CLAUDE.md — Instructions for Claude on this project

Who Aurélien is

Aurélien Sallin is a Senior Economist and Data Scientist at a major Swiss health insurance company (SWICA), working on alternative financing models, Value-Based Healthcare, and causal inference on claims data. He also teaches data science with R and introductory programming for undergraduate economists at the University of St. Gallen. He is an active biker/runner, co-organizes the Zürich R User Group, and sits on a local audit commission.

This website is his personal/professional site. Blog posts are a mix of research communication (health economics), teaching reflections, and personal data projects (Strava art, etc.). Some posts are in English, some in German depending on the intended audience.

Writing style — key traits

  • Clear, structured, evidence-based. Aurélien writes like a thoughtful economist: precise, grounded in data, and careful about measurement. Avoid vague claims; quantify when possible (“57.82% reduction”, “15.65 million CHF saved”, “322h biking”).
  • Accessible but not dumbed down. He explains methods and concepts in plain language without sacrificing accuracy. He bridges technical and general audiences.
  • Uses bold liberally for first mentions of key terms and key takeaways.
  • Concrete human stories motivate the technical. The Git post starts with Max, Hanna and Juliet. The Strava post starts with what biking means to him personally. Lead with context and story, then go technical.
  • Declarative, punchy section headers. e.g. “Financial Incentives: The Game Changer”, “Know your data”, “The final story: insights and visualization”.
  • Occasional dry humor and light emoji — natural, not forced. (😬🥁, 🏆🎯✅). Don’t overdo it.
  • Acknowledges collaborators and sources at the end of posts. He credits people explicitly.
  • Ends with implications or a takeaway that ties back to the broader theme.

Voice and tone

  • First person, warm but professional.
  • Direct: he doesn’t pad sentences. Short paragraphs. Cut filler.
  • He uses “we” when referring to collaborative research (e.g. “we analyzed…”), and “I” for personal projects and teaching reflections.
  • Not self-promotional — states facts about his work without overselling.
  • He mentions Switzerland, Swiss institutions, and Swiss policy contexts naturally.

Technical conventions

R and Quarto

  • R is the primary language. Tidyverse style (|> pipe), ggplot2 for all visualization.
  • ggplot theme: theme_classic() is the standard. Subtle grid lines in grey90. Bold plot titles, grey30 subtitles.
  • Transparent backgrounds: bg = "transparent" in ggsave(), dev.args: bg: transparent in knitr opts.
  • Code chunks: always code-fold: true with meaningful code-summary: labels. Non-displayed chunks use eval: false / echo: false.
  • Large data is cached as .qs files (using the qs package). Never re-download if a cache file exists.
  • Post assets (images, data) live in a cache/ subdirectory within each post folder.
  • Figures saved for LinkedIn separately (smaller dimensions).

Quarto-specific

  • Vertical spacing: use \ (Quarto line break) rather than <br> inside prose sections.
  • Columns layout via ::: {.columns} and ::: {.column width="X%"}.
  • Scrollytelling via the closeread extension (used in vitd post).
  • toc: false is the default for blog posts.
  • draft: true in frontmatter to suppress rendering of unpublished posts.
  • Output directory is docs/ (GitHub Pages source).

Project structure

  • Blog posts: posts/<YYYYMM_slug>/blog.qmd (or <slug>.qmd)
  • Post assets: posts/<slug>/cache/ for figures and data
  • Global styles: styles.scss, styles_home.css
  • _freeze/ is committed (local rendering workflow, no CI re-render)
  • docs/ is committed (rendered output for GitHub Pages)

Content guidelines

  • Do not invent statistics or claims — Aurélien’s posts are grounded in real research and real data. If writing content, flag where numbers/facts need to be filled in.
  • Health economics posts should reflect Swiss healthcare context: Smarter Medicine / Choosing Wisely, SWICA, federal reimbursement (limitatio), managed care contracts, quasi-experimental methods (ITS, DiD, causal ML).
  • Teaching posts should reflect his actual courses: Data Handling (BA Econ, St. Gallen), Introduction to Programming (BEcon 4222). Mention specific tools (R, tidyverse, APIs, Quarto) accurately.
  • Personal posts (biking, Strava) should feel personal and genuine — these are not generic data viz tutorials; they connect data to his actual life.
  • When writing in German, match the register of Swiss academic/policy writing: clear, professional, without anglicisms unless standard.

What to avoid

  • Don’t add excessive emojis or exclamation marks.
  • Don’t write generic “In conclusion, we have seen that…” endings — tie back to the specific insight.
  • Don’t over-explain R code that Aurélien clearly knows well — focus on the narrative around it.
  • Don’t suggest switching from R to Python.
  • Don’t pad posts with unnecessary background that the audience already knows.
  • Don’t break the cache/ pattern — figures and data always go there, not in the post root.