Obsidian vault triage

Triage a messy or imported Obsidian vault into a domain-organized structure with MOC navigation, without losing a single note or piece of information. Use when reorganizing a vault, migrating imports (Evernote, Joplin, Simplenote, Keep), splitting daily notes into atomic notes, or tagging back-catalog notes.

Hard rules (user requirements, non-negotiable)

1. Read before you classify. Read and understand every note before categorizing it. Never file by filename or origin folder alone.
2. Zero loss. No note, attachment, or piece of information may disappear. Content that looks like junk gets an archive/ home, not deletion.
3. Backup before triage. Full snapshot before any write: git init in the vault root, commit everything. The .gitignore excludes only regenerable caches (.smart-env/, .obsidian/workspace*.json). One commit per batch gives instant rollback and a readable progress log.

Target structure

Home.md              # root MOC: links to all domain MOCs, always current
00 Inbox/            # unprocessed captures awaiting triage
10 <Domain>/…        # numbered domain folders (see references/triage-workflow.md)
90 Archive/          # stale/junk, kept, never deleted
_sources/<origin>/   # import resources (e.g. Joplin _resources) stay here

• Numbered prefixes sort naturally in the sidebar; systems near 00, archive near 90.
• Organization is thematic/domain, never by source app. Origin folders (evernote/, joplin/, simplenote/) are import artifacts: their notes get sorted into domains and the folders removed.
• Daily notes live at vault root while active. A triage pass splits a daily note into domain notes; the daily file is removed only after every fragment is verified present in the new notes.

Workflow (batch per domain)

Work in batches: one domain (or one import folder) per batch, one commit per batch. Never attempt a whole-vault move in one pass. The full procedure, including the note-splitting rules and MOC maintenance, is in references/triage-workflow.md. Load it before the first batch.

Gate scripts (run from scripts/, do not reinvent per session):

• linkcheck.py <vault> - every [[wikilink]] and relative ]( target resolves. Zero broken links before a batch commit; classify each reported break as real or false positive (code spans, original import text) before touching anything.
• titlecheck.py <vault> - no note carries a bare section-word title (Summary, Key Points, Produkty, …). Sections stay glued to their parent note; an orphan section-note is a triage failure. Exit 1 = fix the split.
• coverage_audit.py <vault> <mapping.json> - run BEFORE removing a split source. Mapping JSON: {"<source-relpath>": ["<dest-relpath>", …]}. Every non-trivial source line must exist in the destination corpus and every source maps to exactly one destination. Exit 1 = lost content, stop.

A vault git repo should wire titlecheck.py + the staged-file linkcheck as a pre-commit hook so the gates run without agent discipline. Know what the gates CANNOT catch: wrong-domain classification (Gabinet Weterynaryjny in 10 Dev/) passes all three gates - a structurally perfect vault can still be misfiled. Classification is guarded only by the read-per-note rule; gates guard structure and loss. A user-flagged misfile is a CLASS signal: read the whole source domain and refile every sibling sharing the cause (see the workflow's "flagged misfiles travel in packs").

Keep gate docs honest: when a documented script mode (e.g. coverage_audit reading a moved source from git HEAD) is not actually implemented, either ship it or fix the doc - a gate that exits 1 on its documented happy path gets worked around ad-hoc, and the workaround is where mistakes happen.

Import-era frontmatter can wrap a note's ENTIRE body in a block scalar (frontmatter: >- / tags: |-). titlecheck flags these - the body is invisible to a naive --- split, so such a note reads as empty and gets filed as an "empty stub". Never verdict a flagged note without a raw file read, and when unwrapping, audit word-level (old vs new) before committing. After a move batch, run the post-move verification from the workflow's loss audit (byte-diff each moved note vs HEAD~1:<old-path>) - "I only moved them" is a claim to prove, not assert.

Watch out for a Python gotcha in move loops: dumping bodies by splitting on a regex ^---$ mis-parses notes whose BODY contains a literal --- fence (a separator line inside the text) - the split reads only the last chunk as the body. For verdict-critical reads use the raw file (or a split that requires the frontmatter to be the FIRST block).

Domain naming: use the vault owner's language for folder names; use short English tag namespaces (dev/, ai/, home/) for tags: in frontmatter.

Vault-specific hazards (check before any move)

• Moving with shell/git never rewrites links. Only moving or renaming inside the Obsidian app auto-updates them. External moves need the scripted link sweep and broken-link scan from reference step 4.
• Sync folders: a folder containing .stfolder belongs to Syncthing. Pause sync or exclude it before restructuring, or every peer replicates the churn and conflicts.

After triage

Update Home.md and every affected domain MOC in the same batch commit. Update it in the same commit or the map goes stale and misleads.