docs: bring the curriculum up to rhiza v1.7.1 / rhiza-claude v0.13.0

[tschm]

The material was written against rhiza v1.3.3 and rhiza-claude v0.8.2. Four template minors and five plugin minors later, the largest claim in it was false: that core syncs a make layer into every consumer.

What actually changed upstream

Template v1.4.0 (ADR-0011) replaced the synced .rhiza/rhiza.mk and make.d/ — ~1,500 lines copied into every repo — with the pinned rhiza-task CLI, provisioned per invocation by uvx. v1.3.4 did the same for .rhiza/tests/ with the pytest-rhiza plugin. What core still ships is a 71-line shim that pins RHIZA_TASK and forwards to it.

The curriculum now teaches that split — a template distributes configuration; a package manager distributes code — and everything else follows from it.

Lesson by lesson

File	What changed
L5, L7	Bundle model rebuilt from the live manifest: 26 bundles, and tests / marimo / paper are no longer names the template accepts. All five profile expansions were wrong; corrected against the source.
L10	Rewritten. custom-task.mk and custom-env.mk are gone; the extension points are local.mk, [tool.rhiza-task], the new local-setup.sh (v1.7.0) and .rhiza/.env. Every settings key and the six-layer precedence order verified against rhiza-task's Config.
L11	New sections for rhiza-task and pytest-rhiza; skills 8 → 10 (/rhiza:remote, /rhiza:completions); ecosystem diagram redrawn.
L9	The second Renovate custom manager, for the pytest-rhiza pin — including that it silently died when make.d/ retired, which is the failure mode worth teaching.
L8	What crossing v1.4.0 costs, since that update is not routine.
L1–L3, A1	make-layer wording, bundle attributions, and the project-layout tree.
A2	All five downstream configs refetched.
A3	Contribution counts refreshed from the API (tschm 1,910 → 2,338, across two new repos).

Two claims that had become false

• rhiza-claude is not a rhiza-managed repo. It has no .rhiza/template.yml — it adopted the shim Makefile and RHIZA_TASK pin in v0.12.0 without being synced by the template. A2 said the opposite; it now uses this as the instructive partial case.
• rhiza-task keeps its contributor guide at the root, not under .rhiza/. L12's link was dead.

The A2 refresh is worth a look

Three of the five real-world configs moved on in ways that teach something:

• jquantstats dropped an exclusion for a fuzzing workflow the template retired in v1.5.0.
• chebpy dropped book/marimo/notebooks/rhiza.py, an exclusion that had been there since v1.2.5 and never matched anything.
• loman is still at v0.10.3 and names two bundles (marimo, tests) that v1.7.1 would reject — a sharper counter-example than it was a year ago.

Mutation testing, fuzzing, marimushka and rhiza-cli references are dropped throughout; all retired upstream.

Verification

• 42 code-block tests pass (every YAML and TOML snippet parses).
• All 100 external links resolve, bar one VS Code marketplace URL that rejects HEAD — pre-existing and untouched.
• All local links resolve.
• mkdocs build --strict reports the same 7 warnings as before this branch; confirmed by stashing and rebuilding. CI builds non-strict.
• Both Marp decks rebuilt, so the committed HTML matches its source.

🤖 Generated with Claude Code