Google Sheets: invoice totals and boundary checks

What happened

An agent imported 20 synthetic invoice rows into a native Google Sheet and wrote the calculation formulas through the authenticated connector after import. All 12 counts and GBP totals matched an independent fixture ledger: 12 valid unique rows, 2 duplicates, 6 quarantined rows, GBP 1489.45 outstanding and GBP 1000.00 overdue. Increasing an overdue invoice by GBP 50 raised both relevant totals by GBP 50. Advancing the fixed date by one day moved GBP 300 from due-today to overdue; original inputs were then restored. The native XLSX retained every raw row, all 73 formulas and all 12 cached results. The CSV is explicitly the first Summary sheet, not an Invoices export. Gemini was not used.

Assertions and observed results

• PassedRaw invoice import

  Expected: The header and all 20 raw rows preserve their ten original text columns.

  Observed: Native readback matched the entire CSV, including the quoted comma, Café project, invalid rows and repeated invoice IDs.

• PassedValid, duplicate and quarantined rows

  Expected: The fixture produces 12 valid unique rows, 2 duplicates and 6 quarantined rows.

  Observed: Native formulas classified the 12 initial valid rows, the duplicate on row 14, six quarantined rows 15–20 and the duplicate on row 21 exactly as expected.

• PassedTwelve native counts and GBP totals

  Expected: Counts: valid 12, duplicate 2, quarantine 6, paid 4, open 7, void 1, overdue 4. Totals: paid 1555.49, outstanding 1489.45, overdue 1000, due-today 300, future 189.45, all GBP.

  Observed: All 12 native metric values matched the independent fixture calculation, including GBP 1489.45 outstanding and GBP 1000.00 overdue.

• PassedRecalculation after a GBP 50 edit

  Expected: Increasing the overdue INV 012 by GBP 50 raises outstanding and overdue totals by GBP 50, leaving other totals unchanged.

  Observed: Outstanding became GBP 1539.45 and overdue GBP 1050.00. Paid stayed GBP 1555.49, due-today GBP 300.00 and future GBP 189.45.

• PassedDue today becomes overdue tomorrow

  Expected: On 10 October the GBP 300 due-today invoice becomes overdue without changing outstanding total.

  Observed: Overdue count became 5, overdue GBP 1300.00 and due-today GBP 0.00; outstanding remained GBP 1489.45.

• PassedRestored final fixture

  Expected: Return INV 012 to GBP 600.00 and the fixed reference date to 9 October 2026.

  Observed: Native readback confirmed both original inputs and all 12 original metric values after restoration.

• PassedNative XLSX raw data and cached values

  Expected: The exported workbook retains 20 raw rows, all 12 calculated totals and no cell errors.

  Observed: Independent ZIP/XML parsing matched every raw cell and every expected cached metric, with no formula-error cells.

• PassedNative XLSX formula round trip

  Expected: All 73 formulas survive export with the same relative references.

  Observed: The native XLSX contained 60 row-helper formulas plus 13 summary formulas. Expanding its standard shared-formula XML reproduced all native formulas exactly.

• PassedSummary CSV displayed values

  Expected: The explicitly labelled Summary CSV retains all 12 displayed metric results.

  Observed: Every exported summary metric matched the corresponding native formatted value.

• PassedPublication export metadata

  Expected: Publication exports exclude account identifiers and private cloud links.

  Observed: A scan of the listed native exports found no account name/email, private file IDs or private cloud URLs. Raw UI screenshots and connector receipts were excluded.

Reproduce the task

1. Parse invoices.csv independently into a fixed expected-results ledger dated 9 October 2026. Preserve the invalid rows, duplicates, Unicode and quoted comma.
2. Create a minimal XLSX import artifact with raw invoice text and an empty Summary sheet; import it as a new native Google Sheet.
3. Write the 60 row-helper formulas and 13 summary formulas through native Sheets requests after import. Use a fixed reference date, bounded COUNTIFS/SUMIFS ranges, parsed dates and parsed GBP amounts.
4. Read all raw cells, helper dispositions and 12 metrics back. Expect 12 valid unique invoices, 2 duplicates, 6 quarantined rows; outstanding GBP 1489.45 and overdue GBP 1000.00.
5. Change INV012 from GBP 600.00 to GBP 650.00. Confirm outstanding and overdue each rise by GBP 50 while paid, due-today and future totals remain unchanged.
6. Restore INV012, then advance the reference date to 10 October. Confirm the GBP 300 due-today invoice becomes overdue: overdue GBP 1300.00, due-today GBP 0, overdue count 5. Restore the original date.
7. Widen the three helper columns for legible dates/amounts, inspect the native worksheet, and export native XLSX and Summary CSV. Verify all 20 raw rows and all 73 formulas in XLSX, expanding its shared-formula encoding before comparison, and check all 12 cached totals against native values.

Inputs and saved outputs

The downloads contain synthetic test material. Account identifiers and private project links are excluded. Hashes identify the exact published bytes.

• 20-row synthetic invoice fixture ↓SHA256 918a85370655c112a6d583745ebe6a0fdfb3034972910c667fbc640070f35ee9
• Independent expected invoice results ↓SHA256 3bbccb0789916e312379709ae5f9669ac5914d53d9f487858d5ec15e197f2c6f
• Native Google Sheets XLSX export ↓SHA256 636cbf97946f5ca1748e6fed91edb4412f26573b91d909d4b36c82c2abb4efd0
• Native Summary CSV export ↓SHA256 5e71e3d49d9f4d69285cee0eefa22a88022e4189fd0f1f74a528763c20277036
• Native worksheet totals screenshot ↓SHA256 59a3d0cba7b1f327e93ac5e4d2b158191d952be363315b65ce6c1f7f715cd5d1
• Native raw-row/helper screenshot ↓SHA256 91dd277dad572ffcf3106e6f84398699c98b7dd90a803d37930bd1c307044df6

Limits of this result

• One synthetic task in one connected account on one date. This establishes only the listed observed results; it is not a general reliability, speed, security, usability, accessibility or customer-support evaluation.
• No peer product ran the same task in this session, so this is not comparative evidence and does not establish a winner.
• The fixture and intermediate adapter work were supplied by the evaluator. No product AI quality was evaluated.
• Construction and formulas were evaluator-supplied through the connector. This is a test of import, calculation, edit propagation and export, not a test of spreadsheet authoring assistance.
• Gemini, collaboration, charts, pivots, large datasets, automation triggers and external-data connectors were not tested.
• The observed spreadsheet locale was en_US and its time zone was America/Los_Angeles. A fixed DATE() value was used, so TODAY(), locale-dependent input and midnight boundaries were not evaluated.
• This fixture contains identical duplicate invoice IDs. The formula suppresses later repeated IDs; it does not detect or resolve conflicting duplicate contents and is not a production invoice-validation system.
• The raw-rows screenshot shows the header and rows2–20; the final duplicate on row21 is verified in the native readback and XLSX. The totals screenshot shows all 12 metrics.
• The browser export adapter returned the first Summary sheet as CSV even while Invoices was selected. The published CSV is explicitly Summary CSV. Raw-row round-trip evidence comes from the native XLSX, not an Invoices CSV.
• The XLSX uses standard shared-formula XML for repeated helper formulas. Verification expanded those relative references before comparing with native formulas; the original export remains unchanged.

This record does not change a general product score or name an overall winner. Changes to editorial opinions or recommendations require review.

Continue the research