI152 2 mm
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<fsldata:MNI152_T1_2mm_brain_mask.nii.gz: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-no-stdFLIRT's other programsstd2imgcoord without -std (its warning; 1 mm voxels)
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/std2imgcoord-singularFLIRT's other programsstd2imgcoord with a singular matrix: its inverse throws
command linestd2imgcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -xfm sing.mat coords.txt (with coords.txt written first) (with sing.mat written first)
Rejected by both

std2imgcoord exits -6; the same from the replica. terminate called after throwing an instance of ‘std::runtime_error’ what(): inv(): matrix is singular

Rejected by both

std2imgcoord exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::runtime_error’ what(): inv(): matrix is singular

5/5
programs/img2stdcoord-noformFLIRT's other programsimg2stdcoord to a standard image without sform or qform
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<noform:5x6x7:2x3x4:noform.nii: derived from ds005040'"'"'s fMRIPrep run>' coords.txt (with coords.txt written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/img2stdcoord-warp-missingFLIRT's other programsimg2stdcoord -warp of a file that does not exist
command lineimg2stdcoord -img '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -warp none.nii.gz coords.txt (with coords.txt written first)
Rejected by both

img2stdcoord exits 1; the same from the replica. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii.gz Error : No image files match: none An error occured while reading file: none.nii.gz

Rejected by both

img2stdcoord exits 1; the same from the clean room. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii.gz Error : No image files match: none An error occured while reading file: none.nii.gz

5/5
programs/rmsdiff-usageFLIRT's other programsrmsdiff without arguments (its usage, exit 255)
command linermsdiff
Rejected by both

rmsdiff exits 255; the same from the replica. Usage: ~/fsl-6.0.7.23/bin/rmsdiff matrixfile1 matrixfile2 refvol [mask] Outputs rms deviation between matrices (in mm)

Rejected by both

rmsdiff exits 255; the same from the clean room. Usage: ~/fsl-6.0.7.23/bin/rmsdiff matrixfile1 matrixfile2 refvol [mask] Outputs rms deviation between matrices (in mm)

5/5
programs/rmsdiff-rmsFLIRT's other programsthe RMS deviation over 80 mm about a volume's centre
command linermsdiff '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' id.mat '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' (with id.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/rmsdiff-maskFLIRT's other programsthe largest and the RMS displacement over a mask
command linermsdiff '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' aff.mat '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' (with aff.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/rmsdiff-low-determinantFLIRT's other programsa matrix of low determinant (its warning, the matrix on standard error)
command linermsdiff small.mat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' (with small.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/rmsdiff-singularFLIRT's other programsa singular second matrix: rms_deviation's message, exit 251
command linermsdiff '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' sing.mat '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' (with sing.mat written first)
Rejected by both

rmsdiff exits 251; the same from the replica. WARNING:: matrix 2 has low determinant 1.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 0.000000 1.000000 RMS_DEVIATION ERR

Rejected by both

rmsdiff exits 251; the same from the clean room. WARNING:: matrix 2 has low determinant 1.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 0.000000 1.000000 RMS_DEVIATION ERR

5/5
programs/rmsdiff-missing-matrixFLIRT's other programsa matrix file that does not exist
command linermsdiff none.mat '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>'
Rejected by both

rmsdiff exits 254; the same from the replica. Could not open matrix file none.mat Could not read matrix none.mat

Rejected by both

rmsdiff exits 254; the same from the clean room. Could not open matrix file none.mat Could not read matrix none.mat

5/5
programs/rmsdiff-4dFLIRT's other programsa 4D reference (its first volume, the warning)
command linermsdiff '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' aff.mat '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' (with aff.mat written first)
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-usageFLIRT's other programsmakerot without arguments (the compulsory -t, its usage, exit 1)
command linemakerot
Rejected by both

makerot exits 1; the same from the replica. *************************************************** The following COMPULSORY options have not been set: -t,–theta angle of rotation (in degrees) *************************************************** Pa

