Trust Layer
Tamper signals before money moves
File, image and consistency checks flag edits, duplicates and copy-paste jobs — with evidence attached, not a black-box score.
Three signal families
01
File signals
Metadata vs content dates, edit history, embedded software traces.
02
Image signals
Font inconsistencies, resize artifacts, copy-paste geometry.
03
Consistency signals
Duplicates, totals vs rows, names and addresses across the file.
Forensics
Clean scan. Edited scan.
What forensics sees that extraction alone cannot — before payout or approval.
After emphasis: 50%
Font change invisible to reviewers
Duplicate across months
Nobody checks
Font anomaly flagged
Fieldry
Duplicate of INV-038 held
Fieldry
Date mismatch routed
Fieldry
From upload to signal
01
Scan the file
Metadata, fonts and layout geometry checked on ingest.
02
Compare within
Duplicates and cross-field consistency checked across pages.
03
Flag with boxes
Signals arrive with locations and reasons — approve, hold or reject.
Related products
Related use cases
No. Fieldry emits signals with evidence — font anomaly here, duplicate there. The approve/hold/reject decision stays with you.
Edited values (font/geometry breaks), duplicated documents, metadata/content mismatches and internal inconsistencies like totals that don't add up.
No — file and consistency checks run inside the same pipeline. p95 stays under 10s for API.
Inline on fields (status + reason) and as a signals array on the extraction — plus the audit log.
Catch edits before money moves
Tamper signals with evidence.
An edited invoice total trips a font flag. Payment holds while the original is requested.