# Storage Writes

See how parsed chunks are written to SQLite and how deleted files are soft-deleted.

## Transactional upsert

`databaseHandler.Upsert` is the sole content-write entry point used by the filesystem layer. In one SQLite transaction it:

1. Marks all currently active rows for the source as `dismiss = TRUE`.
2. Inserts each current chunk or updates the `(source, chunk)` conflict, including the chunk `total`.
3. Restores current chunks to `dismiss = FALSE`.
4. Preserves the embedding and `is_embed` only when chunk content is unchanged.
5. Clears changed embeddings and resets `is_embed = FALSE`.

This sequence handles shortened files without returning stale trailing chunks and avoids re-embedding unchanged chunks. A parse that yields zero chunks, such as a sheet with only a header row, leaves every row for that source dismissed.

The `file_data` table enforces `UNIQUE (source, chunk)` and keeps a partial index on rows where `is_embed = FALSE AND dismiss = FALSE`, which is the embedding scheduler's selection predicate.

## Removed files

After scanning current entries, the walker compares the previous snapshot. Missing file nodes are recursively passed to `databaseHandler.Dismiss`, which soft-deletes active rows by setting `dismiss = TRUE`. Data remains in SQLite but is excluded by indexing and query reads.

A removed directory is walked through its recorded children, so every file it contained is dismissed individually.