Parsing, cleaning, normalizing: the hands-on workshop from Extract Summit 2026. Everything here runs offline.
- Install uv.
git clone https://github.com/zytedata/messy-bits-workshop && cd messy-bits-workshopuv sync
Run uv sync as soon as you can: it downloads Python and all dependencies,
which takes a while on conference Wi-Fi.
From the directory of the script:
cd demo && uv run 01_parsing.py
cd exercise && uv run starter.py
For the # %% cells in VS Code, open this folder and pick .venv as the
interpreter.
Prefer Jupyter without installing uv/Python? You only need Docker:
cd dublin
docker compose up --build
Open http://localhost:8888 (no token). Stop with Ctrl+C. Edits are saved to your local folder.
Optional, for hook/with_checklist.ipynb:
ANTHROPIC_API_KEY=sk-ant-... docker compose up
On Linux, if your user isn't 1000:1000, run
UID=$(id -u) GID=$(id -g) docker compose up.
hook/: two LLM answers to the same prompt, quoted on the opening and closing slides, plus the prompt that produced them.demo/: the three live-coding scripts and the sample pages they read.exercise/: the hands-on. Start fromstarter.py, seeREADME.mdthere.dublin/: the Dublin version of the workshop: the same demos and exercise as Jupyter notebooks, with the Docker setup and the Scrapy + Jev demo.slides.md: the deck, in Marp format.tests/:uv run pytestchecks that every script still prints what the slides say it does.
This repository is governed. When the portal created it, it set the org
custom property governed=true (plus production, contains_code, and
lifecycle). Org-level rulesets in your GitHub org then apply automatically,
based on those properties — you don't configure branch protection per repo:
| Ruleset | Applies when | Effect on the default branch |
|---|---|---|
baseline |
governed=true |
No force-push, no branch deletion |
require-pr-production |
governed=true AND production=true |
PRs required, ≥1 approval, stale reviews dismissed |
require-secrets-scan |
governed=true AND contains_code=true |
The secrets / detect-secrets check must pass |
.github/workflows/secrets.yml calls the central reusable workflow in
zytedata/governance. The required status check is secrets / detect-secrets.
You cannot weaken this:
- The scan logic lives in the governance repo, not here.
- The check is required by the org ruleset, so a missing/failed check blocks merge.
- New secrets must be explicitly audited into
.secrets.baseline(and reviewed) before a PR can merge.
Run the scan locally before pushing:
pip install detect-secrets pre-commit
pre-commit install
pre-commit run --all-files
# If you add a legitimate, reviewed value that trips the scanner, audit it:
# detect-secrets scan > .secrets.baseline # then review the diff in your PR