CI: "Execution Tests [Latest Anaconda]" fail with `jb: command not found` (jupyter-book 2.x pulled unpinned)

[mmcky]

The daily Execution Tests [Latest Anaconda] jobs — execution-linux.yml, execution-osx.yml, execution-win.yml — have been failing on main every scheduled run since at least 2026-06-03. This is not related to the Anaconda 2026.06 bump (see QuantEcon/workspace-lectures#7): the conda environment resolves and installs cleanly on anaconda=2026.06, the Anaconda channel Terms of Service are accepted (setup-miniconda v4), and the build proceeds — it just can't find the jb command.

Symptom

The Build Lectures (+ Execution Checks) step exits 127:

/home/runner/work/_temp/<...>.sh: line 1: jb: command not found
##[error]Process completed with exit code 127.

Root cause

These workflows install jupyter-book with an unpinned pip install jupyter-book, which now resolves to jupyter-book 2.1.6. jupyter-book 2.x is the MyST-based rewrite and no longer ships the jb console script (it uses the myst CLI), so jb build lectures ... fails with command not found. The Build Cache [using jupyter-book] job (cache.yml) is green because it builds the environment from environment.yml — which pins the 1.x jupyter-book that still provides jb — via activate-environment: quantecon, so it never pulls 2.x.

Workflow	jupyter-book source	Version installed	jb present	Status
cache.yml (Build Cache)	environment.yml (pinned)	1.x	yes	✅ green
execution-*.yml (Latest Anaconda)	pip install jupyter-book (unpinned)	2.1.6	no	❌ exit 127

Fix options

• Pin the install — pip install "jupyter-book<2" ... — in all three execution-*.yml files; or
• Install from the repo environment instead of an unpinned pip line (mirror cache.yml: environment-file: environment.yml + activate-environment: quantecon); or
• Migrate these workflows to the JB2 myst build command, aligning with the in-flight JB2 work ([Experimental] Use jupyter-book 2.0 for building the lectures #363).

Notes

• Timeline: red on every scheduled run back to at least 2026-06-03, which matches jupyter-book 2.x becoming the default pip install target.
• All three OS jobs (Linux / OSX / Windows) share the same unpinned pip line, so they fail identically. The Windows numba/llvmlite special-case comment in the workflow is unrelated to this failure.
• Example failing run (Linux, 2026-07-07): https://github.com/QuantEcon/lecture-python-programming/actions/runs/28879918261