Every field traced back to the source image
Field-level provenance is the trust layer inside Voxket OCR. Every extracted value stores the page it came from and a bounding box for its exact location. In the review screen, click any field and it's highlighted on the source image next to its confidence score, so a reviewer verifies where a value came from in seconds, without hunting through the document.
What this changes for your team
- Review is targeted: attention goes straight to low-confidence fields, with the source one click away.
- Data is defensible: every value has a clean line back to the original form.
- Audits stop being painful: answer 'where did this field come from?' with a boxed source image.
- For master-data teams, it's the difference between 'we think this is right' and 'here's exactly where it came from'.
From setup to running
- 01
Capture at extraction
Each field is stored with a normalized bounding box and a 0–100 confidence score during AI-vision reading.
- 02
Preserve through synthesis
When pages are accumulated into one record, each field keeps its source page, image name, and confidence.
- 03
Surface in review
The results screen color-codes confidence; clicking a field draws the bounding box on the exact region of the source page.
- 04
Inspect deeper
An interactive page viewer renders the full image with clickable bounding-box overlays and zoom.
What's inside
Per-field source linking
Page reference plus bounding box for every value.
Confidence scoring (0–100)
Color-coded for fast triage in review.
Click-to-source
Select a field and see it boxed on the source image.
Interactive viewer
Full-page image with clickable overlays and zoom.
Provenance survives synthesis
Source tracking is preserved when pages merge into one record.
System-default flagging
Fields filled from defaults (no source image) are marked, so reviewers know what wasn't read from the page.
Where it fits
Common questions
What is field-level provenance?
It's a link from each extracted value back to the exact region of the source page it was read from, shown as a bounding box in the review screen.
How do confidence scores work?
Every field gets a 0–100 confidence value at extraction, surfaced with color coding so reviewers can focus on the fields most likely to need a second look.
Can I see exactly where a value came from?
Yes. Click the field and it's highlighted with a bounding box on the source image, alongside its confidence score. An interactive viewer lets you zoom and inspect any field in context.
Does provenance survive when pages are merged into one record?
Yes. Each field keeps its source page, image name, and confidence through the accumulation step.
What if a field wasn't read from the page?
Fields filled from a default (with no source image) are flagged, so reviewers know which values weren't extracted from the document.
Ready to put an AI workforce on your frontline?
Deploy voice, chat, and video agents that work 24/7, act in your systems, and hand off to your team when it counts. Build your first agent in minutes.
White-glove onboarding for teams · Talk to a human, not a form.