.gitignore

# Byte-compiled / optimized / DLL files
__pycache__/
*.py[cod]
*$py.class

# C extensions
*.so

# Distribution / packaging
.Python
build/
develop-eggs/
dist/
downloads/
eggs/
.eggs/
lib/
lib64/
parts/
!book/quarto/contents/*/parts/
sdist/
var/
# Labs ship a pinned mlsysim wheel for Pyodide; ignore other wheel drops here.
wheels/*
!wheels/mlsysim-0.1.1-py3-none-any.whl
!wheels/mlsysim-0.1.2-py3-none-any.whl
!wheels/mlsysbook_labs-0.1.0-py3-none-any.whl
share/python-wheels/
*.egg-info/
.installed.cfg
*.egg
MANIFEST

# PyInstaller
*.manifest
*.spec

# Unit test / coverage reports
htmlcov/
.tox/
.nox/
.coverage
.coverage.*
.cache
nosetests.xml
coverage.xml
*.cover
*.py,cover
.hypothesis/
.pytest_cache/
cover/

# Jupyter Notebook
.ipynb_checkpoints

# IPython
profile_default/
ipython_config.py

# pyenv
.python-version

# Virtual environments
.env
.venv
env/
venv/
ENV/
env.bak/
venv.bak/
bin/
!site/newsletter/bin/
include/
lib/
lib64/
pyvenv.cfg
share/
# Prevent duplicate venvs in subdirectories
site/.venv/
**/.venv/
**/venv/
**/env/

# IDEs
.vscode/
.idea/
*.swp
*.swo
*~

# Jupyter Book / Quarto build artifacts
book/_build/
book/quarto/_build/
book/quarto/_output/
book/quarto/_quarto.yml
# The `binder` CLI may create book/quarto/index.qmd as a transient symlink
# to contents/volX/index.qmd. That file must never be committed: Quarto's
# website auto-discovery picks it up and silently overrides the per-volume
# landing in index-volX.qmd, causing the Vol1 build to deploy Vol2 content.
book/quarto/index.qmd
book/quarto/**/*.html
book/quarto/**/._*_xref.json
book/quarto/**/*_files/
book/quarto/**/*.backup.*
book/quarto/*_figures.txt
# Quarto auxiliary figure dirs for the TinyTorch Quarto site — the
# website project renders through the .qmd files once and leaves
# `<stem>_files/` caches next to each source. They're regenerated on
# demand and confuse the sibling PDF book project's cleanup step.
tinytorch/quarto/**/*_files/
docs/_build/

# NBGrader - only ignore generated outputs, not source/release
assignments/autograded/
assignments/feedback/
assignments/submitted/

# Logs
*.log
book/tools/scripts/testing/logs/

# LaTeX build artifacts
*.aux
*.bbl
*.blg
*.out
*.fls
*.fdb_latexmk
*.synctex.gz
*.toc
*.lof
*.lot
*.pdf
# Callout icon PDFs needed for PDF builds (and parallel worktree builds)
!book/quarto/assets/images/icons/callouts/*.pdf
# MLSys·im docs: tutorial slide PDFs are committed; mlsysim-paper.pdf is built by
# CI (mlsysim-publish-live.yml cp's pdf-artifacts/paper.pdf into place at deploy).
!mlsysim/docs/tutorial_module*.pdf
# StaffML paper: figure PDFs shipped from SVG / generate_figures (not the compiled paper)
!interviews/paper/fig-*.pdf
!interviews/paper/figures/fig-*.pdf
# Periodic Table paper + figures: build outputs. Figure PDFs regenerate from SVG via
# the per-paper Makefile (comment in mlsysim/paper/Makefile: "a stale committed PDF
# cannot mask a freshly edited SVG").
periodic-table/paper/paper.pdf

