Confidence & Provenance

Every cell carries detailed provenance metadata that makes every extracted value auditable and explainable. Hover a cell for a quick confidence score glance; click it to expand the full provenance panel showing the resolution type, Job phase, reasoning trace, and source document references. This transparency is essential for building trust in automated extraction — you can always understand exactly how and why the platform produced a specific value.

When reviewing results, hover over any cell to see its confidence score at a glance. Click the cell to expand the full provenance panel, which shows the resolution type, the phase that produced the value, and a human-readable reasoning trace explaining how the value was derived. Use this information to quickly identify which cells need manual review and which can be trusted as-is.

Confidence scores follow predictable patterns by resolution type. Graph matches from Phase 1 typically score 0.7-0.95 because they are derived from verified registry data. Reference table lookups score 0.95 for exact normalization matches, ~0.70 for fuzzy matches, and 0.50 for AI fallback. Planner-derived values from Phase 2 (resolution type agent_derived) generally score 0.5-0.9 depending on the clarity of the source document and the specificity of the extraction instruction.

• 0.90-0.95 — Step 1 lookup or exact registry transfer. Highest reliability; safe to trust without review in most workflows.
• 0.70-0.89 — Strong graph match or fuzzy registry transfer. Generally reliable; spot-check a sample to validate.
• 0.50-0.69 — AI extraction or fuzzy lookup result. Review recommended; the system found a plausible value but certainty is moderate.
• 0.30-0.49 — Low-confidence AI extraction. The source document was ambiguous or the field instruction was vague. Always review manually.
• Below 0.30 — Very low confidence. The value is likely a best guess. Consider updating the schema instruction or adding a reference table to improve future runs.

Use confidence scores to set your review threshold. Cells above 0.8 are generally reliable and can be trusted without manual verification for most use cases. Cells between 0.5 and 0.8 warrant a quick check. Cells below 0.5 should always be reviewed manually. You can use the full CSV export to filter and sort by confidence, making it easy to batch-review low-confidence cells efficiently.

# The full CSV export adds metadata columns for each field:
# field_name, field_name__confidence, field_name__resolution_type,
# field_name__phase, field_name__reasoning

document_name,invoice_number,invoice_number__confidence,invoice_number__resolution_type,...
Invoice_ACME.pdf,INV-2025-0042,0.95,graph_match,...
Contract_Beta.pdf,CTR-2025-001,0.72,agent_derived,...

Provenance metadata enables powerful downstream analytics. By exporting results with full metadata, you can build dashboards that track extraction confidence by field, document type, or time period. For example, if you notice that the "payment_terms" field consistently scores below 0.6 confidence, this is a signal to refine its extraction instruction or add reference table entries. Teams that track provenance metrics across jobs can quantify the ROI of schema improvements — each instruction refinement or reference table update shows up as a measurable confidence increase in subsequent runs.

Confidence and provenance are also what a reviewer reads when a Spec holds a field for review: the value, its source location, and the reasoning behind it.