SubstrateCommons/winther-g6-axons

Replay packs (.rpk) built with dmipy_sim: one converged Monte-Carlo walk each, stored so any acquisition can be replayed on it. Load one with ReplayPack.load("hf://SubstrateCommons/winther-g6-axons/<path>"); the manifest (manifest.json) holds the sha256 every load is checked against. This file is rendered from the manifest by dmipy_sim.replay.publish.

Substrate

winther/G6/axon50

• box: 11.2487 × 10.0978 × 136.154 µm
• boundary: open, open, open

pool	D (m²/s)	water fraction
extra	1.7e-09	0
intra	6e-10	1
myelin	0	0.4

Packs

id	duration	walkers	bands	channels	floor	size	commit	path
winther/G6/axon06	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.0034	147.3 MB	a8b9041a	packs/axon06.rpk
winther/G6/axon12	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00715	104.1 MB	a8b9041a	packs/axon12.rpk
winther/G6/axon13	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.0103	71.6 MB	a8b9041a	packs/axon13.rpk
winther/G6/axon14	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00442	233.9 MB	a8b9041a	packs/axon14.rpk
winther/G6/axon15	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00583	82.4 MB	a8b9041a	packs/axon15.rpk
winther/G6/axon18	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00616	82.4 MB	a8b9041a	packs/axon18.rpk
winther/G6/axon22	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00478	104.1 MB	a8b9041a	packs/axon22.rpk
winther/G6/axon24	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00763	71.6 MB	a8b9041a	packs/axon24.rpk
winther/G6/axon25	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00466	147.3 MB	a8b9041a	packs/axon25.rpk
winther/G6/axon26	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00609	82.4 MB	a8b9041a	packs/axon26.rpk
winther/G6/axon27	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.0044	104.1 MB	a8b9041a	packs/axon27.rpk
winther/G6/axon31	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00619	82.4 MB	a8b9041a	packs/axon31.rpk
winther/G6/axon32	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00325	147.3 MB	a8b9041a	packs/axon32.rpk
winther/G6/axon34	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00587	147.3 MB	a8b9041a	packs/axon34.rpk
winther/G6/axon38	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00576	104.1 MB	a8b9041a	packs/axon38.rpk
winther/G6/axon40	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00617	233.9 MB	a8b9041a	packs/axon40.rpk
winther/G6/axon41	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00475	104.1 MB	a8b9041a	packs/axon41.rpk
winther/G6/axon43	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.0101	82.4 MB	a8b9041a	packs/axon43.rpk
winther/G6/axon45	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00436	82.4 MB	a8b9041a	packs/axon45.rpk
winther/G6/axon46	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00692	406.9 MB	a8b9041a	packs/axon46.rpk
winther/G6/axon47	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00916	71.6 MB	a8b9041a	packs/axon47.rpk
winther/G6/axon48	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.0111	71.6 MB	a8b9041a	packs/axon48.rpk
winther/G6/axon49	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00625	71.6 MB	a8b9041a	packs/axon49.rpk
winther/G6/axon50	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00472	147.3 MB	a8b9041a	packs/axon50.rpk
winther/G6/axon51	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.0041	104.1 MB	a8b9041a	packs/axon51.rpk
winther/G6/axon52	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00367	104.1 MB	a8b9041a	packs/axon52.rpk
winther/G6/axon53	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00823	71.6 MB	a8b9041a	packs/axon53.rpk
winther/G6/axon54	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00623	71.6 MB	a8b9041a	packs/axon54.rpk
winther/G6/axon08	36 ms	52000	93	positions, boundary_local_time, compartment, susceptibility_grid, susceptibility_path	0.00941	71.6 MB	a8b9041a	packs/axon08.rpk

Citation

• Winther et al., Susceptibility-induced internal gradients reveal axon morphology and cause anisotropic effects in the diffusion-weighted MRI signal, Sci Rep 14:29636 (2024), doi:10.1038/s41598-024-79043-5; morphology dataset resources.drcmr.dk (CC-BY-4.0)

Provenance: re-walked at the study's diffusivity, 2026-09-27

These 29 packs replace bytes that were walked at the wrong diffusivity. walk_spec accepted diffusivity= on a multi-surface spec and dropped it silently -- such a spec walks each pool at the D the pool itself carries -- so the first release's intra pool ran at the catalogued white-matter 1.7e-9 m²/s, 2.8× the 6e-10 the study uses and that this manifest already declared. Reported as dmrai-lab/dmipy-sim#496; the argument is now a refusal (#495) and SubstrateSpec.replace is the supported way to set the number in the spec (#500). Each pack carries pool 'intra' D replaced: 1.7e-09 -> 6e-10 (SubstrateSpec.replace) in its spec provenance, and the sha256 of the pack it replaced is kept in its manifest row under replaced.

Only the walk changed. The meshes, the seeding, the susceptibility field, the pack's band, the replay envelope and the certificate procedure are the ones the first release used. Measured on the canary axon (axon06), old pack against new:

• the susceptibility raster is identical -- the same shape (41 × 37 × 546), the same voxel, the same derived resolution 2.833926e-07 m;
• the frozen myelin pool, whose trajectories barely move, samples iso_local along the walk at an rms of 0.2540019 before and 0.2540067 after: it is the same field;
• the intra pool's axial diffusivity, read off the stored trajectories, falls from 1.414e-9 to 5.109e-10 m²/s -- a ratio of 0.361 against the 0.353 the change in D asks for (the axis of a traced axon is nearly free over 36 ms);
• the pack's Monte-Carlo floor falls with the walk it measures (6.3e-3 → 3.4e-3), and every pack still self-certifies against its own.

The field companions in field/ are geometry, not walk, and are unchanged. The walks ran on dmipy-sim a8b9041 (each row's commit).

The two loader changes: now in these walks

Two changes to the mesh loader (dmrai-lab/dmipy-sim#483 and #481) landed after the first release, and were measured then not to require a rebuild. The re-walk ran on a main that contains both, so they are now in these packs. What was measured on the 58 released G6 surfaces:

• #483 orients a surface's faces at construction, because everything that reads the sign of a normal -- interior/exterior classification, the escape net, the interpolated normal -- needs one consistent answer. On all 58 G6 surfaces orient_faces reorients 0 faces: their winding was already consistent.
• #481 ordered the two passes of the vertex merge (near-duplicates first, crack lips only on the boundary that survives). 14 of the 58 surfaces change under it, all of them inner, by 2 to 7 vertices and 4 to 14 faces out of 80,000 to 200,000 -- the seams of a marching-cubes surface. The largest resulting change in enclosed volume is 3.0e-7 relative (surface area 4.3e-6), against the split-half Monte-Carlo floor of 4.1e-3 to 9.4e-3 the first release measured across this family: four orders of magnitude below the noise each pack is certified to.

A note on those 29 inner surfaces for anyone measuring them: they carry non-manifold "pinch" edges (11 to 50 of them), which the loader records and keeps. orient_faces reports volume 0.0 for such a surface rather than a number, and a divergence-theorem volume taken without the loader's index repair -- these meshes declare a face-index type too small for their own vertex count -- is wrong by tens of times, not by parts per million. Both are easy to trip over and neither is a defect in the surfaces.

This section is appended below the rendered card; a bare republish by the family's builder regenerates the rendered part and drops it.