Public transparency record · sanitized
Public Disclosure Summary (ATRS-style, voluntary)
A plain-language public summary of the medical digital twin, held to a public-sector algorithmic transparency standard. It answers four questions any citizen, reviewer, or journalist can read without being a user: what it does, what it does not do, who is responsible, and how to raise a concern.
Why this exists
Most AI transparency is written for people who already use the system, or is kept in an internal inventory that only the organisation sees. Public-sector transparency standards — such as the UK’s Algorithmic Transparency Recording Standard — ask a harder question: can someone who is affected by the system, but does not use it, and who may be a member of the public, a reviewer, or a journalist, understand what it is and how to challenge it?
The medical digital twin I build is a pre-deployment research prototype, so it is not legally required to publish such a record. I publish this one voluntarily, held to that public-sector standard, because a governance discipline is worth establishing before deployment makes it obligatory — and because the ability to explain a system in plain language to a non-user is itself a test of whether it is well governed.
The summary
1. What it does. It is a supportive tool that monitors health signals, explains what they may mean in plain language, is explicit about what remains uncertain, and flags when a person should seek clinical attention.
2. What it does not do. It does not diagnose. It does not replace a doctor or nurse. It does not decide or change treatment. It is not an emergency service. When information is missing or a question falls outside what it can safely address, it does not give a confident-sounding answer. It also does not simply stop: it says what it could not determine, what would settle it, and what the person can usefully do in the meantime. Declining is easy for any system; being left with nothing to act on is what sends a person somewhere less safe.
3. Who is responsible. Accountability sits with named roles, not with the software: an AI governance lead is accountable for how the system is allowed to behave; a clinical safety owner is accountable for patient safety; the project sponsor is accountable at the institutional level. No part of that responsibility is transferred to the system or to the patient.
4. How to raise a concern. If an output looks wrong, or something about the system is concerning, there is a defined route to reach a responsible human — not only an in-app button, but a monitored channel with a named owner and a response commitment. A concern raised this way is treated as a potential safety signal, not a satisfaction complaint.
What changed, and why
A transparency record that only ever shows its current state is a weak one. It cannot be checked, and it quietly conceals that an earlier version said something different. This section is permanent; every future revision adds to it rather than replacing it.
v0.2 (22 August 2026). Version 0.1 said that when information is missing or a question falls outside what the system can safely address, it “says so plainly rather than giving a confident-sounding answer.” That wording was challenged, and it did not survive the challenge. Declining is a property of any system that stops: a system that ends the exchange and a system that hands something back both satisfy it, and for the person asking they are not the same system at all. Describing the stop as a safety feature, without describing what the person is left holding, overstates it.
Version 0.2 therefore states what a refusal has to hand back: what could not be determined, what would settle it, and what the person can usefully do in the meantime. The internal record is held to a stricter version of this than the page describes — the completeness of that handover is now a monitored indicator rather than a design intention, and the further question of whether sustained use leaves a person less able to manage without the system is recorded as a measure that does not yet exist, rather than one that is passing. The criterion is developed further in forthcoming work, where the person who identified the weakness is credited.
What this demonstrates
This is the public-facing counterpart to the internal governance record. Internally, the same system is documented through a patient consent-and-safety case, a full disclosure inventory across every touchpoint, and a scope-and-authority statement. This summary is deliberately the one document that requires no technical background and no prior use to read — because transparency that only a user can understand is not, by itself, public accountability.