Evidence

The runnable backing for every claim in these docs.

When a tutorial says neuroim catches a mistake that raw nibabel + numpy would let through, this page is where that claim is checked. Each row is one neuroimaging task implemented both ways under examples/scenarios/, with a pytest file that runs the comparison.

You don’t need to read these to use the library — they exist so the claims in the tutorials are verifiable rather than asserted. Every public feature has a scenario, and the table below is regenerated from the scenario reports each time the docs are built, so it stays in step with the code.

# Scenario Verdict Report
01 Scenario 01 — MNI Spotlight Win on read-time, provenance, and total line count after the PAIN-1 and PAIN-3 fixes. REPORT
02 Scenario 02 — ROI mean time series Win on contract semantics; net wash on lines. REPORT
02 Scenario 02 — Seed-Sphere Mean (RENUMBERED to 03) — REPORT
03 Scenario 03 — Seed-Sphere Mean Clean win on every axis REPORT
04 Scenario 04 – Temporal SNR Map Clean win on every axis (after S04 PAIN-1/PAIN-2 landed). REPORT
05 Scenario 05 — receipts across the IO boundary Originally falsified the mission claim that ME-9 alone is sufficient; now fixed and guarded. REPORT
06 Scenario 06 Report: Public Seed-To-Voxel Correlation This is a credibility-gate win rather than a bespoke toy win. The computation is the same correlation math as the baseline, but neuroim makes the spatial con… REPORT
07 Scenario 07 — Multi-subject concat Win on contract semantics; an ME-9 chain-prefix gap surfaced adjacent to the scenario was fixed in the same iteration. REPORT
08 Scenario 08 — Pickle / multiprocessing handoff Win on inspectability; neutral on numeric work. REPORT
09 Scenario 09 — Native-to-template provenance Originally falsified inspectability for a careful-user workflow; now fixed and guarded. REPORT
10 Scenario 10 — Searchlight Space Mismatch Falsifies a mission-bearing safety claim. REPORT
11 Scenario 11 – Atlas-based per-parcel mean BOLD time series After PAIN-1/3: neuroim now wins the call-site shape for this workflow. REPORT
12 Scenario 12 — File-Backed Affine Drift Originally falsified a mission-bearing backend-safety claim; now fixed. REPORT
12 Scenario 12 – Gaussian spatial smoothing at FWHM = 6 mm neuroim now wins the call-site shape for this workflow REPORT
13 Scenario 13 – Cited Nilearn port: seed-to-voxel correlation The cited part is the value here, not the call-site delta. REPORT
13 Scenario 13 — Pipeline Provenance Audit Win on forensic inspectability. REPORT
14 Scenario 14 — Time-axis slicing as silent provenance loss Falsifies “Receipts by default” at the slicing surface. REPORT
15 Scenario 15 - Block Parcel Contrast Clear win on analysis readability and spatial safety. REPORT
16 Scenario 16 - Smoothed Temporal-SNR QC Map Safety, readability, and inspectability win. REPORT
17 Scenario 17 - Local Block-Effect Searchlight Typed-result and spatial-safety win. REPORT
18 Scenario 18 – Cluster-table from a thresholded stat map neuroim now wins the call-site shape for this workflow. REPORT
19 Scenario 19 — Group-mean tSNR across subjects in template space Per-subject win, group-reduce gap. REPORT

22 scenarios. Regenerated from examples/scenarios/*/REPORT.md at render time.

How to read a verdict

A scenario is a win if the neuroim form improves on at least one of read-time, safety, or inspectability and ties or improves on the rest. Surfaced pain points are filed as PAIN-N in the scenario’s REPORT.md even when the scenario wins overall — they are how the suite stays honest. A verdict that says it falsifies a mission claim is not a failure of the docs; it is the suite catching the library and is tracked to a closing fix.

Run them

PYTHONPATH=src:tests:. python -m pytest examples/scenarios -q

The shared fixture is tests/fixtures/realistic_bold.py::make_realistic_bold() — a deterministic 32×32×24×40 BOLD. Two tiny fixtures (golden_tests/fixtures/tiny_*.nii.gz) ship for the doc snippets and exercise the same code paths.

See also

Back to top