GT: Ground Truth Annotation

Record the human-reviewed compliance outcome for each receipt. These annotations form the reference dataset used to evaluate the accuracy, consistency and reliability of the Compliance Intelligence Engine across all experiments.

About this page

Annotate each receipt with a human verdict, supporting reason and ambiguity flag. Fields pre-populate from E1 extraction where available. Use the navigation below to move through the review queue.

Learn more in Research Design
—
Pending Review
receipts
—
Completed
receipts
—
Skipped
receipts
—
Total Receipts
receipts
—
Last Updated
 

Receipt Review

Live preview
—
Select a receipt to preview
Receipt Metadata
Filename—
Uploaded—
File Type—
File Size—
Resolution—
SourceRepository
StatusLoaded

Annotation Summary

Source Repository
AI Extraction (from E1) Populated
Existing Annotation None
Last Modified —
Reviewer —
Save updates this summary. Use Previous/Next to navigate.

Annotation Workspace

Fields below are pre-populated from E1 extraction. Correct any value before saving.

AI Extraction (from E1)

Populated
Vendor—
Date—
Amount—
Currency—
Category—
Points Pay—
Points Earned—
Gift Card Pay—
Other Virtual Pay—
Guidance
  • Verify each extracted field against the receipt
  • Verdict reason must be a single sentence
  • Use "Suspicious" only when policy is clearly violated
  • Mark as ambiguous if reasonable reviewers could disagree
  • Skipped receipts do not contribute to accuracy metrics
Keyboard Shortcuts
←Previous
→Next
⌘+SSave
EscClear changes