Spreadsheet cleanup workflow for a synthetic order ledger

A static before-and-after cleanup proof with deterministic rules, exception queues, validation checks, and an operator-ready handoff.

The job

Customer problem

An operations team spends hours correcting inconsistent dates, duplicate orders, missing owners, mixed status names, and malformed currency values before weekly reporting.

Synthetic inputs

Synthetic order ledger

Twenty fictional rows containing duplicate order IDs, three date formats, blank owners, inconsistent statuses, negative totals, and text in numeric fields.

Normalization rules

Use one date format, one status vocabulary, valid non-negative currency, unique order IDs, and explicit exception reasons.

Reporting contract

Produce clean rows, quarantined exceptions, rule counts, and a reconciliation total without silently discarding data.

What the AI did

Profiled every synthetic column and recorded the detected shape violations.
Applied deterministic normalization rules without overwriting the source representation.
Merged only exact duplicates and quarantined ambiguous conflicts.
Reconciled row counts and valid currency totals before and after cleanup.
Produced an exception queue with one reason and recommended owner action per row.

Findings and deliverables

Status aliases can be normalized safely when the mapping is explicit.
Conflicting duplicate order IDs must be quarantined rather than merged automatically.
Missing owners are a workflow exception, not a value to invent.
Reconciliation totals are the fastest way to catch silent row loss.

Acceptance checks

PASS

No source row disappears without an accounted state

Every row is represented as clean, exact duplicate, or quarantined exception.

PASS

Ambiguous values are never fabricated

Missing owners, conflicting duplicates, and malformed totals remain visible exceptions.

PASS

Before-and-after counts reconcile

The clean set, duplicate set, and exception set sum to the original twenty synthetic rows.

What remains unproven

The synthetic workbook does not reproduce a real customer's formulas, macros, locale, volume, or downstream imports.
A production cleanup would require an approved sanitized sample and explicit output schema.

Next access request

None for the synthetic proof. After approval, request one sanitized sample with no customer or regulated data and a written output schema; no live spreadsheet access is required.