Evidence Grounding in the AI Scribe: How Every Billed Line Gets a Source Span

Audit defense lives or dies on whether you can point at the exact sentence that supports a billed line. Evidence grounding is the feature that makes that possible without asking clinicians to hunt through transcripts.

What a source span actually is

A source span is a character-level pointer from a structured output — a CPT code, a measurement, an ICD-10 diagnosis — back to the token range in the raw transcript or image metadata that produced it. When you click the code in the superbill, the underlying quote highlights.

Why it matters more in wound care than in general scribing

Wound visits stack multiple billable events in minutes: measurement, debridement depth, tissue type, offloading education, skin substitute application. A general scribe that produces a fluent paragraph is not enough — you need line-item traceability. That's what the AI scribe for wound care is built to provide.

How grounding is measured

• Span coverage rate — the share of billable lines that carry a valid source span. Target is 100%; anything less is flagged for clinician attention.
• Span faithfulness — a secondary check that the span actually entails the claim, not just contains keywords.
• Negation handling — spans that include no, without, or denies are treated as counter-evidence, preventing spurious codes.

What it looks like in review

Instead of re-reading the whole note, the clinician sees a superbill where each line is a link. Approve, edit, or reject in one click. The signed source notes approach originally built for wound trials is the same substrate powering the everyday superbill.

How it flows into claims

Grounded lines pass into WISER claims and compliance with their evidence attached, so a downstream auditor — internal or CMS — can trace any code back to the visit in seconds.

See it on a real visit with WoundScribe AI.