# Database
*.db
*.sqlite

# OS
.DS_Store
Thumbs.db
.envrc

# Temporary files
*.tmp
*.temp

# Binary design sources
*.ai

# Backup files
*.bak
*.backup

# Misc. configurations
.claude
.claude/
.cursor
.cursor/
.cursor~HEAD
.gemini
.gemini/
.ai/

# MLSysBook extension working directories (worktrees, reports, logs)
.mlsysbook/

# Binder layout diagnostics (PDF/table contact sheets and generated QMD)
book/.layout/
book/quarto/.layout/

# Project specific
tito-cli.log
COMMIT_LOG.txt

# Tito CLI user state and cache (local to each user)
.tito/

# Downloaded datasets (not source-controlled, too large)
/data/
datasets/
milestones/datasets/

# BUT: Include TinyTorch's curated educational datasets (tiny, ship-with-repo)
!datasets/tinydigits/
!datasets/tinytalks/
!datasets/tinymnist/
!datasets/README.md
!datasets/DATASET_ANALYSIS.md
!mlsysim/mlsysim/datasets/

# BUT: Include StaffML app source directories
!interviews/staffml/src/data/
!interviews/staffml/src/lib/

# Temporary generated files
progress.json
**/VERIFY_*.py
**/NEW_CITATIONS.bib
**/proposed_figures.tex

# Generated notebooks (built from src/*.py source files)
# The modules/ directory contains generated notebooks for learners
modules/*
!modules/README.md

# Generated website module docs (copied from src/*/ABOUT.md during build)
# Single source of truth: src/*/ABOUT.md
docs/modules/*
docs/chapters/modules/*

# Site build artifacts
site/_build/

# Generated tinytorch package (auto-generated from src/ via nbdev_export)
# Single source of truth: src/*/*.py files
tinytorch/tinytorch/*
!tinytorch/tinytorch/README.md

# Generated benchmark outputs
.tito/benchmarks/

# Misc. generated files
/.luarc.json
book/quarto/index.idx
book/quarto/index.ilg
book/quarto/index.ind
CLAUDE.md

# Newsletter drafts: tracked in git on feature branches, but `draft: true` in
# frontmatter keeps them invisible in the rendered site listing. Drafts should
# live on feature branches and only merge to dev when moved to posts/.

/.quarto/
**/*.quarto_ipynb

# Quarto / mlsysim rendered site output
_build/
mlsysim/docs/assets/
mlsysim/docs/site_libs/
mlsysim/docs/**/*.html

# mlsysim scratch / test plans
mlsysim/test_*.yaml
mlsysim/check_*.py
# Local uv lockfile; CI uses pip from pyproject.toml, not `uv sync --locked`
mlsysim/uv.lock

# Marimo session state (local to each user)
**/__marimo__/session/

# Slides Quarto build cache
slides/.quarto/

# Internal planning docs, review notes, and AI prompts (not public)
labs/plans/
site/newsletter/CONTENT_PLAN.md

# Release-review working tree: tracking docs, manifests, prompts, and
# per-pass output dirs from the multi-pass release audit. Local-only.
/review/

# Auto-regenerated on every `news pull` / publish-live render. Tracking it
# in git would create cherry-pick conflicts every time it drifts (subscriber
# count, issue count) without adding any signal — pull.py rewrites it from
# scratch each run.
site/newsletter/_stats.yml
mlsysim/core/optimization/PLAN.md
book/quarto/contents/vol1/REVIEW_NOTES.md
book/quarto/contents/vol2/context_notes.md
book/tools/review_prompts.md
book/tools/scripts/figure_narrative_audit.md
book/tools/scripts/tbl_lst_narrative_audit.md

# One-off scripts in repo root
/fix_*.py
/apply_*.py
/test_*.py
/find_*.py
/generate_*.py
/check_*.py
/cut_*.py
/sharpen_*.py

# Generation logs (from quiz/question generation scripts)
generation_log*.txt

# PID files
flash_pid.txt
*.pid

# Resized image thumbnails (generated by build scripts)
*.resized

# Generated reports and state snapshots
# Default output from book/tools/audit/scan.py (repo root)
/audit-ledger.json
book/tools/baseline_state.json
book/tools/scripts/reference_check_report.txt

