FRAME Audit: OwnChart

URL: https://github.com/nickpdawson/OwnChart
Date: 2026-05-28
Analyst: FRAME v1.0 / Synambix
Documents reviewed: README.md, PHILOSOPHY.md (21 sections, 21.9 KB — the most substantive patient agency doctrine in any repo in this batch)

One-Line Verdict

FOUNDATIONAL | Score: 22/24 | The deepest patient agency philosophy in the batch, an evidence contract that operationalizes CLAIM, and a self-hosting commitment with no commercial dilution — held back only by alpha maturity and single-maintainer sustainability.

Dimension Scores

Dimension	Score	Evidence
Patient Agency Index	3/3	”The institution is never the customer. The patient is the user, the owner, the corrector, and the canonical authority on their own story.” User correction is canonical (§5); patient memory is first-class evidence (§6); truly self-hosted, no SaaS path
Architecture Integrity	3/3	Consent gate on egress path; no third-party telemetry (§15) — no Sentry, Datadog, Mixpanel, PostHog; SHA-256 content-addressed immutable sources; every LLM job creates a ModelRun audit record; prompts externalized as versioned YAML
Technical Maturity	2/3	Alpha state; demo at demo.ownchart.me; iOS in TestFlight; Docker Compose deploy path; per-capability status. But multiple docs marked “coming soon,” no CI badge, test count not published.
Safety & Disclaimer Posture	3/3	”Not medical advice” stated; “AI never instructs you to start, stop, or change medication” (§7); self-harm trigger routes to crisis support and human referral, never instructions; consent gate explicit in architecture
Interoperability Stack	3/3	SMART on FHIR; CCDA/XML; HealthKit/Apple Health; DICOM planned; Epic, athena, ModMed, NextGen, Cerner connector docs; FHIR on export surface — patient cannot be locked in
CLAIM Alignment	3/3	§19 is literally CLAIM operationalized: “AI should increase agency not dependency, make hidden structure visible, help the user think question and advocate.” Evidence contract with five labeled classes. Hugo Campos and Critical AI Health Literacy credited explicitly by name.
Sustainability Model	2/3	PolyForm Noncommercial (alpha); single maintainer (nickpdawson); no community channel visible; 17 stars. MIT as possible future license noted. CLA not required (unlike UHR).
Scope Honesty	3/3	”Alpha software. Expect rough edges.” Capabilities clearly labeled. Coming-soon docs explicitly marked. Clear “what OwnChart is not” in README.
TOTAL	22/24

Key Strengths

PHILOSOPHY.md is a category-defining document. 21 sections, 21.9 KB of patient agency doctrine. This is what it looks like when a project actually works out its theory of change before writing code. Section §21 (“Doctrine travels with the fork”) is particularly important: “Strip the consent gate and you have an EHR scraper. Add telemetry and you have a SaaS.” This is the sharpest articulation of the principal/agent failure mode in any repo analyzed.
Evidence contract operationalizes CLAIM. Five labeled classes (Source-backed, User-canonical, Inferred, Statistical, Unknown), mandatory source citation, “why do you think that?” as a first-class UI interaction — this is the most mature epistemic architecture in the batch.
User correction is canonical over institutional record. §5 and §14 together describe a system where the patient’s assertion supersedes the institutional record for display purposes while preserving the original source. This is radical in the context of existing EHR design.
No third-party telemetry, by design. §15 lists every analytics vendor by name and explicitly excludes them. This is not a privacy policy hedge — it is an architectural commitment verified by no external scripts in the deployment model.
Export surface maintains FHIR compliance — the patient can never be trapped. This is rare.

Key Gaps

Alpha maturity with underdeveloped install path. Several key docs (User Guide, Install Guide, Operations Runbook, Configuration Reference) are marked “coming soon.” The product is not yet accessible to non-technical patients or caregivers.
Single-maintainer risk. No org, no team, no community channel. The philosophy is exceptional; the bus factor is 1.
No CI badge or published test coverage. Unlike Tula (0.97 aggregate eval score) or OpenKP (567 tests), OwnChart’s technical health is not publicly measurable.
Household/caregiver support is roadmap, not live. The caregiving use case is explicitly named as important but not implemented.

CLAIM Assessment

Contextual Grounding: Exemplary. Every AI output is grounded in the user’s specific data with mandatory citation. The evidence hierarchy in PHILOSOPHY.md §3 defines precisely how sources are ranked by authority — primary event records beat EHR summaries beat model inference. This is contextual grounding as architectural principle, not feature description.

Interrogative Stance: Exemplary. The evidence contract mandates that every claim supports the question “why do you think that?” with source links, page/section, extracted excerpt, confidence level, and correction history. Confidence labels are human-readable (“Confirmed / High / Medium / Low / Possible / Unknown”) not numeric, because the design explicitly acknowledges that patients need to challenge, not read p-values.

Associative Integration: Strong. FHIR on both import and export surfaces means the patient is never trapped in OwnChart’s data model. Multi-provider AI architecture including local models means the patient can route outputs wherever they choose. Saved conversations are searchable and reusable — the patient’s research accumulates over time as a connected body.

Judgment Layer Activation: Strong. AI produces candidates, never commits. “AI never silently mutates the canonical layer” (§7). Every AI suggestion is accepted, edited, or rejected by the user. User corrections override source records. The patient is explicitly designated as “the canonical authority on their own story.”

Methodological Transfer: Strong, though not fully operationalized. The PHILOSOPHY.md teaches the epistemology (§3, §16 — provenance as a first-class data type, evidence hierarchy, the difference between precision and truth). Whether the product UI actually teaches these concepts to the patient or keeps them as backend architecture is the remaining question.

Patient Agency Verdict

OwnChart is the most philosophically coherent patient agency project in this batch. PHILOSOPHY.md §21 states: “Patient-owned means: the patient. Owns it.” And then proceeds through 20 preceding sections to define exactly what that means in architectural terms — data immutability, user correction as canonical, no telemetry, consent gate on every LLM call, FHIR export, local-model support for zero-egress operation.

The project is held at alpha by the honest acknowledgment that the install path and documentation are not yet accessible. But the direction is clear and the design is right. Where Tula builds a governance-mature health coordination system, OwnChart builds a patient reasoning platform. They are complementary projects solving different parts of the same problem.

Evidence Flags

Stars discrepancy between visits: README fetch showed 5 stars; PHILOSOPHY.md fetch showed 17 stars. This suggests rapid growth during the analysis window. Not a flag for misleading content — noted for calibration.
PolyForm Noncommercial for alpha, MIT noted as possible future. The license may change. Worth monitoring for derivatives that strip the consent gate — §21 explicitly warns about this risk.

Recommendation

CITE + ENGAGE. OwnChart’s evidence contract and PHILOSOPHY.md are citable in CLAIM framework writing as the clearest existing operationalization of critical AI health literacy principles in a working (alpha) patient tool. Nick Dawson acknowledges Hugo Campos and Josh Mandel by name — the intellectual lineage is already mapped. Direct engagement could accelerate CLAIM alignment across the patient AI ecosystem.