Data Cleanup (Apple Numbers)

Diagnose and repair common data-quality problems in a live Numbers table using the Digits MCP tools (numbers_*). Cleanup is destructive by nature, so the rule is: diagnose fully, propose, get sign-off, then fix in one batch.

Workflow

1. Locate the table. numbers_list_documents → numbers_list_sheets → numbers_list_tables. Confirm which table to clean and its dimensions.
2. Read everything. numbers_read_table with include_formulas: true. A formula that returns an error or a blank looks the same as a typed blank until you check the formula grid.

3. Profile and report problems — don't fix yet. Scan for:

   • Duplicate rows (exact, or matching on a key column the user names).
   • Blanks in columns that should be complete.
   • Inconsistent labels — "NYC" / "nyc" / "New York City", trailing spaces, mixed casing — that fragment what should be one category.
   • Text in number columns — "$1,200", "1200 ", "N/A", "tbd" — the usual reason a SUM won't compute. (numbers_read_table returns these as strings rather than numbers, which is your tell.)
   • Type mismatches within a column (some cells numeric, some text).

   Present findings grouped by problem type, with cell references and counts, leading with what's already clean.

4. Propose a fix plan and get explicit sign-off. For each issue, say exactly what you'll do: which rows you'll delete, which labels you'll normalize to which canonical value, which strings you'll convert to numbers. Deleting rows and overwriting cells is not reversible except by Cmd+Z, so the user must approve the plan before you touch anything.
5. Fix in the safest order, batched:
   • Value fixes first. Normalize labels and convert text-numbers to real numbers in a single numbers_set_cells call (Digits has no transaction API, so one batched write is the closest thing to atomic — prefer it over many small writes that could half-apply).
   • Row deletions last, since deleting shifts indices. Delete from the bottom up (numbers_delete_row at the highest index first) so earlier indices stay valid as you go.
6. Verify. Re-read the table. Confirm duplicates are gone, the problem columns are now uniformly numeric, and any totals/SUMs compute. Report before/after counts.
7. Save only with permission. Edits are in-memory and undoable until the user saves. Ask before numbers_save_document.

Rules

• Diagnose and propose before changing anything; never silently "fix" data.
• Deletions go bottom-up so row indices don't shift under you.
• Convert "$1,200" → 1200 as a plain number; formatting is the cell's job.
• Pick one canonical spelling per category and tell the user what you chose.
• Batch value edits into one numbers_set_cells call.
• When in doubt about whether two rows are "the same," ask — don't assume.