# Timestamped audit run output — regenerated on demand, never committed
book/tools/audit/out/
# Audit-ledger output (regenerated by `binder check` runs)
book/tools/audit/audit-ledger-*.json

# Stray patch.py files at repo root (from one-off bib fix runs)
/patch.py

# TinyTorch user-local config
tinytorch/.tinytorch/

# Next.js
.next/
node_modules/

# LaTeX layout preview rasterizations (regenerate with pdftoppm)
interviews/paper/layout-review/

# Local PDF page exports for layout review (not part of the LaTeX source)
interviews/paper/page-*.png

# StaffML paper: LaTeX/BibTeX outputs (rebuild with make paper or pdflatex + bibtex)
# (Global *.aux / *.log / *.pdf also apply; listed here so intent is obvious.)
interviews/paper/paper.pdf
interviews/paper/paper.aux
interviews/paper/paper.log
interviews/paper/paper.out
interviews/paper/paper.bbl
interviews/paper/paper.blg

# Debug dump from SVG parsing experiments
interviews/paper/svg_structure.txt

# Output dirs from release-prep audit scripts and SVG audit artefacts (ephemeral, not committed)
/.audit/
/_audit/

# Editorial audit reports (internal scratch; not committed to the book repo)
book/quarto/audits/

# Math-rendering audit artifacts (tools/audit/*.py); regenerated on demand
/audit-math-report.json
/audit-math-report.md
/audit-pdf-report.json
/audit-pdf-report.md
/audit-pdf-spot-check.md
/audit-pdf-output/

# StaffML corpus.json — build artifact regenerated from YAML by
# `vault build --local-json`. Kept locally for offline development
# and static-export deploys; production uses the Cloudflare D1 worker.
# Pre-commit / CI workflows generate it on demand; not tracked in repo.
interviews/vault/corpus.json
interviews/staffml/src/data/corpus.json

# SocratiQ bundle — build artefact produced by `socratiq/` Vite build
# (`npm --prefix socratiq run build:vite`), which copies the output into
# `book/quarto/tools/scripts/socratiQ/bundle.js`. The mirror at
# `book/tools/scripts/socratiQ/bundle.js` is a symlink to that file. The
# bundle is ~7 MB and has accumulated ~57 MB across 7 historical versions.
# Going forward, the bundle should be regenerated rather than re-committed.
# (This rule does NOT untrack the existing tracked bundle.js — that cleanup
# is a separate, coordinated change. It only stops *new* drops from being
# added by accident.)
book/quarto/tools/scripts/socratiQ/bundle.js
book/tools/scripts/socratiQ/bundle.js
# Generic bundle.js anywhere under scripts/ai_menu/dist/ (legacy build path
# from issue #1175 — no longer in HEAD but defensively excluded).
scripts/ai_menu/dist/bundle.js
scripts/ai_menu/dist/*.bundle.js

# vault-cli check output — dump from `vault check` runs, regenerated on demand
interviews/vault-cli/check_results.json

# Math/sub-agent audit reports — ephemeral, not committed
/audit/

# Generated dataset CSV — not committed; regenerated locally during chapter work.
# Has slipped into staging via `git add vol1/` multiple times; explicit ignore
# prevents recurrence.
all_ai_models.csv
book/quarto/contents/vol1/nn_computation/all_ai_models.csv
book/quarto/all_ai_models.csv
book/quarto/data/all_ai_models.csv

# Fmt / notation audit scratch outputs (one-off queues; summarize in docs)
book/tools/audit/fmt/*_adjudication_queue.txt
/build_log.txt
/scripts/converge_precision.py
/scripts/debug_precision.py
/scripts/fix_ml_systems.py
/scripts/refine_ml_systems.py
/scripts/remove_allow_zero.py
/scripts/remove_drop_zero_robust.py
/scripts/restore_precision.py
/scripts/surgical_fix_ml_systems.py
/scripts/test_ml_systems_values.py
/scripts/unblock_build.py