lds-syllabus

Imported from ELC/lds-syllabus (AGENTS.md). Install upstream with npx skills add ELC/lds-syllabus. Copyright stays with the author.

AGENTS.md

Guidance for AI agents working on jupyter-book-template — Jupyter Book 2 template for MyST notebooks and GitHub Pages deployment, powered by MyST. All book configuration lives in a single file: book/myst.yml.

Repository layout

src/
  core/                 # Pandera schemas, Settings, Protocols, feature engineering
  simulation/           # Synthetic data generation
  analysis/             # Case bootstrap confidence intervals
  prediction/           # MAPIE split conformal prediction intervals
  evaluation/           # Regression metrics and interval coverage reports
  visualization/        # Altair charts and shared theme constants
tests/                  # Pytest suite (mirrors src/ 1:1, 100% coverage)
book/
  myst.yml              # Project + site configuration (TOC, theme, plugins)
  chapters/             # Book content (.md, .ipynb)
  assets/               # Site branding (favicon, logos)
  plugins/              # Custom MyST plugins
  bibliography.bib      # Citations
pyproject.toml          # Project metadata (jupyter-book-template), deps, Poe tasks
.agent/skills/          # Agent skills (e.g. sync-from-template)
.github/workflows/      # GitHub Pages deploy (main branch)
.binder/Dockerfile      # Binder environment (main branch)

Build output goes to book/_build/ (gitignored). Run book commands from the repo root via Poe; tasks set cwd = "book".

Project source code

Library code lives under src/ as installable top-level packages (core, simulation, analysis, prediction, evaluation, visualization). They are built with uv_build and installed editable on uv sync — chapters import them directly, never via sys.path.

• Public DataFrame types are defined in core.schemas using Pandera DataFrameModel.
• Do not use bare pd.DataFrame in function signatures; use pandera.typing.DataFrame[Schema].
• Tests mirror src/ 1:1 under tests/; run with uv run pytest.
• Lint, format, and typecheck settings live in pyproject.toml; pre-commit local hooks run uv run ruff / pylint / mypy so uv.lock is the single source of tool versions.
• CI runs poe check, then pytest, then poe build-book (uv run poe ci).

SOLID extension points

Forks should extend behaviour through these seams rather than editing call sites:

Pipeline data flow

flowchart LR
  Settings --> generate_dataset
  Settings --> expand_features
  expand_features --> ensemble_pipeline[Pipeline: features + VotingRegressor]
  random_forest_regressor --> VotingRegressor
  svm_regressor --> VotingRegressor
  VotingRegressor --> ensemble_pipeline
  ensemble_pipeline --> compare_models
  generate_dataset --> split
  split --> compare_models
  compare_models --> regression_pipeline
  regression_pipeline --> fit_pipeline
  fit_pipeline --> predict
  predict --> regression_metrics
  regression_pipeline --> confidence_intervals
  regression_pipeline --> fit_conformal
  fit_conformal --> conformal_intervals
  confidence_intervals --> plot_intervals
  conformal_intervals --> plot_intervals
  conformal_intervals --> interval_metrics
  confidence_intervals --> interval_metrics

Initialize the template

Use this checklist when turning the template into a new book project.

GitHub setup

1. Fork the repository (GitHub only allows using templates you own).
2. Use main as the default branch (content, CI, and site deployment).
3. Enable Template repository under Settings → General.
4. Enable GitHub Actions if disabled.
5. Enable GitHub Pages with source GitHub Actions under Settings → Pages.
6. Clone the fork locally.

When creating a new repo from the template, select it from the template dropdown.

Connect to upstream (before customizing)

Repos created via Use this template start with a single commit and no shared Git history with this repository (GitHub discussion #50012). Merge from upstream immediately after cloning, before replacing placeholders or editing book content. At that stage the tree still matches the template snapshot, so the merge is usually conflict-free; waiting until after customization forces large add/add conflicts across pyproject.toml, README.md, workflows, and other infra files.

From the child project root (not this template repo):

git remote add template https://github.com/ELC/jupyter-book-template.git
git fetch template main
git merge template/main --allow-unrelated-histories --no-edit
git push origin main

Later updates use a normal merge (git fetch template main && git merge template/main --no-edit). Full workflow, conflict rules, and troubleshooting: .agent/skills/sync-from-template/SKILL.md.

Agents setting up a new book project must run this upstream merge (or confirm it was already done) before any placeholder or content edits.

Local environment

Requires Python 3.13 (see .python-version).

uv sync --all-groups

Replace placeholder values

Search the repo for REPLACE WITH and update every match:

Also update manually in book/myst.yml:

• project.copyright
• project.authors
• project.github
• site.options.analytics_google (or remove the key to disable Google Analytics)

Update README badges and links (project URL, Binder, Colab, GitHub Pages URL). Replace the template README with project-specific documentation when ready.

