Typed schema map
Define the accepted date, identifier, category, quantity and amount shapes before transforming rows.
Sanitized spreadsheet controls
Clean the sheet without hiding what changed.
One sales, product or inventory CSV/XLSX becomes a typed clean dataset, a reviewable exception ledger and a set of totals that reconcile back to the accepted input.
1sanitized input file
10kmaximum input rows
5visible control layers
0live-system access
The delivery
Every removal or correction leaves evidence.
Rejected-row queue
Separate malformed records with stable reason codes instead of silently dropping them.
Duplicate decision log
Use an agreed business key and record which row was retained, merged or flagged for review.
Control totals
Reconcile row counts and agreed numeric totals across input, accepted, rejected and duplicate outputs.
Correction ledger
List each normalized field family, rule version and changed-row count in a reproducible run report.
Owned synthetic proof
Inspect a fee discrepancy build before sharing a file.
The exact deterministic proof reconciles 483 synthetic source rows to 472 accepted and 11 rejected rows, produces 20 stable review flags, verifies an at-least-95-percent spot check and matches Pandas and SQL control totals.
Download the fee audit proof ZIP · Open the machine receipt · Read the five-control note
This is proof of method only, not a customer project, prior-client result, accounting approval or production-accuracy claim.
Included
One bounded cleanup.
- One authorized, sanitized CSV or simple XLSX
- Sales, product, inventory or similar nonregulated business records
- One schema and up to 10,000 rows
- Agreed typing, duplicate and reconciliation rules
- One written correction round
Excluded
No personal-data or live-system promise.
- Names, email addresses, phones, addresses or payment data
- Credentials, CRM login, private URLs or production writes
- Web/contact verification, lead enrichment or scraping
- Forecasting, valuation, accounting approval, tax or financial advice
- Regulated, government or official-institution work
Start with metadata
Describe the shape, not the records.
The public form asks only for file type, row-count band, record family and primary control need. Do not attach a dataset or reveal field names, company names or customer information.