Appendix E — Checkpoints at a Glance, and Answers

Keywords

Quarto, reproducible publishing, coding agents, Markdown, Python, EPUB, data analysis, k-means clustering

Every checkpoint

Each checkpoint is a folder in quarto-workshop/checkpoints/ holding a project as it stands at the end of one hands-on block, with a README listing the steps to catch up: back up your folder (all but .venv/) to a new time-stamped folder beside it, copy the checkpoint in (and, for the water project, data/ if it is missing), render and check a named output. Paste Catch me up: follow ../quarto-workshop/<path>/README.md into your agent, or follow the README by hand, then carry on from the section whose catch-up box you used. The README names the workshop slide at the same point. There is no checkpoint 03.

Reading on your own, diagnose before you catch up. Check where you are (pwd), what changed since your last commit (git status, then git diff) and the error’s first line (Reading an error, Appendix D), and try the one fix it points to. Catch up only if that fails. Afterwards, open the output the README names and see that it shows what the book describes at the catch-up box.

Chapter 5 has no checkpoint either. Its project is small and typed by hand, and its starter, starters/first-report/ in the same repository, is its fresh start: to begin again, rename your copy with mv ~/first-report ~/first-report-old and copy the starter again as in Section 5.1. Within the chapter, your last commit is the known-good state: git restore takes a file back to it.

Table E.1: The London checkpoints, the section whose catch-up box brings you to each, and what it holds.
Path Catch-up box Holds
checkpoints/01-london-markdown Section 6.10 london.md with every Markdown construct, the rules, the environment, data/ and images/
checkpoints/02-london-qmd Section 7.7 london.qmd with front matter, a computed sentence and chart, and six formats
checkpoints/04-london-audiences Section 8.3 the dashboard, the Word brief and its email, and the A5 flyer
examples/london Section 8.4, Section 9.4 the finished London project: website, square advert and white paper
Table E.2: The water checkpoints, the section whose catch-up box brings you to each, and what it holds. They hold sources only: their catch-up copies the data.
Path Catch-up box Holds
checkpoints/05-water-explore Section 18.4 the rules, the environment and explore.qmd
checkpoints/06-water-prepare Section 19.4 scripts/analysis.py with prepare()
checkpoints/07-water-cluster Section 20.6 k-means in the script and clusters.qmd
checkpoints/08-water-interpret Section 21.6 outcome rules, named profiles, and the heatmap and outcomes on the page
checkpoints/09-water-causation Section 22.4 the within-cluster check on the page
checkpoints/10-water-report Section 23.1 report.qmd in HTML, Word and Typst PDF
checkpoints/11-water-slides Section 23.2 slides.qmd in Reveal.js and PowerPoint
checkpoints/12-water-epub Section 23.3 the report’s optional EPUB
checkpoints/13-water-website Section 23.4, Section 23.5 the website, its blog post and listing
examples/water Section 23.7 the finished water project, with the dashboard and white paper
Table E.3: The evaluation checkpoints of the fast path and Part III. They carry their data, and their catch-up runs scripts/evaluate.py before rendering, because outputs/ is computed, not stored.
Path Catch-up box Holds
starters/eval-repo Section 3.2 (its fresh start) the evaluation repository before Quarto: code, tests, data and a notebook
checkpoints/14-eval-start Section 3.6 the rules, the packages Quarto needs, report.qmd and its guard
checkpoints/15-eval-results Section 10.3 model-card.qmd, references.bib and a report with numbered sections
checkpoints/16-eval-slides Section 11.6 the deck, acme.scss and its contrast test, scripts/deck_pdf.py and the PowerPoint template
checkpoints/17-eval-site Section 12.5 _brand.yml and the logo, the Word and PowerPoint templates, the website and scripts/check_site.py
checkpoints/18-eval-whitepaper Section 13.5 whitepaper.qmd, references.bib with a CSL style, scripts/check_refs.py and the house template
checkpoints/19-eval-params Section 14.5 update.qmd, segment.qmd and scripts/render_params.py
checkpoints/20-eval-ci Section 15.6 team rules in AGENTS.md, a pull request template, and the check, publish and monthly workflows
examples/feedback Section 14.5 the educator’s feedback sheets, from a synthetic gradebook
examples/eval Section 15.6 the finished evaluation project

Answer: changing a London input

The task (Section 9.5): set the Tower of London’s visits_2025 to 8,000,000.