Rejected by both

makerot exits 1; the same from the clean room. *************************************************** The following COMPULSORY options have not been set: -t,–theta angle of rotation (in degrees) *************************************************** Pa

5/5
programs/makerot-helpFLIRT's other programsmakerot -h (its usage, exit 1)
command linemakerot -h
Rejected by both

makerot exits 1; the same from the replica. Part of FSL (ID: 2412.6-dirty) makerot Copyright© 2003, University of Oxford (Mark Jenkinson) Usage: makerot [options] --theta=angle Compulsory arguments (You MUST set one or more of): -t,–theta an

Rejected by both

makerot exits 1; the same from the clean room. Part of FSL (ID: 2412.6-dirty) makerot Copyright© 2003, University of Oxford (Mark Jenkinson) Usage: makerot [options] --theta=angle Compulsory arguments (You MUST set one or more of): -t,–theta an

5/5
programs/makerot-defaultFLIRT's other programs30 degrees about z through the origin
command linemakerot -t 30
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-axis-centreFLIRT's other programsan axis that is not of unit length (FSL-76) through a centre
command linemakerot -t 30 -a 1,2,3 -c 4,5,6
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-longFLIRT's other programsthe long options with their values after =
command linemakerot --theta=-45.5 --axis=0,0,4 --centre=1:2:3
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-covFLIRT's other programsthe centre of an image's volume (--cov)
command linemakerot -t 20 -a 0,1,1 --cov=@epi.nii
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-cov-missingFLIRT's other programs--cov of a file that is not an image (FSL aborts)
command linemakerot -t 20 --cov=none.nii
Rejected by both

makerot exits -6; the same from the replica. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

Rejected by both

makerot exits -6; the same from the clean room. Image Exception : #63 :: No image files match: none Image Exception : #22 :: Failed to read volume none.nii Error : No image files match: none terminate called after throwing an instance of 'std::runt

5/5
programs/makerot-zero-axisFLIRT's other programsa zero axis (its error, exit 1)
command linemakerot -t 10 -a 0,0,0
Rejected by both

makerot exits 1; the same from the replica. ERROR:: Axis cannot be zero

Rejected by both

makerot exits 1; the same from the clean room. ERROR:: Axis cannot be zero

5/5
programs/makerot-short-axisFLIRT's other programsan axis of one value (FSL-77: the buffer's zeros)
command linemakerot -t 30 -a 5
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-stale-axisFLIRT's other programsthe axis given twice, the second of one value (FSL-77: the first's values)
command linemakerot -t 30 -a 1,2,3 -a 7
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-bad-thetaFLIRT's other programsa theta that is not a number (Couldn't set_value, exit 1)
command linemakerot -t abc
Rejected by both

makerot exits 1; the same from the replica. Part of FSL (ID: 2412.6-dirty) makerot Copyright© 2003, University of Oxford (Mark Jenkinson) Usage: makerot [options] --theta=angle Compulsory arguments (You MUST set one or more of): -t,–theta an

Rejected by both

makerot exits 1; the same from the clean room. Part of FSL (ID: 2412.6-dirty) makerot Copyright© 2003, University of Oxford (Mark Jenkinson) Usage: makerot [options] --theta=angle Compulsory arguments (You MUST set one or more of): -t,–theta an

5/5
programs/makerot-outFLIRT's other programsthe matrix written to a file (-o)
command linemakerot -t 30 -a 1,1,0 -o out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 9 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
programs/makerot-out-unwritableFLIRT's other programsan output file that cannot be written (the message, exit 0)
command linemakerot -t 30 -o nodir/out.mat
Bit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
programs/makerot-tokenFLIRT's other programsa word that is not an option (unrecognised token, exit 1)
command linemakerot -t 30 extra
Rejected by both

makerot exits 1; the same from the replica. Part of FSL (ID: 2412.6-dirty) makerot Copyright© 2003, University of Oxford (Mark Jenkinson) Usage: makerot [options] --theta=angle Compulsory arguments (You MUST set one or more of): -t,–theta an

