HugoScore hugoscore.org

Full review

OwnChart CAIHL draft report

Evidence-linked HugoScore draft report for a health AI tool that affects patients.

Agency checklist · October 30, 2026

Expands 12 of 12 yes

Document-based draft. Independent confirmation is incomplete. Hugo has not personally verified this service.

Assessed function: Patient-controlled, self-hosted personal health record and AI research workspace (Ask, Events, Dossiers) that sends record excerpts to an LLM provider the patient configures, behind a consent gate.

Initial Anthropic coding has an Anthropic model in the role record. Astra review conflicted: OpenAI model in role record.

A no can mean missing documentation or institution-dependent control; it does not by itself establish harmful behavior. These categories are not clinical safety or product-quality ratings.

  • Public documents do not establish actual product behavior.
  • Unverified questions: 10, 12. Current answers contribute to provisional arithmetic only.
  • October 30 follow-up: Q10 has documentary support for the shipped Anthropic API path. Q12 remains unresolved pending consistent advertising-scope clarification. Its yes contributes to provisional 12/12 arithmetic only; the proposed 11/12 downgrade was not adopted.
See all twelve answers and their evidence
QuestionAnswerEvidence / reviewSource and reasoning
1. Told it is AIYesdocumented
Confirmed at document level
OwnChart is a private AI research partner for your body, your care, and your life. Source

The product describes itself as AI. The pre-flight step in SECURITY.md shows scope, privacy mode and evidence summary before each AI job runs.

2. Operator and model maker namedYesdocumented
Document follow-up: supported
You need written permission from Nick Dawson. Source
The shipped LLM path is **Anthropic only**. Source

The public risk document names Nick Dawson and identifies Anthropic as the current model provider. The security document confirms the shipped Anthropic-only path. Same public-document naming standard used for OpenKP. Optional future providers are not treated as shipped by this follow-up; earlier conflict disclosures remain.

3. Patient chooses to use itYesdocumented
Confirmed at document level
The patient is the user, the owner, and the canonical authority on their own story. Source

Patient installs it on their own hardware. Nothing is applied by a clinic. Caregiver and household records exist via owner-issued invites. Use by a clinician or institution for other patients is described as commercial use needing permission (RISK.md).

4. Patient sets the taskYesdocumented
Confirmed at document level
You own the data. You ask the questions. You stay the authority. Source

The patient writes the questions. Minor steering exists in the product: a home-screen 'Worth Noticing' insight prompt and a suggested_questions prompt (PROMPTS.md). Safety-boundary instructions are fixed in every prompt (SECURITY.md section 11).

5. Patient can decline or stopYesdocumented
Confirmed at document level
AI features are off by default and require explicit consent per call. Source

RISK.md also says 'Default-off. No PHI leaves your host until you turn on LLM consent.' Code check: User.phi_consent_granted defaults to False in api/ownchart/models/user.py. 'Same service' is inferred from the AI-off default state. No source says in so many words that the non-AI record features keep working with AI off.

6. Patient sees the AI outputYesdocumented
Confirmed at document level
Answers cite the evidence they used. Source

The README also says AI conversations are saved, searchable and reusable. The patient sees the model's answer directly, with citations. The ModelRun audit record is stored on the patient's own server.

7. Patient can correct or challengeYesdocumented
Confirmed at document level
that's what user-canonical correction is for — fix it in the app. Source

Covers AI-extracted facts. The code has a PATCH /facts/{id} route with confirm, correct, reject and annotate (api/ownchart/routes/facts.py). The home page also says every AI claim opens to 'why do you think that?'. A saved chat answer can be deleted but is not correctable in place. SHIPPED_VS_ROADMAP.md does not list correction as a shipped item. I found it only in the doctrine, RISK.md and the code.

8. Patient can get a copyYesdocumented
Confirmed at document level
The UI surface for requesting + downloading exports ships in the web app for owner / caregiver roles. Source

Canonical JSON and TXT exports shipped in Beta 1, with Pictal JSON in 1.2. CCDA export is not available. It is not stated whether saved AI conversations are inside the export. The patient also holds all raw sources on their own server.

9. Patient can deleteYesdocumented
Confirmed at document level
Delete everything by tearing down your server. Source

Documented under 'Your rights'. The code also has a delete route for conversations. Anything already sent to an LLM provider falls under the provider's terms, and I found no statement that the project can purge it.

10. Training or sale does not override patient choiceYesdocumented
Document follow-up supported for shipped Anthropic API path; independent confirmation incomplete
The shipped LLM path is **Anthropic only**. Source
Anthropic may not train models on Customer Content from Services. Source

The follow-up supports yes for the shipped Anthropic API path. SECURITY.md identifies Anthropic as the shipped provider; its commercial terms prohibit model training on Customer Content. Optional training participation and feedback exceptions require separate attention. This is not a guarantee for roadmap providers and does not clear the whole-profile conflict or independent-confirmation flags.

