[yarikoptic-gitmate]

Summary

• BIDS 1.11.0 allows + in labels (bids-standard/bids-specification#1926). This adds dandischema.utils.sanitize_label() and BIDS_LABEL_REGEX. The function replaces every character other than an ASCII letter, an ASCII digit or + with +, one character at a time, so fly_226 becomes fly+226. It is meant to be the single definition shared with dandi organize (dandi/dandi-cli#1981).
• sanitize_value() replaces only characters that sanitize_label() replaces too, so sanitize_label(sanitize_value(v)) == sanitize_label(v) for any v. A label produced by any dandi-cli version and the identifier it came from therefore map to the same value.
• aggregate_assets_summary() uses that property. A sub- or sample- label from a file name now adds a subject or sample only if its sanitize_label() form differs from that of every identifier in the metadata. Before, the raw label was compared with the sanitize_value()-ed identifier. Names written by older dandi-cli versions, which kept , and spaces, therefore counted each subject twice, and with the new numberOfSubjects <= numberOfFiles assertion that breaks aggregation.
• sanitize_value() itself is unchanged.

Effect on DANDI

I recomputed aggregate_assets_summary() with master and with this branch for all 606,651 assets of the 1,001 public draft Dandisets, using their dandisets/<id>/draft/assets.jsonld manifests (2026-11-05). Exactly 6 Dandisets change, and all other counts, including numberOfSessions and numberOfSamples, are identical:

Dandiset	Files	Subjects on master	This PR	File-name label vs metadata identifier
000053	359	35	34	Ford Prefect vs Ford Prefect (sanitize_value() → Ford-Prefect)
000217	1,121	50	25	Mouse 2054 vs Mouse 2054 (sanitize_value() → Mouse-2054)
000246	2,812	195	186	mouse1-DC1-1L-C2-1R-1 Left vs mouse1_DC1-1L_C2-1R-1 Left
000550	17	AssertionError	17	Fig6-5xBi-0,83MHz-08 vs Fig6_5xBi_0,83MHz_08
000631	15	AssertionError	15	600ns-4kV-0,8MHz-BP-8-24-21-BPAE-20 vs 600ns_4kV_0,8MHz_BP_8-24-21_BPAE_20
000703	22	AssertionError	22	…-nac (2) vs …_nac (2)

The assertion still fails for 001248, 001677 and 001869. Their file names and NWB subject_id genuinely disagree: a placeholder subject_id, R2302 vs R23-02, and fly-226 vs fly_0.

Design notes

• One character at a time, nothing collapsed or stripped. Collapsing runs and stripping a trailing + would merge distinct subjects in public data: …2000um2_-0.4V vs Eref… and …2000um2_0.4V vs Eref… (000983), and 4484 vs 4484? (001195).
• Metadata identifiers are not merged by sanitize_label(). 001702 has distinct subjects 5XFAD+_… and 5XFAD-_… (transgene ±). Sanitizing both sides symmetrically would count 67 subjects instead of its 82. Only file-name labels are matched against metadata in sanitize_label() form; everything else is compared as before.
• Session counting from file names is unchanged.
• Not changed: the numberOfSubjects <= numberOfFiles assertion. dandi-archive would still fail to publish the 3 Dandisets above, so whether that check should become a warning is a separate decision.

Testing

• pytest dandischema: 333 passed, 17 skipped. On master: 320 passed. The new tests cover sanitize_label(), the sanitize_value() round trip over all printable ASCII characters, and aggregation for the 000550, 001702 and 001869 cases.
• codespell, flake8, mypy dandischema and pre-commit (black, isort) are clean.
• The archive-wide comparison is described above.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PvaGRZLHuwDgtWrKMFiMZ8

[codecov]

Codecov Report

❌ Patch coverage is 72.91667% with 13 lines in your changes missing coverage. Please review.
✅ Project coverage is 49.08%. Comparing base (eef36dd) to head (a8c80d1).

Files with missing lines	Patch %	Lines
dandischema/metadata.py	0.00%	10 Missing ⚠️
dandischema/utils.py	0.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #447      +/-   ##
==========================================
+ Coverage   48.56%   49.08%   +0.52%     
==========================================
  Files          19       19              
  Lines        2477     2522      +45     
==========================================
+ Hits         1203     1238      +35     
- Misses       1274     1284      +10     

Flag	Coverage Δ
unittests	49.08% <72.91%> (+0.52%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.