In short
- Not a medical device and not a diagnosis. Baseline does not diagnose, treat, monitor or prevent any condition, and is not a replacement for professional care. It is not an emergency or crisis service.
- Research stage, synthetic data only. There is no production service. Both prototypes run on fictional people only, and the therapist view refuses to run with real data. No study with participants has started.
- No certification or approval. We hold no security certification, no regulatory approval and no completed data-protection assessment.
- Treated as sensitive. Information about someone’s wellbeing can count as special-category (health) data under the GDPR. We therefore design as if it does.
This page is an engineering summary, not legal advice, and it will change as the work and the legal assessment progress.
The therapist view (Perspective)
How the current prototype is built, and what that does and does not mean.
| Design choice | Where it stands | Status |
|---|---|---|
| Fictional patients only | The store refuses to load unless every record is marked synthetic. There is no real-data mode and no way to create a non-synthetic patient. A production build refuses every page unless synthetic demo mode is switched on explicitly. | Prototype |
| Runs locally | The demo binds to the local machine only. Data stay in a local file; nothing is sent to an external service. | Prototype |
| No sign-in yet | The therapist role is a demo switch, not authentication. Patient links carry a one-session token that does not expire. This is not acceptable for real patients and is not intended to be. | Not built |
| Patients see no scores | A patient completes questions and sees a “you are done” screen, with no scores, map or interpretation. Who may see what in real use is not decided. | Prototype |
| No verdicts | No norms, percentiles, cut-offs, diagnostic or risk scores and no composite scores are computed. Generated text is tested against a list of forbidden conclusion terms. | Prototype |
| No AI model | No language model is connected. Free text is shown verbatim to the therapist and not analysed or sent anywhere. | Design |
| Separate records | Responses, results and identity are kept in separate collections so that access rules can be decided later. Clinician notes never appear in the patient view. | Prototype |
“Prototype” means demonstrated in early software on synthetic data. It does not mean production-ready, tested for security or independently reviewed.
Personal baseline engine: commitments and status
These are the principles the engine prototype is built toward. The right-hand column says how far each one has got.
| Commitment | Where it stands | Status |
|---|---|---|
| Hosting in the EU | Not deployed. The choice of host and any AI provider is constrained by this goal. | Planned |
| Encryption at rest and in transit | Not built. The prototype keeps data in memory only. | Not built |
| Identity and wellbeing data kept apart | The prototype stores them separately, linked only by a pseudonymous ID. | Prototype |
| Opt-in per data source | Each source requires consent; withdrawing a source deletes its data. | Prototype |
| Export and delete everything | Implemented in the prototype as functions; not yet offered in a product. | Prototype |
| Minimise raw data | The iOS prototype sends daily totals rather than raw health samples; journal text is encoded and not stored; calendar data is counts and hours only. | Prototype |
| Review and correct what Baseline believes | A person can mark a detected change accurate or incorrect. Correcting individual features is not built. | Partial |
| AI-generated text is labelled | Explanations are marked as AI-generated in the prototype interface. | Prototype |
| Language model sees as little as possible | Only the detector’s structured output would be sent to a language model, not text or sensor data. No provider has been chosen. | Design |
| No advertising, data sale or employer access | A policy commitment of the team, not a technical or contractual guarantee. | Policy |
Out of scope for the first version: GPS, message content, and microphone or any passive audio.
Questionnaire licences
Established questionnaires keep their original wording, response format and scoring, and each is scored separately. We checked their terms of use on 1 November 2026; this is our reading, not legal advice.
- Primals Inventory (PI-18). Free for research, clinical or personal-reflection use; commercial use requires contacting the author first. Used here only in an internal research prototype. Permission is needed before any commercial deployment.
- Mini-IPIP. Part of the public-domain International Personality Item Pool.
- Human Values Scale (PVQ-21). Licensed for non-commercial research in its documented form only. Because the prototype would use it inside a composite assessment for product development, it is not administered and none of its items are stored until the authors give written consent.
- Word selection, daily check-in and free text. Written by Baseline, experimental and without psychometric evidence. They are never scored together with the established questionnaires.
Frameworks we are assessing
Being on this list means we are looking at it, not that we meet it.
- GDPR. We assume wellbeing information is special-category data. A data protection impact assessment, a lawful basis, and consent wording for research use versus product use are still to be settled, and are needed before any real data are processed.
- EU Medical Device Regulation (MDR). Whether software of this kind could qualify as a medical device depends on its intended purpose. This matters most for any output a therapist could read as guidance, such as a flag or an analytic statement. The question is open. Until it is resolved, Baseline is positioned as non-clinical and makes no diagnostic or treatment claims.
- EU AI Act. We are reviewing how it would apply if a language model were integrated, including its restrictions on emotion recognition in workplaces and education. This is one reason we commit to no employer access.
- Research ethics. Any study with participants would need prior ethics review and informed consent. None has been submitted or approved.
We have not claimed, and do not currently claim, conformity with any specific technical standard or certification scheme.
This website
The site is static. The only personal data it collects is the email address you type into the research-preview form. The form is processed by Netlify Forms, a third-party service, and we use the address only to contact you about the research preview. This site also loads fonts from Google Fonts, and one page loads math-rendering files from a public CDN; those providers may see your IP address. A full privacy notice for the website is still to be written.
The charts on this site are simulated illustrations and the screenshots show fictional patients, not data from real people. The papers we list are external research and are not evidence that Baseline works.
Corrections
If you think something on this site overstates what Baseline is or does, please tell us so we can fix it.