TRANSPARENT TECHNICAL VALIDATION

Technical validation benchmark

Synthetic-corpus results, not a customer productivity claim.

The corpus contained 12 independent dataset pairs across six ERP-like schemas. The untouched holdout passed 100% user-field preservation, ERP-field updates, new/closed-job detection, duplicate detection and unmatched detection, with zero silent wrong merges.

Measured processing for five repeat datasets had a median of 0.56 seconds, including two XLSX reads, preflight, deterministic merge and an 11-sheet export. Initial profile setup took 50 seconds in the controlled run. Human selection and review time varies by workbook.

A public manual-work baseline reported roughly two hours for daily production re-entry, but it was not a recruited-user stopwatch study. Real demand and repeat use remain unproven.