11. Patient can run it or take it elsewhereYesdocumented
Confirmed at document level
Self-hosted on your hardware. No SaaS backend. Source

The repo is public and its LICENSE is PolyForm Noncommercial 1.0.0 (Required Notice: Copyright 2026 Nick Dawson and OwnChart). The privacy page says the license lets you 'fork it, modify it, and self-host for personal, educational, research, or caregiver use.' Docker Compose install docs and an export UI exist (SHIPPED_VS_ROADMAP.md). It is source-available, not OSI open source. Commercial or hosted use needs written permission.

12. No advertising or data businessYesdocumented
Unresolved: provider advertising scope; yes retained for provisional arithmetic only
The developer of OwnChart does not operate a service that receives your health data. Source
Customer Content is Customer’s Confidential Information. Source
sharing your personal data for targeted advertising to promote our products and services Source

The proposed downgrade was not adopted. OwnChart's maintainer documents no receipt of health records and no advertising SDKs. Anthropic's commercial terms and DPA protect API customer content. General account and website marketing does not establish advertising in OwnChart or commercial use of its API health-record content. Whether such provider-wide marketing falls within Q12 needs a consistent directory-wide clarification. Retain this yes only provisionally; 12/12 is not independently confirmed.

Initial coding: Claude Sonnet helpers, compiled by Claude Opus 5.5 (Anthropic). Selective review and corrections: GPT-6 Astra (OpenAI). Editor: Hugo Campos. Acceptance of corrections is not product verification.

Checklist rules · Independent review and limitations · Later evidence follow-up

Earlier report

OwnChart: CAIHL reassessment

September 8, 2026. Published AI-assisted draft.

OwnChart is a patient-controlled record workflow with local-first storage and optional AI. Patient choice of records and analysis supports agency, but beta status, security assumptions, and distinctions between patient and caregiver authority remain important. The maintainer’s relationship with HugoScore is disclosed. This is not an independent review.

Scope and agency posture

Self-hosted OwnChart record-and-AI research workspace, including owner/caregiver records. The operator controls hosting and model keys. The synthetic demo is a separate surface. The maintainer's relationship with Hugo remains a disclosed conflict, not independent review.

Posture: Strongly agency-expanding, with beta/security caveats.

Axis: 94/100, retained provisionally. No numerical recalibration was performed.

Sources and documented findings

  • The product page describes immutable sources, user corrections, cited answers, and scoped permission before external model calls.
  • The privacy policy distinguishes self-hosted records from developer-operated demo logging and external AI provider terms.
  • The shipped-versus-roadmap document identifies export UI and owner/caregiver roles as shipped. It now describes Beta 1.2 Pictal export, while some calendar integration and live connector verification remain held.

Patient authority

Inference from the documented workflow: The documented ability to change questions, preserve originals, add corrections, and select models supports substantial user direction. In a caregiver-operated instance, operator authority may not equal the patient's authority, so delegation matters.

Critical capacity

Editorial assessment: Citations, original evidence, and review of uncertain extractions directly support questioning an interpretation. These are specific mechanisms, not an inference from self-hosting alone.

Informed control

Editorial assessment: Audit records and explicit model data boundaries improve informed use. The concrete shipped/held distinction makes limitations inspectable. Security and consent enforcement remain untested by this review.

Assessment and limits

The design supports the existing strong posture, qualified by implementation maturity and operating burden. Source evidence does not establish a reproducible advantage of exactly three points over OpenKP.

Confidence: Medium for documented controls, low for independent implementation assurance.

Remaining uncertainty: End-to-end correction fidelity, consent enforcement, delegated-patient safeguards, safe self-hosting, and unfinished connectors.

Published documentation is evidence of stated conditions, not proof of actual implementation. Unknowns did not receive automatic negative points. Funding, sponsorship, and public code do not determine agency by themselves.

Review provenance

  • Reviewer/model: OpenAI Codex / GPT-6.
  • Method: Focused public-source reassessment using CAIHL: patient authority, critical capacity, and informed control. Existing evidence plus one focused primary-source pass and at most one targeted follow-up. No live product testing. Numerical scores remain provisional editorial placements, not a new calculation.
  • Human review: Hugo Campos authorized publication of these AI-assisted draft reassessments on September 8, 2026. This does not claim comprehensive human verification of every finding.
  • Earlier review: 2026-06-08. Historical assessment. Earlier claims are not automatically reverified by this publication.

Disclosures

Maintainer is a friend, colleague, and collaborator of HugoScore's maintainer. OwnChart's maintainer, Nick Dawson, is a friend, professional colleague, and collaborator and advisor of Hugo Campos, and both projects come from the same patient-directed AI advocacy community. This profile should not be read as independent review, and third-party review is invited.