[e-zz]

Scope (WIP)

Issue: #11 — MCP upsertNodes persists a block but its title does not enter the incremental search index.

• Mark API upserts that reuse the EDN importer and carry that marker into the final batch transaction metadata.
• Preserve imported-data? for genuine imports, including mixed batches, so normal import handling is not reclassified as an API upsert.
• Add focused tests using a disposable DataScript connection, the production Node SQLite search tables/triggers and a real FTS query. Tests cover add, edit of the same UUID (old title removed), genuine import and mixed-batch metadata. Renderer/worker transport and unrelated post-commit effects are stubbed; this is not full desktop E2E.

Verified progress

• Before this change, the targeted test observed a final transaction still marked imported-data? and no FTS match (5 tests, 13 assertions, 3 failures).
• Independently rerun on this branch: node static/tests.js -n frontend.worker.db-listener-mcp-search-test — 6 tests / 28 assertions / 0 failures / 0 errors.
• node static/tests.js -n frontend.worker.db-listener-test — 14 / 51 / 0 / 0.
• Targeted clojure -M:clj-kondo --lint across the four changed files — 0 errors / 0 warnings. git diff --check passed.

Still open — do not merge yet

• No real MCP HTTP request, desktop GUI search box or [[...]] autocomplete verification; no complete graph close/reopen check. The test uses the actual SQLite FTS path but stubs transport and unrelated UI effects.
• Full repository suite has not been accepted as green. The separate worker report records broader-suite failures also present against baseline, not an all-green release gate.
• An untracked, debug-only probe test was excluded from this commit/PR.

This is a draft progress branch, not an issue closure. Keep #11 open until disposable desktop add/edit/search and reopen are observed and the relevant regression gates are accepted.