# Verify a synthetic reconciliation task


Use the [task-package specification](/data/task-packages) to choose the package layout and record evidence for the exact revision.

Create an original synthetic ledger with no real accounts or transactions. The worked example applies the spreadsheet cell-checking pattern using a small tabular fixture. [SpreadsheetBench project](https://spreadsheetbench.github.io/)

## Author and verify

Specify the output cells or named columns, the currency unit and how refunds are represented. State how to round before anyone solves the task. A Tier 1 verifier should compare each required cell and the total. A matching grand total alone can hide errors that cancel out. A judge is useful only for a separate explanatory rubric.

## Calibrate and correct

Pin the input fixture, rounding rule and verifier version. Run cases with positive and negative amounts and with equal totals produced from different rows. Treat a wrong value as a failed check. Record spreadsheet or runner startup errors separately. Preserve the exact input and corrected verifier if a bad row passes.

## Worked example

Use Python 3.12. This example checks the exported cell values using an invented sale and refund. A fabricated allocation with the same total should fail.

```bash
python3.12 - <<'PY'
from decimal import Decimal

amounts = {"sale": "1.25", "refund": "-0.30"}
cells = {key: int(Decimal(value) * 100) for key, value in amounts.items()}
cells["net"] = sum(cells.values())
expected = {"sale": 125, "refund": -30, "net": 95}

def passes(actual):
    return actual == expected

assert passes(cells)
assert not passes({"sale": 130, "refund": -35, "net": 95})
assert not passes({"net": 95})
print("reference pass, same-total and missing-cell controls fail")
PY
```

For a workbook task, pin the calculation engine and export the evaluated cell values before running the check. This local example checks values after calculation.
