The casebook convention¶
Some discussion threads in this repository are casebook threads: reconstructed conversations written for teaching, published by the maintainer, with illustrative roles. This document exists so nobody has to guess which ones.
Why they exist¶
The most useful thing a technical discussion can show is not the right answer. It is the route to it: the plausible wrong turn, the correction that came with a measurement, and the moment someone changed their mind. A repository that has only correct answers teaches half the subject.
A new repository has no such conversations, because it has no participants yet. Rather than wait, the maintainer writes them — openly labelled, so the reader knows exactly what they are reading.
The rules¶
- Every casebook thread is labelled. The first line of the opening post carries the
banner, and the thread carries the
casebooklabel. - Every number is real. A casebook comment may not contain a figure that cannot be reproduced by running a named cell in this repository. Roles are illustrative; results are not. This is the line between a teaching transcript and fiction, and it is not negotiable.
- The wrong answers are ones people actually give. Not strawmen. Most are drawn from the issue tracker — #4 and #7 both began as confident, reasonable, wrong hypotheses.
- Disagreement resolves on evidence, never on authority. The maintainer does not win an argument by being the maintainer. If a thread ends with someone deferring rather than measuring, it is a bad thread.
- Not every thread resolves. Some end unresolved and linked to an open issue, because that is what a real tracker looks like.
The three tiers¶
| Tier | What it is | Where |
|---|---|---|
| A · Casebook | Written by the maintainer, roles illustrative, banner and label present | Q&A, Design Reviews, Interview Prep, Show and Tell |
| B · Live | Genuinely open. Real questions from real people, answered as they arrive | Exercise Clinic, Errata, Office Hours |
| C · Curated | A live thread that turned out well — re-titled, summarised, answer marked, linked from the docs | Anywhere |
Over time the centre of gravity moves from A to C. That is the intent.
The roles¶
The same voices recur, so their perspective is predictable:
| Role | Perspective |
|---|---|
| Priya | Sharp on method. Asks whether the reasoning is sound, not whether the code runs |
| Marcus | Reaches for the plausible textbook answer first. Frequently wrong, always in an instructive way |
| Dana | Owns the eval harness. Arrives with numbers and an interval |
| Rafael | Search infrastructure background. Thinks about what breaks at scale |
| Noor | Regulated industry. Asks about permissions, audit and retention before anything else |
What this is not¶
It is not a substitute for real participation, and it does not pretend to be. There are no additional accounts: every casebook comment is published by the repository owner, which is visible on every comment. If you want to disagree with one, the reply box works normally — that is the point.