ds002685 Failure report

## Dataset

- **Dataset:** https://github.com/OpenNeuroDatasets/ds002685
- **Pipeline:** mriqc 24.0.2 (repronim/containers bids-mriqc)
- **Cluster:** Dartmouth Discovery (4h time limit, 4 CPUs, 16GB)
- **Processing level:** subject (all sessions in one job)
- **Subjects submitted:** 2 (via `--count 2`)
- **Total subjects:** 13

## Results

| Job | Subject | Status | Duration | Cause |
|-----|---------|--------|----------|-------|
| 7825481_1 | sub-01 | **failed** | ~3.5h | AFNI INT64 datatype error |
| 7825481_2 | sub-02 | **timed out** | 4h (limit) | SLURM killed at time limit |

Both jobs failed. Neither produced results.

## Failure 1: sub-01 — AFNI can't handle INT64 NIfTI

mriqc's anatMRIQC pipeline calls AFNI's `3dFWHMx` to compute
smoothness on the conformed T2w. AFNI does not support INT64:

```
bid.e7825481_1:116-124:
++ 3dFWHMx: AFNI version=AFNI_24.2.03 (Aug 20 2024) [64-bit]
++ Authored by: The Bob
*+ WARNING: Using the 'Classic' Gaussian FWHM is not recommended :(
** AFNI can't handle NIFTI datatype=1024 (INT64) in file
.../conform/sub-01_ses-00_acq-spc_T2w_conformed.nii.gz
** FATAL ERROR: Can't open dataset
```

The failing file is `sub-01_ses-00_acq-spc_T2w` — an SPC (3D SPACE)
T2-weighted acquisition. All 13 subjects have this file. This means
**every subject will fail the same way** on this particular scan.

**Questions:**
- Is this an mriqc issue (should it cast the datatype before passing
to AFNI) or an AFNI limitation we can't work around?
- Can mriqc be configured to skip specific modalities/acquisitions?
- Does mriqc continue processing other modalities after one fails,
or does it abort the entire subject?

## Failure 2: sub-02 — SLURM time limit exceeded

sub-02 was still processing (downloading data + running mriqc on
8 sessions) when SLURM killed the job at the 4-hour limit:

```
bid.e7825481_2:82:
slurmstepd: error: *** JOB 7825481 ON q01 CANCELLED AT 2026-03-23T00:36:51 DUE TO TIME LIMIT ***
```

This is because for this dataset 1 job per subject won't work well.

| Subject | Sessions | NIfTI files |
|---------|----------|-------------|
| sub-01 | 18 | 283 |
| sub-02 | 8 | 125 |
| sub-04 | 23 | 349 |
| sub-05 | 23 | 350 |
| sub-06 | 23 | 352 |
| sub-07 | 23 | 353 |
| sub-08 | 12 | 183 |
| sub-09 | 22 | 332 |
| sub-11 | 22 | 334 |
| sub-12 | 22 | 323 |
| sub-13 | 22 | 331 |
| sub-14 | 22 | 330 |
| sub-15 | 23 | 347 |

Every subject has one `acq-spc_T2w.nii.gz` file (in ses-00/anat/).

sub-02 has 8 sessions with 125 NIfTI files. Processing all sessions
sequentially in one 4-hour job is insufficient. Subjects with 23
sessions (349+ files) would need much more time.

**Mitigation:** Use `--processing-level session` instead of
`subject`. Each session becomes its own SLURM job, parallelized
across the cluster. We can also bump the reserved time.