Root cause identified. Summary of the `ECONNREFUSED 127.0.0.1:15432` errors:

**What's happening:** `palacering.env` has `DATABASE_URL` pointing to a local Postgres on port 15432. That Postgres isn't running — `brew services list` shows `postgresql@14` as `none`. No Docker container for it either.

**Impact:** The `/api/usage/log` and `/api/usage/index` endpoints try to connect and fail. The usage monitor stats page (`/api/usage`) returns empty. There's also a silent data loss bug — when the DB is down, the SDK's usage fallback to JSONL only triggers on network errors, not HTTP 500s, so logged entries may be lost.

**Palace Code dashboard** is unaffected — threads, detail panel, session tails, Linear sync all work correctly (confirmed by the self-heal audit above).

**To fix the Postgres errors:** run `brew services start postgresql@14` and verify it starts on port 15432. The database `palacering` with user `palacering` would need to exist. Alternatively, if the usage DB was abandoned in favor of JSONL files, remove `DATABASE_URL` from `palacering.env` — then `db` becomes `null` and both endpoints return gracefully without errors.

Let me know which way you want to go on this.