Rejected by both

makerot exits 1; the same from the clean room. Part of FSL (ID: 2412.6-dirty) makerot Copyright© 2003, University of Oxford (Mark Jenkinson) Usage: makerot [options] --theta=angle Compulsory arguments (You MUST set one or more of): -t,–theta an

5/5
difftest/applyxfm-blur-extraslicerandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
difftest/applyxfm4d-double-paddingrandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
difftest/applyxfm4d-int-spline-smallrandom testingRejected by both

applyxfm4D exits -11; the same from the replica.

–2/2
difftest/applyxfm4d-nan-backgroundrandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
difftest/background-level-overflow-survivesrandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
difftest/background-nan-orderrandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
difftest/basescale-init-floatrandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
difftest/bbr-no-boundary-points-nanrandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
difftest/extracttxt-dc-negativerandom testingRejected by both

extracttxt exits 1; the same from the replica. dc: stack empty sed: invalid option – ‘2’ Usage: sed [OPTION]… {script-only-if-no-other-script} [input-file]… -n, --quiet, --silent suppress automatic printing of pattern space --debug annotate p

–5/5
difftest/paramsubset-singular-costrandom testingBit-identical

every file FSL writes, byte for byte, with FSL’s exit status and messages

–6/6
verbose/6-search-debugsaveverbose levels-verbose 6 -debugsave on the search alone: its inner level 4 and the grids it saves
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 6 -debugsave -omat out.mat -schedule s.sch (with s.sch written first)
–Open difference

Within the record’s thresholds, not bit-identical: costs.nii.gz: the same grid and header, correlation 0.999 (or within 1e-4 when constant): 1.0; sc.nii.gz: the same grid and header, correlation 0.999 (or within 1e-4 when constant): 0.0; tx.nii.gz: the same grid and header, correlation 0.999 (or within 1e-4 when constant): 1.0; ty.nii.gz: the same grid and header, correlation 0.999 (or within 1e-4 when constant): 1.0.

10/10
verbose/7verbose levels-verbose 7 on the default schedule: the search's evaluations, starts and tolerances
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 7 -dof 6 -omat out.mat
–Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ. Printed numbers differ in their last digits. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

5/5
verbose/5-optimisersverbose levels-verbose 5: gridmeasurecost, Brent, Powell's correction, a parameter subset
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 5 -omat out.mat -schedule s.sch (with s.sch written first)
–Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ. Printed numbers differ in their last digits.

5/5
verbose/7-subsetverbose levels-verbose 7: Brent's and a subset's starts, tolerances and doubled evaluations
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 7 -omat out.mat -schedule s.sch (with s.sch written first)
–Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ. Printed numbers differ in their last digits.

5/5
verbose/3-scales-optionsverbose levels-verbose 3: forced and input-only scales (the evaluation counts), every setoption echoed, a short tolerance list's index error
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 3 -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
verbose/2-noclampverbose levels-verbose 2 -noclamp: the images' ranges unclamped
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 2 -noclamp -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
verbose/2-noresampleverbose levels-verbose 2 -noresample: no subsampling lines
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 2 -noresample -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
verbose/2-applyxfmverbose levels-verbose 2 -applyxfm: the images read and the resampled volume
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 2 -applyxfm -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -out out
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
verbose/2-outverbose levels-verbose 2 -out without -omat: the transformed input, then the final result
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 2 -out out -schedule s.sch (with s.sch written first)
–Open difference

Within the record’s thresholds, not bit-identical: out.nii.gz: correlation with FSL’s image inside the reference’s mask: 1.0.

