'master' (or even 'git-annex') might still be the default branch for some datasets

[yarikoptic]

Ran into https://github.com/OpenNeuroDatasets/ds003604/ which still has 5 yo 'master' to be the default branch although there is 'main' now there too!

I didn't inspect other datasets in the org so there might be more

[yarikoptic-gitmate]

Survey of all OpenNeuroDatasets repos (2315 total, probed via git ls-remote on 2026-07-27):

Default Branch Distribution

Default branch	Count	%
main	1247	53.9%
master	615	26.6%
(empty repo, no branches)	451	19.5%
git-annex	2	0.1%

Breakdown of the 615 master-default repos

Category	Count	Action needed
Only master exists (no main)	472	Create main from master, change default
Both exist, identical SHA	74	Just change default to main
Both exist, diverged SHAs	69	Investigate which branch has the real content, then fix

Impact

The openneuro-studies tooling (and likely other consumers) uses get_branch_sha() which tries the repo's default_branch first, then falls back to main, then master. When the default is master and work continued on main, the tooling picks up the stale master SHA — effectively freezing the dataset at whatever commit master points to.

For the 69 diverged repos, master is likely an old snapshot from before the main migration, while main has the real updates. This means tools that respect default_branch (GitHub API, git clone, etc.) serve stale data.

All 69 diverged repos (master ≠ main)

ds000031   ds000228   ds000254   ds001306   ds001371   ds001486
ds001511   ds001553   ds001555   ds001747   ds001796   ds001894
ds002041   ds002241   ds002382   ds002522   ds002608   ds002674
ds002685   ds002721   ds002731   ds002898   ds002938   ds002994
ds003011   ds003029   ds003037   ds003061   ds003082   ds003126
ds003146   ds003176   ds003194   ds003195   ds003381   ds003430
ds003463   ds003498   ds003505   ds003511   ds003516   ds003550
ds003553   ds003554   ds003563   ds003568   ds003592   ds003604
ds003626   ds003633   ds003702   ds003708   ds003720   ds003721
ds003739   ds003751   ds003768   ds003774   ds003777   ds003782
ds003787   ds003812   ds003816   ds003825   ds003844   ds003885
ds003887   ds003949   ds004078

74 identical repos (just need default branch changed)

ds000105   ds000172   ds000201   ds000204   ds000214   ds000220
ds000221   ds000224   ds000243   ds000244   ds000246   ds001399
ds001420   ds001563   ds001751   ds001848   ds001907   ds001972
ds002013   ds002169   ds002236   ds002274   ds002278   ds002345
ds002425   ds002603   ds002606   ds002675   ds002684   ds002718
ds002741   ds002785   ds002790   ds002878   ds002879   ds002893
ds003052   ds003097   ds003145   ds003171   ds003340   ds003357
ds003382   ds003427   ds003437   ds003453   ds003470   ds003484
ds003508   ds003620   ds003673   ds003763   ds003764   ds003766
ds003770   ds003775   ds003778   ds003800   ds003805   ds003810
ds003822   ds003826   ds003830   ds003846   ds003849   ds003850
ds003858   ds003950   ds003959   ds003967   ds003969   ds003972
ds004019   ds004056

Also notable: 2 repos (ds003367, ds007283) have git-annex as their default branch, which means a regular git clone would check out the git-annex tracking branch instead of actual data.

Script used: enumerated ds000001–ds009999 via git ls-remote --symref (no API rate limits).

[yarikoptic]

and this is presumably the script which was used https://www.oneukrainian.com/tmp/analyze_default_branches.py

[yarikoptic]

an interesting finding -- in https://github.com/OpenNeuroDatasets/ds001796/ main is present but behind master branch... uff... I guess likely there are more

[yarikoptic]

I just spotted that our https://registry.datalad.org/overview/ had nemar datasets without basic BIDS metadata extracted... looks like they all (?) having git-annex branch as the default one :-/ like https://github.com/nemarDatasets/on003516 and so on

are those managed via openneuro as well or some other means and issue should be filed there too?? (attn @arnodelorme)