TEACHING STANCE
Competence means knowing what to check.
The camp is orientation, not compressed fall statistics. Students should leave able to navigate an R project, make a small evidence trail, supervise a coding agent, and recognize where their current knowledge ends.
- 01Begin with a policy question, not a tour of functions.
- 02Live-code the reasoning, including errors and repairs.
- 03Require a short human prediction before an agent pass.
- 04Ask for evidence of verification, not polished output.
- 05Keep optional extensions ready for experienced students.
The LLM is the language engine; the harness is the workshop of context, tools, permissions, memory, and feedback; the agent is the engine acting inside that workshop; the student is the investigator. Reuse the mapping: prompt = work order, diff = receipt, tests = measuring instruments.
noindex discourages search indexing but does not create privacy. If these notes must be private, protect the instructor/ path through the hosting provider or move it outside the public deployment.
SIX MEETINGS
The same investigation gains one new layer each time.
| Date | Meeting | Core move | Agent role |
|---|---|---|---|
| Thu Aug 13, 1-4 | Lesson + Lab 1 | Inspect and describe evidence. | Read-only explainer |
| Wed Aug 19, 3-5 | Lesson 2 | Transform, join, test, document. | Supervised data engineer |
| Thu Aug 20, 4-5 | Lab 2 | Update and compare versions. | API assistant with stop rules |
| Mon Aug 24, 1-4 | Lesson + Lab 3 | Plot, model, qualify. | Skeptical second reader |
| Tue Aug 25, 3-5 | Lesson 4 | Render and audit a handoff. | Evidence-citing auditor |
| Wed Aug 26, 4-5 | Lab 4 | Exchange, run, repair. | Independent second reviewer |
The increase in agency is deliberate. Do not skip the read-only and planning rungs. Students should experience that better supervision comes from tighter context, permissions, and checks—not simply from a longer prompt.
ROOM PRACTICE
Keep both ends of the experience range moving.
| Moment | Low-floor support | High-ceiling extension |
|---|---|---|
| Inspection | Provide exact commands and ask for plain-language interpretation. | Write reusable assertions and missingness summaries. |
| Wrangling | Work from the frozen wide file. | Rebuild the long file from the API. |
| Plotting | Modify a supplied ggplot. | Compare scales, samples, and robust alternatives. |
| Agent use | Use provided prompt frames. | Ask the agent to propose tests, then improve them. |
| Handoff | Keep one R script with comments. | Render a documented Quarto field notebook. |
Working groups are temporary learning structures. Remind students that there is no group submission and that keeping a personal script protects their own learning trail.
ANTICIPATE
Make the likely mistakes teachable.
- “The code ran, so it is correct.”
- Ask students to state what the output should be, test a key, and inspect a surprising row.
- “Missing means zero.”
- Compare a genuinely observed zero with
NAand discuss how each enters summaries. - “The latest dataset is better.”
- Show that new releases can revise history, change coverage, or alter definitions.
- “The regression proves the policy worked.”
- Return to the claim ladder and ask what assignment mechanism or counterfactual is missing.
- “The agent knows the package.”
- Open the official documentation, test a minimal example, and compare versions.
- “My project runs.”
- Restart R, remove workspace objects, and render from the top.
BEFORE EACH CLASS
Keep the room focused on reasoning.
- □Render and open the lesson deck at its stable URL.
- □Run the demonstration script from a clean R session.
- □Confirm the frozen dataset and dictionary links work offline.
- □Prepare one intentional error and one verification demonstration.
- □Keep printed or static prompt frames available if an AI service fails.
- □State the privacy boundary and no-grading policy aloud.
The complete minute-by-minute facilitation architecture, interaction vocabulary, and lesson-specific teaching plans live in the teaching plan. The research basis and implementation rationale live in the evidence brief.