Re-lock every notebook for Colab's Python 3.13 runtime

[bendichter]

Colab switched its default runtime to Python 3.13 on 2026-08-19 (the runtime now reports 3.13.15 (main, Aug 6 2026)). Our install cells were compiled against the 3.12 runtime, and 34 of the 35 pin sets stopped resolving there: ml-dtypes on 3.13 requires numpy>=2.1, and the pinned numpy==2.0.2 has no cp313 wheels, so the first cell failed before any notebook code ran.

What Changes

The Colab constraint file (.github/colab-preinstalled.txt) is refreshed from the current runtime's package list. lock_notebook.py, the Dockerfile base image, and the test and build workflows all move to Python 3.13, and every one of the 44 bootstrap notebooks is re-locked.

Re-locking is driven by requirements.in, and 29 of the 35 image groups predated that contract and had none. A new script, .github/scripts/backfill_requirements.py, derives one for each: it scans the notebooks (and their helper modules) for imports, resolves import names to distribution names inside the group's published container image, and carries over ndx-* packages and pkg @ git+ pins, which an import scan cannot see because they are used through a file's cached schema rather than imported. The generated files are committed, so this is a one-time backfill; future changes go through the normal contract.

Groups that use pynwb are held at pynwb<4, hdmf<5, which is the stack the fleet was last verified on. The re-lock should fix Colab, not double as a pynwb major upgrade; that can be done deliberately per group later. The demos notebook keeps the nwbwidgets-era stack (pynwb<3), and 001550/PaganLab additionally needs pynwb>=3.1 plus its ndx-* extension packages, which I verified locally before generalizing the bounds.

Verification

All 35 pin sets resolve for Python 3.13 on linux in a uv pip install --dry-run with wheels required for every compiled package. 001550/PaganLab ran green end to end locally on 3.13. CI on this PR executes all 44 notebooks on 3.13 runners, which is the real test; I will triage anything that fails here. After merge, the image pipeline rebuilds the whole fleet on the 3.13 base.

The Get-to-know-a-Dandiset tutorial had five outputs with an invalid jetTransient key that made nbformat validation fail; those outputs are removed.

🤖 Generated with Claude Code

https://claude.ai/code/session_011NuStoQykizhQCx2KBrMCN

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[github-actions]

Preview for this PR has been removed (PR closed).

[github-actions]

Executed notebooks

CI executed the notebooks changed in this PR through a Jupyter kernel. Review the outputs here (each link is the notebook with its outputs, as a reader would see it):

• ✅ 000108/chunglab/demo/dashboard.ipynb
• ✅ 000402/MICrONS/demo/000402_microns_demo.ipynb
• ❌ 000409/IBL/bwm_usage_notebook.ipynb
• ✅ 000458/AllenInstitute/reanalysis.ipynb
• ✅ 000458/FlatironInstitute/000_lindi_vs_fsspec_streaming.ipynb
• ✅ 000458/FlatironInstitute/001_summarize_contents.ipynb
• ✅ 000559/dattalab/markowitz_gillis_nature_2023/reproduce_figure1d.ipynb
• ✅ 000559/dattalab/markowitz_gillis_nature_2023/reproduce_figure_S1.ipynb
• ✅ 000582/Sargolini2006/000582_Sargolini2006_demo.ipynb
• ✅ 000718/CaiLab/zaki_2024/000718_demo.ipynb
• ✅ 000727/clandinin/simple_data_access/reading_data.ipynb
• ✅ 000728/AllenInstitute/visual_coding_ophys_tutorial.ipynb
• ✅ 000947/TurnerLab/public_demo/000947_demo.ipynb
• ✅ 000971/lernerlab/seiler_2024/fiber_photometry_example_notebook.ipynb
• ✅ 000971/lernerlab/seiler_2024/optogenetics_example_notebook.ipynb
• ✅ 001038/DombeckLab/001038_demo.ipynb
• ✅ 001075/001075_paper_figure_1d.ipynb
• ✅ 001084/HoweLab/001084_demo.ipynb
• ✅ 001172/HigleyLab/001172_demo.ipynb
• ✅ 001528/HnaskoLab/Lotfi_2025/001528_demo.ipynb
• ✅ 001538/catalyst_neuro/zhai_2025/figure_1E_dspn_somatic_excitability.ipynb
• ✅ 001538/catalyst_neuro/zhai_2025/figure_2GH_oepsc_analysis.ipynb
• ✅ 001538/catalyst_neuro/zhai_2025/figure_5F_acetylcholine_biosensor.ipynb
• ✅ 001538/catalyst_neuro/zhai_2025/how_to_use_this_dataset.ipynb
• ✅ 001550/PaganLab/01_behavior_demo.ipynb
• ✅ 001550/PaganLab/02_optogenetics_demo.ipynb
• ✅ 001636/TurnerLab/motor_cortex/antidromic_detection_tutorial.ipynb
• ❌ 001636/TurnerLab/motor_cortex/turner_m1_glm.ipynb
• ✅ 001636/TurnerLab/motor_cortex/turner_m1_peth.ipynb
• ✅ 001636/TurnerLab/motor_cortex/turner_m1_usage.ipynb
• ✅ 001687/wanglab/figure2-distance-tuning/figure2_distance_tuning.ipynb
• ❌ 001712/IBL-Widefield/public_demo/anatomical_localization_widefield.ipynb
• ❌ 001712/IBL-Widefield/public_demo/processed_widefield.ipynb
• ✅ 001712/IBL-Widefield/public_demo/raw_widefield.ipynb
• ✅ 001754/CatalystNeuro/001754_demo.ipynb
• ✅ dandi/DANDI User Guide, Part II.ipynb
• ✅ dandi/DANDI User Guide, Part I.ipynb
• ✅ demos/NWBWidget-demo.ipynb
• ✅ tutorials/bcm_2024/analysis-demo.ipynb
• ✅ tutorials/cosyne_2023/simple_dandiset_search.ipynb
• ✅ tutorials/neurodatarehack_2024/advanced_asset_search.ipynb
• ✅ tutorials/neurodatarehack_2024/simple_dandiset_search.ipynb
• ✅ tutorials/open_data_quick_start_2026/Get-to-know-a-Dandiset.ipynb

Last updated: commit 7af70c71f1a7a209f7d1a72f362274c1f3d57f56. The executed copies live only in the preview and are removed when the PR closes; nothing is committed to the branch.