6/6
verbose/3-2dverbose levels-2D -verbose 3: the tripled slice, no sform messages
command lineflirt -in '<slice_in.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<slice_ref.nii: derived from ds005040'"'"'s fMRIPrep run>' -2D -verbose 3 -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
verbose/3-debugverbose levels-verbose 3 -debug: each setscale's debug lines before the input's
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 3 -debug -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

11/11
debug/scales-debug-debug over forced and input-only scales: the pair's input at each setscale
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -debug -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

11/11
verbose/2-basescale-trverbose levels-verbose 2 with a base scale (0.6 mm voxels) and a 4D input of TR 2: the sampling and initial matrices in scaled mm, the levels' TR
command lineflirt -in '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1_small.nii: derived from ds005040'"'"'s fMRIPrep run>' -init '<init.mat: derived from ds005040'"'"'s fMRIPrep run>' -verbose 2 -omat out.mat -schedule s.sch (with s.sch written first)
–Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ. Printed numbers differ in their last digits.

5/5
verbose/3-empty-referenceverbose levels-verbose 3 with an all-zero reference: the centre's COG warning where its line prints
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<zero:t1.nii:zeroweight.nii: derived from ds005040'"'"'s fMRIPrep run>' -verbose 3 -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
real/anglerep-quaternion-dof3real registrations-anglerep quaternion at 3 dof: FSL-90's starts and the dof message start by start, the invalid starts the identity (§8.1 step 9)
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -anglerep quaternion -dof 3 -omat out.mat
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
programs/pointflirt-no-points-verboseFLIRT's other programspointflirt -v without points: FSL's lines and headings, then its crash (FSL-204; refused in fix and in reproduce)
command linepointflirt -v -i none1.txt -r none2.txt
–Expected divergence

FSL crashes (FSL-204: undefined behaviour); the clean room refuses in fix and in reproduce

2/2
programs/pointflirt-nan-wordFLIRT's other programsa nan word in a coordinate file reads 0 (§2.2): the fit of the points as read
command linepointflirt -i in.txt -r ref.txt (with in.txt written first) (with ref.txt written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
programs/pointflirt-overflowFLIRT's other programsa coordinate whose square overflows: FSL's SVD fails (signal 6; the clean room exits 1)
command linepointflirt -i in.txt -r ref.txt (with in.txt written first) (with ref.txt written first)
–Rejected by both

pointflirt exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds

4/4
programs/img2imgcoord-warp-3dFLIRT's other programsimg2imgcoord -warp of a 3D image, verbose: the lines so far, FSL's read error, exit 1 (§15.2: exactly three volumes)
command lineimg2imgcoord -src '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -dest '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -warp '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -v c.txt (with c.txt written first)
–Rejected by both

img2imgcoord exits 1; the same from the clean room. An error occured while reading file: epi.nii

5/5
programs/std2imgcoord-warp-4dFLIRT's other programsstd2imgcoord -warp of a 4-volume series: FSL's read error, exit 1
command linestd2imgcoord -img '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -std '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -warp '<bold4.nii: derived from ds005040'"'"'s fMRIPrep run>' c.txt (with c.txt written first)
–Rejected by both

std2imgcoord exits 1; the same from the clean room. An error occured while reading file: bold4.nii

5/5
programs/pairreg-verboseFLIRT's other programspairreg -v at 12 dof: every line of the three schedules the clean room carries (§15.8), echoed as FSL's
command linepairreg '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1_moved.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' '<t1_moved.nii: derived from ds005040'"'"'s fMRIPrep run>' pair.mat -v
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 12 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

5/5
debug/bbr-no-boundary-debug-debug -cost bbr with an all-zero white-matter segmentation (the clean-room catch-up, its step 4): the set-up's lines and error at every evaluation, NaN costs that move nothing, the zero debug volume, -omat and -out written
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -cost bbr -wmseg '<zero:wm.nii:zerowm.nii: derived from ds005040'"'"'s fMRIPrep run>' -debug -out out -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

13/13
debug/noresample-debug-debug -noresample (the clean-room catch-up, its step 4): the levels are the reference itself, its forms as read; no global_refweight1 without weights
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -noresample -debug -omat out.mat -schedule s.sch (with s.sch written first)
–Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ. Printed numbers differ in their last digits.

