docs: seven corrections, including one that misdirects contributors

[EricAndrechek]

Found by reading every doc in the repository against the code and the CI config, rather than grepping it. Touches no workflow file, so it is independent of #22.

1. CONTRIBUTING.md actively misdirects — highest severity here

"Markdown formatting and structure … run in CI's prose job but do not block merges yet, for the same reason: this tree's markdown was written hard-wrapped and most of it hasn't been through a dprint fmt pass."

Both halves are false now. prose is a required status check on main — confirmed against the branch-protection API, which lists twelve — and ci.yml's own comment says the reformatting already happened and "Both checks pass, so both gate." A contributor reading this today believes a markdownlint failure cannot stop their PR.

It also never mentioned the public job, so the most surprising gate in the repository was undocumented for the people most likely to trip it. Added, with the fix: say what a thing is, not where it lives.

2. docs/reference/artifact.md — the spec contradicts the code it cites

Step 4 said four symbols are mandatory, while naming chtypes.NewRegistry in go/chtypes/multiversion.go as "the reference algorithm." That code requires seven:

// go/chtypes/multiversion.go:141-144
if (!out->version || !out->init || !out->rows || !out->compile
    || !out->freep || !out->validate || !out->schema_free) {

Python (_native.py:156), TypeScript (ffi.ts:135) and Rust (ffi.rs:175) require four and degrade the rest to unsupported. So this is not only a doc error — it is a real parity divergence nobody had catalogued: an artifact missing chs_free, chs_validate_type or chs_schema_free loads in three bindings and is flatly refused by the fourth.

The doc now describes both behaviours and points at #13. Picking a winner is that issue's job, and the finding is filed there. No artifact this repository builds is affected — every one exports all seven — so the blast radius is third-party and historical artifacts only.

The same file numbered two different items 6, because one 6. was buried inline at the end of item 5. Renumbered; the list is eleven steps.

3. docs/reference/bindings.md — a symbol that exists nowhere, and a miscount

• It cited chs_schema_compile_with_settings as the symbol whose absence makes the probe answer false. That identifier exists nowhere — not in any binding, not in include/chtypes.h — and every binding's own doc-comment says the probe is always true post-consolidation.
• It announced "Two independent detectors" and then numbered three.
• It presented 't'/'f'/'e'/'d' as the universal verdict spelling. TypeScript renders 'true'/'false'/'error'/'decline'. The four states and the fail-closed rule genuinely are identical; the spelling is not, and two SDKs cannot share a log format across it. Now disclosed where verdicts are described, pointing at SDK parity: 12 divergences between the four bindings that need a decision, not a fix #13.

4–7. The rest

file	was	now
reference/python.md	listed Schema, Filter, Block as the context managers	Registry is one too (registry.py:1160); Library deliberately is not, with the reason
install.md	advertised ffi-rs's prebuilt matrix (darwin x64, musl…)	notes chtypes artifacts ship for three platforms only — an Intel Mac or musl user installs cleanly, then chtypes fetch has nothing for them
index.md	Reference table never linked reference/artifact.md	linked — it is what a reader debugging a fetch or verify problem wants
limitations.md	gated filters on an internal rig "gating green"	gated on a release saying so, which a public reader can actually check

Verification

All four parity suites pass — they read bindings.md, and the Python one asserts the manifest and that document agree on the object-model table, so an edit there is not free. lint-public, lint-prose, markdownlint and dprint check all clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_018ckJDLjWhAAttQVrYnQEJD