Verify the setup

uv run poe build-book    # build static HTML with executed notebooks
uv run poe serve-book    # build, then preview at http://localhost:8000
uv run pytest            # run pytest with 100% coverage
uv run poe ci            # pre-commit checks + tests + build (same as CI)

CI (.github/workflows/ci.yml) runs on pushes to main, sets BASE_URL=/<repo-name> for project Pages, and deploys book/_build/html.

Test a branch before merging a PR

When a change affects Binder, pyproject.toml / uv.lock, or project.jupyter, validate on the feature branch before merge:

1. Ensure the branch is pushed to GitHub (MyBinder only builds remote refs).

2. Temporarily set in book/myst.yml:

   project:
     jupyter:
       binder:
         ref: <branch-name>

3. MyBinder (not CI) builds from .binder/Dockerfile on the pushed ref. Use a commit-SHA launch URL (https://mybinder.org/v2/gh/ELC/jupyter-book-template/<sha>) after each push; branch URLs can show logs from an older cached build. Stale logs show RUN uv sync on Dockerfile line 9—the fixed file uses UV_PROJECT_ENVIRONMENT and rm -rf .venv instead.

4. For manual checks, align README Binder/Colab URLs and book/myst.yml binder.ref with the branch.

5. Run uv run poe build-docker locally (same as the check-docker CI job). The image installs deps into .venv/ (see UV_PROJECT_ENVIRONMENT in .binder/Dockerfile); root .hidden hides .venv and venv in JupyterLab.

6. Before merge, remove project.jupyter.binder.ref (or set ref: main), revert README badge URLs, and do not leave a feature branch pinned in myst.yml.

Agents must not leave project.jupyter.binder.ref set to a non-main branch in changes intended for merge unless the user explicitly asks to keep it.

Add new pages

1. Create the content file

Add a new file under book/chapters/. Supported formats:

• .ipynb — traditional Jupyter notebooks (Binder/Colab friendly).
• .md — MyST markdown notebooks (plain text, git-friendly). Copy the YAML frontmatter block from book/chapters/01_markdown.md if the page contains executable code cells.

Use a numeric prefix for ordering (e.g. 03_my_chapter.md).

2. Register the page in the table of contents

Edit project.toc in book/myst.yml. Paths are relative to book/. File extensions are required.

Flat entry:

project:
  toc:
    - file: chapters/00_how_to_use.ipynb
    - file: chapters/03_my_chapter.md

Nested section:

    - title: My Section
      children:
        - file: chapters/03_part_a.md
        - file: chapters/04_part_b.ipynb

The first TOC entry is the book landing page.

For large books, split the TOC into a separate file and reference it with extends: in myst.yml (see MyST TOC docs).

To auto-generate a starting TOC from filenames:

cd book
uv run jupyter book init --write-toc

Re-order the generated entries as needed.

3. Optional: top navigation bar

Add links under site.nav in book/myst.yml:

site:
  nav:
    - title: My Chapter
      url: /chapters/my-chapter

URLs use slugs derived from filenames (extension omitted, numeric prefix stripped). The existing entry /chapters/markdown maps to chapters/01_markdown.md.

4. Citations and assets

• Add BibTeX entries to book/bibliography.bib and cite with MyST syntax ([@key] or @key). Numbered references are enabled via site.options.numbered_references.
• Put site-wide branding files in book/assets/ (favicon, logos).
• Reference images from chapter content with paths relative to the chapter file, or colocate them in book/chapters/.

5. Build to validate

uv run poe build-book

Notebooks are executed at build time (--execute). Fix any execution errors before committing.

Configure the site

All configuration is in book/myst.yml under two top-level keys:

project: — content and metadata

site: — theme and publishing

Custom plugins

The Ethical Ads plugin lives in book/plugins/:

• ethicalads.mjs — MyST directive (:::{ethicalads}:::)
• ethicalads-widget.mjs — anywidget renderer

When changing the book title, update both project.title in myst.yml and book_title in ethicalads.mjs.

Dependencies

Python packages for notebook execution are declared in pyproject.toml under [project] dependencies. Add libraries there when chapters import new packages, then run uv sync.

Commands reference

Conventions for agents

• Do not edit book/_build/ — it is generated output.
• Keep changes focused: new pages need both a content file and a myst.yml TOC entry.
• Run uv run poe build-book after adding or changing executable notebooks.
• Do not commit secrets (analytics IDs are fine; API keys are not).
• Only create git commits when explicitly asked.
• Prefer MyST/Jupyter Book 2 docs over Jupyter Book 1 patterns (_config.yml, _toc.yml are legacy and not used here).

Further reading

• Jupyter Book — Table of contents
• MyST configuration
• MyST deployment (BASE_URL)
• Template README — human-oriented fork/setup instructions