10/10
schedule/perturb-scaledschedulesmeasurecost and gridmeasurecost of a scaled and skewed row at 6, 7 and 12 dof (the clean-room catch-up, its step 4): a perturbed row is composed again with all twelve parameters, whatever the command's dof
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
out/display-range-out's header-applyxfm onto FSL's MNI152_T1_2mm (display range 3000 to 8000): -out's range 0 and 0, as FSL writes it whatever the reference's (the clean-room catch-up, its step 5)
command lineflirt -in '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref $FSLDIR/data/standard/MNI152_T1_2mm -applyxfm -init eye.mat -out out (with eye.mat written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
debug/bbr-setoption-debug-debug -wmseg: the boundary-based cost set up by setoption and by a new pair
command lineflirt -in '<epi.nii: derived from ds005040'"'"'s fMRIPrep run>' -ref '<t1.nii: derived from ds005040'"'"'s fMRIPrep run>' -wmseg '<wm.nii: derived from ds005040'"'"'s fMRIPrep run>' -debug -omat out.mat -schedule s.sch (with s.sch written first)
–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

11/11
programs/midtrans-overflowFLIRT's other programsmidtrans of a transform of 1e200 scales (its determinant overflows: used, §15.5): FSL writes rows of NaN; refused (FSL-241)–Expected divergence

midtrans writes rows of NaN with exit status 0 (FSL-241, rule B); the clean room refuses in fix and in reproduce

3/3
programs/midtrans-overflow-svdFLIRT's other programsmidtrans of a translation of 1e300: FSL's decomposition fails and it reports a matrix-size error, exit 0; refused (FSL-241)–Expected divergence

midtrans reports its failed decomposition as a matrix-size error with exit status 0 (FSL-241, rule B); the clean room refuses in fix and in reproduce

3/3
difftest/applyxfm4d-int32-paddingrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 2,286 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
difftest/applyxfm4d-spline-doublerandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
difftest/applyxfm4d-spline-floatrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

6/6
difftest/debug-bbr-first-pairrandom testing–Open difference

Within the record’s thresholds, not bit-identical: debug.nii.zst: correlation with FSL’s image inside the reference’s mask: 1.0.

13/13
difftest/debug-bbr-no-boundaryrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 13,586 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

13/13
difftest/debug-empty-weight-qfacrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 2,568 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

12/12
difftest/debug-neurological-levelsrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

12/12
difftest/debug-noresamplerandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

10/10
difftest/nan-matrices-char-dof7random testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 16 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

5/5
difftest/nan-matrices-forcescalingrandom testing–Rejected by both

flirt exits 1; the same from the clean room. ERROR:: Unrecognised value (ANALYZE) of environment variable FSLOUTPUTTYPE Legal values are: NIFTI, NIFTI_PAIR, NIFTI2, NIFTI2_PAIR, NIFTI_GZ, NIFTI_PAIR_GZ, NIFTI2_GZ, NIFTI2_PAIR_GZ, NIFTI_ZST, NIFT

4/4
difftest/nan-matrices-subsetrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 16 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
difftest/nan-matrix-paddedrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 5,898 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
difftest/out-completed-sform-negative-zerorandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages. In fix, its default, 191 values change, from sinf, cosf, atan2f, cos, exp and logf computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

6/6
difftest/pairreg-search-equal-costs-sortedrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5
difftest/search-cost-minima-fallbackrandom testing–Open difference

Within the record’s thresholds, not bit-identical: stdout: the same lines (numbers aside): numbers differ. Printed numbers differ in their last digits.

6/6
difftest/search-equal-costs-sortedrandom testing–Bit-identical

Every file FSL writes, byte for byte (images: every value and header field, descrip aside), with FSL’s exit status and messages.

