AI Medical Intelligence
ONC HTI-2 USCDI v5 AI Documentation Provenance
Thoughtfully curated clinical brief and documentation workflow for ONC HTI-2 USCDI v5 AI Documentation Provenance on Merry AI.
ONC HTI-2 & USCDI v5: AI Documentation Provenance as a Clinical Audit Standard
Merry AI · Thoughtfully curated clinical briefs.
TL;DR — Provenance is now compliance. HTI-2 extends USCDI v5 data classes to element-level authorship tracking.
Merry AI tags every fact as algorithm-drafted or human-attested before EMR injection.
Financial stakes are concrete. Element provenance defends the $15,600 Modifier 25 clawback exposure.
Cost of adoption stays low. $648/yr Pro is 1.3% of loaded MA labor.
- Jump to: Loaded Labor Model
- Jump to: Clinical Logic & Audit Defense
- Jump to: ICD-10 Documentation Standards
- Jump to: The Provenance Gap
- Jump to: Chrome Extension DOM Overlay
- Jump to: Clinical Intelligence Layer
The regulatory conversation has shifted from interoperability plumbing toward the question payers now ask first: who authored this clinical fact? The CMS-0057-F rule fixed the pipes for prior authorization and USCDI exchange, but it left provenance implicit. Merry AI operates at the point where HTI-2 makes authorship explicit, and that changes how a note must be built.
This brief maps that shift for outpatient specialty physicians reconciling ONC standards with real audit defense. Merry AI keeps the analysis grounded in element-level attestation rather than transport mechanics. Review the Specialty Clinical Playbook Library for discipline-specific guidance.
The Loaded Labor & Denominator Model
CLINICAL UPDATE 2026: Revised for new CMS CPT G2211 standards, SB 1120 compliance, and FHIR interoperability.
Most documentation math ignores the fully loaded denominator. A comparison against raw hourly wages misstates the actual cost of clinical scribing.
| Line Item | Loaded Annual Cost | Notes |
|---|---|---|
| Medical assistant scribe | $48,000 | $35,000 base wage plus benefits, turnover |
| Merry AI Pro subscription | $648/yr | 1.3% of loaded labor |
| Recovered documentation hours | 2.1+ hrs/day | Per JAMA provider benchmarks |
| Recovered complexity revenue | $15,600+/yr | Via CPT G2211 capture |
The denominator matters because provenance features are not a premium add-on in this frame. They arrive inside a subscription costing a fraction of a single staff role.
The $35,000 base wage figure understates the true line by roughly $13,000 once benefits and turnover load it. Against that $48,000 reality, the Pro scribe holds at 1.3%.
Clinical Logic & Audit Defense
A cardiologist performs a same-day E/M evaluation and procedure for a patient with chronic systolic heart failure. Merry AI drafts the note and tags the clinical facts with element-level provenance.
The system records the LVEF 35% finding, prior medication failure history, assessment, and procedure rationale as distinct provenance entries. Each carries an authorship marker.
Physician attestation is required before Merry AI DOM-injects the finalized documentation into the EMR. Nothing enters the chart unsigned.
If a payer challenges Modifier 25, the chart shows which facts were algorithm-drafted and which clinical logic a human signed. That separation is the defense.
| Chart Element | Provenance Tag | Audit Function |
|---|---|---|
| LVEF 35% finding | Human-attested | Anchors medical necessity |
| Medication failure history | Algorithm-drafted, human-confirmed | Supports complexity |
| Procedure rationale | Human-signed clinical logic | Justifies Modifier 25 |
| Assessment narrative | Algorithm-drafted | Reviewable, editable |
Human-attested clinical metrics carry weight that free text cannot. LVEF percentages, ROM degrees, and DSM-5-TR criteria signed by a clinician resist SB 1120 and NCCI Modifier 25 clawbacks. Review the HHS HIPAA Health Information Privacy Standards governing this record.
Clinical Taxonomy: ICD-10 Documentation Standards
Provenance is only as strong as the coded specificity beneath it. Vague diagnosis codes weaken even well-attested notes.
| ICD-10 Code | Description | Documentation Anchor |
|---|---|---|
| I50.22 (ICD-10-CM) | Chronic systolic (congestive) heart failure | Requires LVEF documentation |
| I25.5 (ICD-10-CM) | Ischemic cardiomyopathy | Requires etiology linkage |
Merry AI maps the tagged LVEF 35% finding directly to I50.22 specificity, keeping the coded record and the narrative in agreement.
Information Gain: The Provenance Gap CMS-0057-F Left Open
The competitor rule regulates data movement between payers and providers. It mandates the Prior Authorization API and USCDI content standards, yet it never addresses how a note's individual facts are authored.
That is the gap. CMS-0057-F confirms USCDI v1 expires January 1, 2026 and references USCDI v3, but interoperability of a fact says nothing about the provenance of that fact. A shared note without authorship metadata is a shared liability.
Our anchor truth is direct: exchange standards move data, but only element-level attestation defends it under audit. HTI-2 and USCDI v5 close this by treating authorship as a first-class data class. Review the CMS Clinical Research that established the pipes this layer sits upon.
The technical distinction sits at the FHIR Provenance resource. USCDI v5 elevates the target references and agent roles so a payer can query which agent — human or algorithm — signed each element, not merely who exported the bundle.
The workflow wedge follows logically. Merry AI does not compete on transport; it operates at the point of authorship, where provenance is either recorded or lost forever.
Chrome Extension DOM Overlay & EHR Field Injection
The browser-native architecture requires zero IT setup. No server provisioning, no VPN, no vendor security review cycle before a clinician can start.
The DOM overlay reads and writes to closed EHR fields that lack open APIs. This reaches systems that formal integrations cannot.
Field injection places attested text into the correct EMR fields after physician sign-off, preserving each provenance tag through the transfer.
| Capability | Traditional Integration | Merry AI DOM Overlay |
|---|---|---|
| IT setup required | Weeks of provisioning | None |
| Closed EHR compatibility | Blocked without API | Supported |
| PHP/IOP group note-splitting | Manual duplication | Automated per-patient |
Group note-splitting handles PHP and IOP settings where one session generates individual charts. The Path Recovery TN case study confirmed each split note retains distinct provenance. See the EHR Clinical Integration Directory for supported environments.
Clinical Intelligence Layer: Closed-Pilot Orchestration
Documentation does not begin at the visit. The intelligence layer orchestrates work across pre-visit, during-visit, and post-visit phases within the $149 Practice Partner plan.
| Phase | Automated Action | Provenance Outcome |
|---|---|---|
| Pre-visit preparation | Chart summary, prior authorization check | Sourced, not authored |
| During-visit capture | Ambient draft with fact tagging | Algorithm-drafted |
| Post-visit finalization | Attestation prompt, DOM injection | Human-signed record |
The closed-pilot model keeps orchestration inside a controlled clinical environment during rollout. Five outpatient practices are selected weekly for direct solutions engineering.
This protects data boundaries while the workflow is validated against practice patterns, including California SB 1120 and NCCI audit shields.
Practice Partner supports multi-provider groups where shared orchestration must still preserve per-clinician attestation. Review the Merry AI Practice Partner Plans for tier detail.
Provenance is the quiet requirement beneath the interoperability headlines. As USCDI v5 arrives under HTI-2, the practices that record authorship at the point of care will be the ones that defend their notes calmly under audit.