diff --git a/.planning/phases/10-schema-migration/10-CONTEXT.md b/.planning/phases/10-schema-migration/10-CONTEXT.md
new file mode 100644
index 0000000..3e23819
--- /dev/null
+++ b/.planning/phases/10-schema-migration/10-CONTEXT.md
@@ -0,0 +1,71 @@
+# Phase 10: Schema & Migration - Context
+
+**Gathered:** 2026-03-09
+**Status:** Ready for planning
+
+<domain>
+## Phase Boundary
+
+The database supports multiple libraries and phantom tracks — existing users upgrade seamlessly. Delivers: `libraries` table, `audio_files.library_id` FK, `playlist_tracks` phantom metadata columns, config migration from TOML to SQLite, and atomic migration guarantees. No UI, no CRUD API, no scan pipeline changes — just schema and migration.
+
+Requirements: DATA-01, DATA-04, LIB-04, LIB-05, LSCAN-05
+
+</domain>
+
+<decisions>
+## Implementation Decisions
+
+### Migration experience
+- Silent auto-migrate on startup — no user interaction, no progress indicator, no confirmation dialog
+- Migration runs automatically when the app detects the schema version is behind
+- On migration failure: show error dialog and refuse to start — no degraded/read-only mode
+- Automatic database backup before migration runs (copy .db file before any schema changes)
+- Schema version tracked via integer (SQLite `user_version` pragma or schema_version table) — app checks on startup, runs pending migrations sequentially
+
+### Default library identity
+- Migrated library name derived from the directory name (e.g., `/home/user/Music` becomes "Music")
+- `music_directory` key removed from TOML config after successful migration — libraries table is the sole source of truth
+- Old config key ignored if still present (no crash on stale config)
+- Fresh installs start with an empty libraries table — no default library auto-created, user adds their first library when they want to scan
+- Libraries table is minimal: name, path, created_at — no scan metadata columns yet (Phase 11 can add those)
+
+### Phantom track schema
+- Rich cached metadata on `playlist_tracks`: title, artist, album, duration, genre, cover art path
+- Eager population: metadata columns filled on every playlist_tracks insert (not lazily on library removal)
+- Phantom tracks identified by NULL `audio_file_id` — no separate `is_phantom` boolean column needed
+- Migration adds new columns via ALTER TABLE ADD COLUMN (not table rebuild) — existing playlist_tracks rows get NULL metadata columns, backfilled from audio_files data
+
+### Migration rollback strategy
+- One-way migration — downgrade to pre-multi-library versions is unsupported
+- Pre-migration backup is the user's safety net for rollback
+- Backup file naming is timestamp-based (e.g., `yellowjacket.db.bak.20260309`) — multiple backups can coexist
+- No automatic backup cleanup — user manages old backup files
+- Migration events (start, success, backup path, errors) logged at INFO level to standard app log
+
+### Claude's Discretion
+- Exact column types and constraints for the libraries table
+- Index strategy for library_id FK on audio_files
+- Whether to use SQLite `user_version` pragma vs a dedicated schema_version table
+- Migration transaction boundaries (single transaction vs per-step)
+- Backfill query strategy for populating phantom metadata on existing playlist_tracks rows
+
+</decisions>
+
+<specifics>
+## Specific Ideas
+
+No specific requirements — open to standard approaches
+
+</specifics>
+
+<deferred>
+## Deferred Ideas
+
+None — discussion stayed within phase scope
+
+</deferred>
+
+---
+
+*Phase: 10-schema-migration*
+*Context gathered: 2026-03-09*