5/5

Speed

Wall time of the whole command (reading, registering, writing), one run each, on fMRIPrep’s call on the record’s three fMRIPrep runs and on the record’s real registrations (benchmark). These timings are informal: the machine was running other work beside them, and the clean room was timed at its part 2’s third step, before the specification’s answers changed its arithmetic. FSL and the replica run on one thread; the clean room on one and on 14, with the same result at any thread count.

The clean room’s boundary-based cost computes the boundary points’ terms in parallel. It runs fMRIPrep’s call 1.2 to 1.3 times as fast as FSL on one thread and 12 to 14 times on 14, and the field map’s search 3 times as fast on one thread: it skips the steps that cannot reach the mask. The small correlation-ratio registrations are line searches of a few tens of thousands of points, one evaluation after the other: on one thread at FSL’s speed (0.9 to 1 times), 3 to 4 times as fast on 14. SIMD sampling is the next step.

Speed, both ports against the original: FSL flirt 6.0.7.23

Seconds for the whole command (reading, registering, writing), one run each, on a log scale. The factor is how many times faster the clean room is on 14 threads than FSL on one.

  • FSL flirt 6.0.7.23, 1 thread
  • replica, 1 thread
  • clean room, 1 thread
  • clean room, 14 threads
  1. fMRIPrep's BBR call, ds005040FSL flirt 6.0.7.23, 1 thread 46.7 sreplica, 1 thread 56.9 sclean room, 1 thread 38.7 sclean room, 14 threads 3.76 s12×
  2. fMRIPrep's BBR call, ds003592FSL flirt 6.0.7.23, 1 thread 35.4 sreplica, 1 thread 43 sclean room, 1 thread 28 sclean room, 14 threads 2.71 s13×
  3. fMRIPrep's BBR call, ds000228FSL flirt 6.0.7.23, 1 thread 18.3 sreplica, 1 thread 22.6 sclean room, 1 thread 14.7 sclean room, 14 threads 1.35 s14×
  4. default 12 dof (BOLD reference to T1w)FSL flirt 6.0.7.23, 1 thread 0.969 sreplica, 1 thread 1.31 sclean room, 1 thread 1.1 sclean room, 14 threads 0.302 s3.2×
  5. 6 dofFSL flirt 6.0.7.23, 1 thread 0.723 sreplica, 1 thread 0.784 sclean room, 1 thread 0.694 sclean room, 14 threads 0.201 s3.6×
  6. BBR with a field map and its mask (-pedir -2)FSL flirt 6.0.7.23, 1 thread 256 sreplica, 1 thread 269 sclean room, 1 thread 80.8 sclean room, 14 threads 7.04 s36×
  7. -applyxfm (trilinear)FSL flirt 6.0.7.23, 1 thread 0.221 sreplica, 1 thread 0.21 sclean room, 1 thread 0.115 sclean room, 14 threads 0.0392 s5.6×
0.01 s0.1 s1 s10 s100 s1000 s

From the benchmark report docs/benchmarks/fsl-flirt.json (AMD Ryzen 9 7945HX with Radeon Graphics, 32 logical CPUs, shared with other runs, so the times may be a little long; each program in FSL’s clean environment, outputs NIfTI-1 gzipped; informal: one run each, LARMORX at commit 04f5462, part 2’s third step, before the specification’s answers changed the arithmetic: not timed again).

Corrections by default

Where FSL has a deterministic defect, both ports correct it by default, fix, and warn, in the same way unless said otherwise below. upstream_defects="reproduce" (on the command line --upstream-defects reproduce from the next release, LARMORX_UPSTREAM_DEFECTS=reproduce in 0.1.0rc1) gives FSL’s behaviour: the replica’s byte for byte, the clean room’s within its record’s thresholds (register).

