OpenKP and OwnChart evidence follow-up, October 30, 2026
Prepared by GPT-6 Astra (OpenAI). Public-document review only. No product interaction, login, purchase or health information was used. Hugo permitted GitHub access after the original review. The HugoScore repository remains private; the website remains public. This follow-up preserves the original review and the ten accepted corrections as dated records.
Result
OpenKP changes from 9/12, Leans expands, to 10/12, Expands, the highest category. Q2 changes from no to yes. Q7 and Q12 remain no. A top category does not require a perfect twelve answers: ten yes answers and Q11 yes qualify. This change applies the written naming rule consistently, not a preference for the project's author.
OwnChart remains provisionally 12/12, Expands. Its Q2 naming evidence is now checked. Q10 and Q12 retain their unresolved review flags; this targeted follow-up is not a complete independent confirmation of OwnChart. Your Healthcare Agent remains 10/12, Expands, with Q7 and Q12 no.
The role record for OpenKP's reviewed configuration is patient-operated local software, supplied by Hugo Campos, with Claude from Anthropic. No OpenAI role was identified in that configuration. OwnChart documents a shipped Anthropic path and future other-provider paths; the existing OpenAI conflict disclosure remains visible. Hugo's OpenKP authorship and the original Anthropic-coder relationship remain disclosed. Hugo has not personally verified any service.
Deciding evidence and disputed answers
Every source below was accessed October 30, 2026. GitHub access means the public file was fetched through the GitHub connector, without cloning or running a product. Public web text means the page was read through the web tool.
| Question | OpenKP | OwnChart comparison | Public source and exact passage | Access mode |
|---|---|---|---|---|
| Q2: operator and model maker named | No becomes yes. Public documentation is sufficient; homepage placement is not a rule. | Yes retained under the same standard. | OpenKP DESIGN.md: “Author: Hugo Campos” and “A compromised Anthropic API (Claude Desktop handles this, not us)”. OwnChart RISK.md: “You need written permission from Nick Dawson.” Its SECURITY.md: “The shipped LLM path is Anthropic only.” | GitHub public files |
| Q7: correct or challenge output | No retained as not documented in the three inspected project documents. Preview approval is documented; output correction was not found. | Yes remains supported by an explicit in-app route. Neither behavior was tested. | OpenKP installation README: “returns a preview without touching Kaiser”. OwnChart RISK.md: “that's what user-canonical correction is for” and “fix it in the app”. | GitHub public files |
| Q12: no advertising or data business | No retained for the assessed Claude consumer configuration under the adopted broad advertising rule. This is not a finding that OpenKP sells records. | API content has different terms. Its broader Q12 remains unverified; content protection alone does not settle account-data or marketing scope. | Anthropic privacy policy: “sharing your personal data for targeted advertising to promote our products and services”. Commercial terms: “Customer Content is Customer’s Confidential Information.” | Public web text |
For Q7, the inspected OpenKP documents were the root README, DESIGN.md and openkp/README.md. This bounded search supports the checklist's not-documented answer, not a claim that conversational correction is technically impossible. The host-terms normalization explicitly covers Q8 to Q12; it does not independently resolve Q7.
The original Q2 note applied a stricter standard because of the editor's connection. That was not justified by the question's wording. Conflict disclosure and consistent evidence standards must coexist.
A separate documentation concern remains: the OpenKP root README says “Everything stays on your machine.” That wording must not be used to infer that a cloud assistant never receives the data: Anthropic's privacy policy covers inputs supplied through connected services. This follow-up did not edit OpenKP or test its data flows.
Record and implementation
The machine-readable evidence-followup-2026-10-30.json records three resolved access flags: OpenKP Q2, OpenKP Q7 and OwnChart Q2. Only OpenKP Q2 changes an answer. The remaining primary-table unverified count is 15, down from 18. The original accepted-corrections.json is unchanged. The original review is unchanged. All classifications remain document-based drafts.
Current counts: 6 Constrains, 3 Leans constrains, 19 Mixed, 21 Leans expands, 7 Expands, 3 Not assessed. The compiler applies this follow-up after the ten accepted corrections and checks its baseline answers. Current answers and sources are visible in each tool's checklist table. Original report bodies remain historical evidence.
This follow-up is included in release 225439b, committed and pushed on October 30. Deployment verification is recorded in the release handoff (project record, not published). See current results and checklist rules.
Validation: compiler freshness, JavaScript syntax, classification/filter parity, score rendering and all 54 Python tests pass. The original review and all 59 preserved historical report bodies are unchanged. These are HugoScore implementation checks, not product tests.