What changes. Every page that loads attractions.csv (Table 9.3) is rebuilt with the new count, and what it shows depends on its code. The guide, in every format it writes (web page, Word and PDF, and PowerPoint, slides and EPUB if made), the dashboard’s value boxes and bar chart, and the partner brief (london-email.docx) find the top attraction with idxmax() and the total with .sum(), so the Tower of London becomes the most visited attraction and the total rises. The square advert, if made, uses only idxmax(), so it names the Tower too. The white paper names no top attraction: its sums by admission change, but its table keeps ALVA’s rank_uk column from the file, so it would still rank the Tower ninth in the UK beside the most visits of the ten. A copied column goes stale too.

What does not. The flyer, which copies its facts and quotes no visit counts; the covering email (london-email-message.md), which quotes no numbers and is not rendered; the dashboard’s line chart and the 3-D map, which read visits.csv, not attractions.csv; any output not rebuilt; and the live site, until it is published again.

Why. Every render re-runs every page’s code (freeze: false), and each changed output is linked to the input by a line of code that loads it. A strong answer names that line.

Follow-up questions worth asking. Where would the change appear if the number had been typed into the prose instead of computed? (Nowhere: the typed number would silently disagree with the chart.) What would change if visits.csv were edited instead? (The dashboard’s line chart, the 3-D map and the white paper’s trend.) Why does the render re-run every cell? (Because nothing is cached; see Section 23.4.)

Answer: changing a water setting

The task (Section 23.6): set MAX_MISSING = 1 instead of 2 in scripts/analysis.py. Computed here by running the finished analysis both ways:

Table E.4: The water analysis with each setting.
MAX_MISSING = 2 MAX_MISSING = 1
rows dropped for missing features 2 654
rows kept 11,235 10,583
municipalities 1,090 1,088
k 4 4

The prediction. Every row missing more than one of the nine features goes. Department capitals have no zone-level coverage, so their rural zones reach that rule with both coverage values missing (imputation comes after the rule): every one of their rural zone-years goes, Tunja’s eight among them, while Tunja’s urban zone, which takes the municipal total, stays.

What the rebuild shows. The stricter setting drops 652 more rows, every one of them missing both coverage values: the capitals’ 135 rural zone-years and 517 other zone-years whose coverage split is missing that year, seven of them capitals’ urban zones with no municipal total either. Nothing else goes. Tunja’s urban zone stays in cluster 1 in every year. k stays 4, so the blog post’s assert k == 4 still passes (a setting that changed k would make the post fail its render, by design), the names stay, and 98% of the remaining rows keep their cluster.

What changes. pre-render re-runs scripts/analysis.py, which rewrites every file in outputs/ (clusters.parquet, cluster_profiles.csv, k_selection.csv, stability.csv, outcomes_by_year.csv, irca_infant.csv, irca_infant_bins.csv and summary.json), and freeze: false re-executes every page, so every page that reads outputs/ changes (Table 23.1): the clusters page, the report in every format, the slides, the blog post, the dashboard and the white paper, which also quotes the setting itself.

What does not. data/, the page sources, the post’s title (k is still 4) and, until it is published again, the live site. Nor does the explore page: it re-runs, but reads only data/water_municipal.csv.

Why. Compute once, publish many: one setting in one script, read by every publication. Strong answers name the mechanism (pre-render, outputs/, freeze: false) and check at least two outputs against the new row counts. Undo with git restore scripts/analysis.py.