  • FLIRT ends with exit status 0 after any error past its command line (an image or matrix it cannot read, a matrix of the wrong size, an unknown -bbrtype). LARMORX exits 1 with FLIRT’s message, and 0 in reproduce (FSL-68).
  • What FSL leaves undefined LARMORX refuses in fix and in reproduce, with the reason: a schedule’s matrix name it cannot interpret, setoption bbrstep below 1, a matrix file with a word that is not a number, -pedir without a field map (FSL-70 to FSL-73).
  • FSL’s background value of a volume 3 voxels wide (a level of FLIRT’s pyramid on a small image, or the padding of such an input) comes from a buffer too small for the border it writes. FSL corrupts its heap there, and then aborts, crashes or finishes. LARMORX takes the value over the whole border by default, and with reproduce the values that fit FSL’s buffer (FSL’s result where it finishes); it names FSL-75 in both. A single-voxel volume, which has no border, is refused. The same buffer is corrected in applyxfm4D and in mcflirt’s replica.
  • FLIRT shares with mcflirt the integer outputs’ clamping (FSL-09), the treatment of infinite voxels (FSL-14) and the order of NaN voxels in a background value (FSL-22).
  • makerot divides its axis by the square root of its length, so a rotation about an axis that is not of unit length turns by the angle times that root: corrected by default (FSL-76). An axis or centre of other than three values, which FSL reads past, is refused (FSL-77).
  • midtrans exits with status 0 after an error of its computation, a singular matrix: exit status 1 by default (FSL-78).
  • applyxfm4D’s spline interpolation of an integer image keeps the spline’s coefficients and values in the image’s integer type, a coarser and shifted spline: computed as for a float image by default (FSL-79). With -3D and sinc it uses a blackman kernel of width 7 instead of the rectangular kernel of width 9 it asks for: the rectangular kernel by default (FSL-80). Both are named only where the written image changes.
  • FSL 6.0.7.23’s aff2rigid stops with a Python error under NumPy 2 before writing its matrix: the matrix is computed by default (FSL-81), FSL’s error with reproduce.
  • The boundary-based cost’s normals are scaled by the reference level’s voxel sizes, not the segmentation’s (FSL-67). Measured on fMRIPrep’s three calls and an anisotropic segmentation, computing them in millimetres moves the results by 0.02 mm or less and brings no gain, so they stay FLIRT’s in both modes.
  • With -anglerep quaternion FLIRT composes the search’s starting points, which hold Euler angles, as quaternions, so most of them lose their rotation: composed as Euler angles by default, FLIRT’s composition with reproduce, named in both (FSL-90).
  • A -refweight or -inweight with no voxel of at least 0.01, and an input or reference of a single voxel, make every cost constant and FLIRT’s matrix an artefact of its search: refused by default, FLIRT’s run with reproduce (FSL-91, FSL-92).
  • A voxel size of 0 in a header makes the base scale 0, and FLIRT never finishes: counted as 1 by default, as reading the image counts it, and refused with reproduce; -basescale 0 is refused in both (FSL-93).
  • applyxfm4D reads an image stored as unsigned bytes as signed ones, so its values 128 to 255 enter the interpolation, the background value and the spline’s coefficients as negative numbers (a profile 100, 140 interpolated 0.4 of the way gives 13 instead of 116), while flirt -applyxfm reads the same image as stored. The clean room reads it as stored by default and as FSL does with reproduce, named where a written value changes (FSL-111, found by its runs of FSL). The replica does the same since 2026-11-03: its default names FSL-111 where the bytes written differ, and its reproduce reads as FSL does, without a notice.
  • pointflirt with two coordinate files that hold no point crashes FSL after its messages. Both ports refuse it in fix and in reproduce after FSL’s lines (FSL-204; the replica since 2026-11-03, the clean room since 2026-11-05).
  • midtrans given transforms whose values reach beyond double precision in its averages (a scale of 10200, a translation of 10300) writes rows of NaN, or reports its failed decomposition as a matrix-size error, with exit status 0. The clean room refuses them in fix and in reproduce (FSL-241, found by its runs of FSL on 2026-11-05).

