docs: record ADR-0016 and archive consolidate-vault-writes

[AlexMost]

Group 3 (PR 3 of 3) of the consolidate-vault-writes opsx change. Documentation only — no src/ or test/ file changes.

PR 1 (#122) folded FsVaultWriter into FsVaultProvider; PR 2 (#123) resized VaultProvider to the six note-file methods and moved the aggregates out. This PR records the decision behind both and brings the docs back in line.

ADR-0016

docs/adr/0016-one-disk-module-owns-note-writes.md, status Accepted. Per design D9 it states:

• one disk module (FsVaultProvider) owns note writes over a vault root, over shared readRaw / writeRaw helpers with one fs-error mapping;
• the VaultWriter seam is deleted, not kept — single consumer, failed the deletion test, and its divergence from the provider's mapping is what shipped edit_note write failure escapes as a bare Error without a code (ADR-0003 violation) #113;
• what the surviving VaultProvider seam is actually for: a stub point for the operations tool tests, not an abstraction over a plausible second backend. docs/architecture/vault-provider.md had been claiming the latter ("a REST API, an Obsidian plugin bridge"); that claim is now gone;
• the explicit non-goal that FsVaultReader's errors-as-data convention is untouched.

ADR-0009 is not rewritten

Its body is unchanged and it stays Accepted. Its INDEX.md row gains "VaultProvider method set refined in part by 0016", matching the 0003 → 0015 precedent for the same kind of narrowing.

Its bullet (a) — "listTags counts frontmatter tags only, not inline #tags" — was flagged as a stale claim. It turns out to already carry its own correction: an italic _Follow-up (2026-08, change inline-tags-in-list-tags)_ note sits directly beneath it, added by that change under this repo's append-don't-rewrite convention. It is a pre-existing item this change did not cause and does not need to fix; left alone deliberately, and recorded in the retrospective.

Architecture docs

File	Change
vault-provider.md	Six-method interface listing; the edit_note-goes-through-VaultWriter paragraph replaced; the two private resolution modes (resolveExisting / resolveNew) documented; resolveIdentifierPath references removed; "what it deliberately does not do" corrected — normalization is now at the tool layer for every write tool
disk-write-path.md	edit_note section added; the existing-note error taxonomy stated against the distinct create-time one, as a table
note-path-resolution.md	The rule paragraph and the "does not do" list no longer name FsVaultWriter
vault-registry.md	writer table row deleted; the reader/provider enumeration fixed
obsidian-lib.md	vault-writer.ts bullet deleted; vault-aggregates.ts bullet added (the same directory gained a file in PR 2, and this doc enumerates it); vault-overview.ts deps refreshed; entry-construction sentence fixed
module-structure.md	FsVaultWriter mermaid node + edge removed; the two prose sentences corrected
vault-reader.md	Not named by the plan. Task 3.6's docs/-wide sweep caught it asserting the reader feeds FsVaultProvider's listTags/listProperties — the reader still scans, but it now feeds free functions

docs/guide/ and README.md came back clean from the sweep, as expected: their list_tags / list_properties mentions are user-facing tool docs and remain correct. No edits were invented there.

Archive

openspec archive synced two ADDED and two MODIFIED requirements into openspec/specs/headless-vault-operations/spec.md. The spec's ## Purpose enumeration gains edit_note by hand — a Purpose block is not expressible as a requirement delta, so that is a deliberate edit on the synced spec, not a sync gap (task 3.7).

verify.md and retrospective.md are included; all 19 tasks are ticked.

Gates

• openspec validate --all — 18/18 valid (19 before archive removed the change item)
• npm test — 110 files, 1371 tests passing (flat vs PR 2 by design; no src//test/ changes)
• npm run lint — clean
• npm run typecheck — clean

Closes #114

🤖 Generated with Claude Code