Safety and governance
Clinical judgement stays with clinicians.
Every proposal Reedley prepares is designed to carry its evidence, pass a review floor and record who decided what; here is what exists, what is underway and what remains.
- Human review designed into every clinical action, with draft floors practices cannot lower.
- Provenance on every line: the passage, entry or rule behind it.
- Append-only history of decisions, with the record as it stood (development server store; in memory in the browser build).
Statements reviewed 14 November 2026.
Today, in one paragraph
Reedley is a product in development, tested on synthetic data.
No GP practice uses it. No real patient data has been processed. No integration with EMIS Web, EMIS-X or SystmOne exists or is claimed. No Clinical Safety Officer has been appointed, so none has reviewed the draft hazard log, and no medical-device assessment has been made. The proposed company, Reedley Clinical Systems Ltd, is not yet incorporated. We will update this page as each of these changes.
Assurance timeline
What has happened, and where each standard stands.
Dated events are evidenced in the repository. The present state is read from the assurance register as at 14 November 2026; no planned item carries a date.
- 4 Oct 2026
- Reedley brand identityThe Reedley identity was established, including the rule that the brand’s Reed Green may signal workflow progress but never clinical severity or patient outcome.
- 5 Oct 2026
- Build block 2: one product, one lifecycleThe earlier document demonstrator became part of a unified product shell around one shared work engine and one lifecycle.
- Build block 3: single-pass document reviewDocument review was rebuilt as a single pass, with identity resolution, related correspondence, coding proposals and review policy floors.
- Build block 4: freshness, access policy and request preparationThe fourth build block added freshness tracking that marks derived information stale, an access policy within the practice, and new prescribing and results workflows — all on synthetic data. These rules are enforced by Reedley’s own screens, not yet by a server.
- 8 Oct 2026
- Public website previewA first preview of this website, including an interactive demonstration on synthetic data, a trust centre and an honest assurance register.
- Structured internal challenge reviewEight perspectives, thirty findings, run with AI-assisted review tools; not an independent clinical, safety or regulatory review. All findings addressed or listed in the launch checklist; request preparation downgraded to “Needs clinical sign-off”.
- 31 Oct 2026
- Statements re-checked against the buildEvery public statement re-checked against the build. Internal AI-assisted reviews found 2 critical and 21 important wording problems; all fixed. Two content rules added: evidence must prove the event, and received is not reviewed.
- 1 Nov 2026
- Statements rewritten after internal reviewsInternal AI-assisted reviews from four perspectives — clinical, NHS digital and security, user experience and conversion, accessibility and performance — found 12 critical and 70 important problems. The statements were rewritten and re-checked against the build the same day. Not an independent review; the founder’s review of the new wording is pending, and no Clinical Safety Officer has been appointed.
- 14 Nov 2026
- Restart baseline and first model runThe clinical build was re-baselined on synthetic data. A first model-in-the-loop run on synthetic development documents was recorded; the locked holdout was not used. Access, freshness and closure rules were moved into the clinical logic on a branch that is awaiting review and not yet part of the development build.
- Next · no dates are committed
- Pilot partners and a Clinical Safety OfficerFounder, then independent walkthroughs on synthetic data; then a governed pilot. Nothing moves to real data without the approvals described on the pilot page.
Present state · as at 14 November 2026
- DCB0129In progress
- DCB0160Not applicable
- Medical device qualification (MHRA, UK MDR 2002)Not yet assessed
- DTACPlanned
- DSPTPlanned
- ICO registration and supplier DPIA (UK GDPR)Planned
- Cyber Essentials and Cyber Essentials PlusPlanned
- Independent penetration testPlanned
- Clinical-system integration (for example IM1 pairing)Not yet assessed
- WCAG 2.2 level AA (this website)In progress
- Real-world evaluation with practicesPlanned
What the statuses mean
Completed and evidenced; in progress; planned; not yet assessed; not applicable (with the reason). Nothing here indicates approval, certification or endorsement by the NHS, NHS England, the MHRA or any other body. The full register has the evidence and the next step for each item.
Assurance status
Every relevant standard and certification, with a factual status and what comes next.
ReadClinical safety
How clinical risk is being managed, DCB0129 and DCB0160 explained, and what is not yet in place.
ReadResponsible AI
Where AI is used, what it must never do, how it is evaluated and who is accountable.
ReadSecurity
Our security approach for the product and this website, separating what exists from what is planned.
ReadIntegration and compatibility
EMIS Web, SystmOne and the routes to integration — with no claims we cannot evidence.
ReadReport a vulnerability
How to tell us about a possible security issue, and what to expect.
ReadTrust
Built to be checked, not just believed.
Clinical safety, responsible AI and information governance are foundations, not features. Our Trust Centre states what exists, what does not yet, and what comes next.
Clinician-centred design
Designed for the people who sign: dense but legible, one primary action per view, evidence never more than a click away.
Human oversight
Every proposal needs a named person. Review levels are designed with floors that practice settings cannot lower (draft levels, awaiting CSO sign-off).
Explainability and traceability
Every statement links to its source. Confidence is shown as a reason — “dose stated explicitly on p.2” — never a percentage.
Clinical safety, stated as it is
A draft hazard log and an assurance register that says what does not exist yet. No Clinical Safety Officer has been appointed; one will own the hazard log before any real-world use.
Responsible handling of information
Synthetic data only today. Practice-level isolation, role-based access and explicit grants are designed in from the start.
Clear accountability
Append-only history of who decided what, when and on what evidence. Decisions belong to named people.
Assistance, not autonomy
Reedley suggests. A clinician decides, and is recorded.
Human responsibility is part of the architecture, not a setting. The development build is designed and tested against these boundaries; in that build they are enforced by Reedley’s own screens, not yet by a server, and they shape how we describe it.
Reedley is designed to
- Read correspondence and results with the reviewer, citing the exact passage for every finding
- Compare what a document says with what the record shows, and say where they differ
- Propose work, coding and record updates for a named person to accept, edit or reject
- Route work to the right person with a due date and the reason it exists
- Look for evidence that work actually happened before treating it as complete
- Keep an append-only history of who decided what, when, and on what evidence
Reedley does not
- Prescribe, or add, stop or change a medicine in the clinical record
- Diagnose, or interpret what a result means for a patient
- File, code or send anything without a person’s decision
- Treat conditional wording such as “consider” as an instruction
- Verify work on a tick. A person’s report moves an item to actioned; verification comes from the record where Reedley can read it; where it cannot, a named person’s structured statement is accepted today, and that rule is under review.
- Replace the clinical system, which remains the record