Known limits

Both ports are experimental. The records’ standard comparison holds fMRIPrep’s calls on three fMRIPrep runs. Their full comparison (fMRIPrep’s 22 calls on every fMRIPrep run with BOLD runs) and the data panel’s other images run in the final validation phase.

Both ports have every command-line program and script of FSL’s FLIRT package, with two exceptions. The coordinate programs’ -warp, FNIRT’s warp fields, is refused with a message until FSL’s warp-field libraries are ported. The scripts flirt_average and epi_reg are not ported yet: they need FSL’s other programs (fslmaths, FAST, the field maps). standard_space_roi, which needs them too, is the replica’s since 2026-11-04 and the clean room’s since 2026-11-05, with its cases in BET’s records (BET’s evidence).

With NaN or infinite input voxels in reproduce, a NaN result can carry the other sign than FSL’s (nan against -nan): an all-NaN matrix from a schedule that optimises relative to a row, and NaN voxels of applyxfm4D’s spline interpolation. The values are NaN in both.

The clean room misses its record’s thresholds in 1 case (above). Beyond them, these are open: the registrations’ paths on random, often ill-posed pairs and the messages whose number follows them, the weighted costs’ last bits, FSL’s sub-threshold costs on a test volume without intensity range, and the sign of a zero skew under gimbal lock (questions for the specification, specs/flirt.md §8.1). Its random testing so far is two runs, of 24 and 31 minutes (above); the long 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, and both ports’ programs have their cases.

History

  • 2026-11-01: the replica, its parts 1 and 2: flirt and convert_xfm with every cost function, the boundary-based cost and the schedules, and FLIRT’s other programs avscale, img2imgcoord, std2imgcoord, img2stdcoord, rmsdiff and makerot.
  • 2026-11-02: the replica’s part 3: midtrans, pointflirt, applyxfm4D in every data type and the scripts pairreg, extracttxt and aff2rigid. The clean room, flirt and convert_xfm; in its part 2 the same day, FLIRT’s other programs, -debug, the verbose levels and the speed. Six of the replica’s cases came with the specification’s answers to the clean room’s questions: FSL-90 to FSL-93 and the order of NaN voxels in FLIRT’s padding (FSL-22).
  • 2026-11-03: three of the replica’s cases came with the answers to the clean room’s part 2’s questions: applyxfm4D on uint8 images, trilinear and with -3D spline (FSL-111), and pointflirt with two coordinate files that hold no point, where FSL crashes and the replica refuses in fix and in reproduce (FSL-204). The replica’s record of 305 cases. The specification’s answers of that day (§14): the windowed sinc and the paths through the C library’s single-precision sine, cosine and arctangent in reproduce, -debug’s lines and volumes, avscale’s residuals, applyxfm4D’s spline in double precision, leastsq’s sums.
  • 2026-11-04: the clean room’s reproduce computes the C library’s functions as glibc 2.35 does: 41 cases give FSL’s files since, among them the default registration (its matrix 0.033 mm from FSL’s before, FSL’s own 0.005 mm) and -applyxfm’s sinc. fMRIPrep’s three calls came from 0.0101, 0.0011 and 0.006 mm to 0.0065, 0.0002 and under 0.00001 mm.
  • 2026-11-05: the clean-room catch-up. The specification’s answers of 2026-11-03 are in the clean room since. -cost leastsq passes, the search orders its starting points of equal cost as FSL’s sort does, pointflirt without points and FSL-241’s transforms are refused. ds005040’s matrix moved from 0.0065 to 0.0113 mm. The random-testing run of 13,907 calls.
  • 2026-11-06: the specification’s answer on the weighted correlation’s count (§6.4).
  • 2026-11-07: weighted normcorr passes; the clean room’s record made again, 339 cases.

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.