Autonomous result & the safety net
A result that no clinician will ever read has to carry, on its own, everything a clinician would have supplied: what was actually assessed, what was not, how long it can be relied on, and the route back into care when something changes. It is the only screen in this system with nobody standing behind it.
Overview
Every other pattern here has a clinician in it somewhere. Even Confidence disclosure, which is entirely about calibrating trust in an algorithm, assumes a qualified reader who can weigh the number and disagree with it. Remove that reader and a set of things stop happening quietly.
Nobody notices the result was implausible. Nobody says "yes, but you should come back sooner than that." Nobody remembers that this person mentioned their vision had changed. Nobody creates the follow-up. Every one of those has to become a property of the interface, because there is no longer anyone to hold them.
An autonomous screening device sits at the far end of the harm-latency axis. When AcuteLine is wrong, the cath lab is activated or it is not and everybody finds out within the hour. When a screening result is wrongly negative, it surfaces as sight loss two years later, seen by clinicians who never used the software, and it is never traced back.
There is no natural feedback loop. So the interface has to manufacture one — a stated interval, a recall the system owns, and a route back that does not depend on anybody remembering to build it.
No reader, no assumption
Everything a clinician would have added is stated explicitly, because the person who would have added it is not there.
Dated, not final
A screening result describes one moment. It carries the date it stops being a reasonable basis for doing nothing.
The route back is part of the result
What to do if something changes before then is not an appendix. It is the half of the result that prevents harm.
Anatomy
Your eye screening result
No signs of diabetic retinopathy needing treatment were found in either eye.
Photographs of both eyes were checked by software. A person did not look at them, because nothing was found that needed a closer look.
| Part | Rule |
|---|---|
| The finding | First, in plain language, in one sentence. Scoped to the condition actually screened for — never "your eyes are healthy". |
| Who assessed it | Stated, not implied. "A person did not look at them" is information the recipient is entitled to and would otherwise assume the opposite of. |
| What was not checked | Required. The single most load-bearing element on the page — see below. |
| Expiry | A date, not an interval. "In 12 months" requires arithmetic from a date the reader no longer has in front of them. |
| The route back | Specific symptoms, a specific person to contact, a specific urgency. Never "consult your doctor if concerned". |
| Provenance | Software version and assessment date, so a result can be reconstructed years later when the model has changed. |
What was not checked
A person handed a negative eye-screening result reasonably concludes their eyes are fine. That conclusion is wrong, and it is wrong in a way that will keep them from an optometrist while something the device never looked for gets worse.
An autonomously issued result states what was not assessed, adjacent to the finding, at the same visual weight. Not in a footnote, not in an accompanying leaflet, not behind "learn more". The negative space around a screening result is the part that gets misread, and it is the only part the device can do anything about.
This is the same instinct as out-of-scope disclosure, hardened by the absence of a reader. A clinician receiving a scoped result already knows the scope. A patient does not, and has no way to find out.
- Name the conditions, do not gesture at them. "Other eye conditions were not checked" is not usable; glaucoma, cataract and macular degeneration are.
- Never let the scope be inferred from the service name. People do not distinguish "diabetic eye screening" from "an eye test", and the result is where that gets corrected.
- Say a person did not look. Recipients assume human review by default. A device that lets them keep assuming it has misrepresented itself by silence.
- Do not soften with reassurance the device cannot support. "Everything looks great" is a claim about a person's eyes; the device has a claim about four photographs.
Dated, not final
A screening result is a statement about one morning. Its usefulness decays, and the interface is the only thing that can say by how much and by when.
The recall is created before the result is shown, not after — so a result can never exist without one. A follow-up that depends on someone remembering to create it is not a safety net; it is an intention.
| Outcome | What the system creates |
|---|---|
| No referable disease | Routine recall at the programme interval, booked at the moment of issue, with the date on the result. |
| Referable disease | A referral with an owner and a due date — not a recommendation to the person to arrange something. |
| Ungradable | Referral to a human grader, plus a recall so the person does not fall out of the programme while the referral is in flight. See Capture & quality gate. |
| Partially assessed | Both: a result for the assessed eye and a referral for the other, each with its own date. |
| Recall could not be created | The result is not issued. The encounter is held and escalated to the programme coordinator. |
The route back
The interval assumes nothing changes. Sometimes something changes, and the entire safety of an annual screening programme rests on the person recognising it and acting before their next invitation. That sentence is the highest-stakes copy in the product, and it is usually the weakest.
If your vision becomes blurred, patchy or suddenly worse, or you see floaters or flashes, contact your optometrist or GP the same week.
Named symptoms, a named contact, a named urgency. A person can check themselves against this.
If you have any concerns about your eyes, please consult your healthcare professional.
Legally safe, clinically useless. It asks the person to self-diagnose "concern", which is exactly the judgement they lack.
- Symptoms, not conditions. "Signs of retinopathy" is not something a person can notice. Blurred vision is.
- A specific urgency. "Same week", "same day", "today" — not "promptly", which people scale to their own anxiety rather than to the risk.
- A route that exists. Naming a service the person cannot actually reach within the stated urgency converts a safety net into a formality.
- Always present on a negative result. This is precisely where it feels redundant and precisely where it does the work.
- Never behind a disclosure control. Not an accordion, not a tooltip, not a second page. See Accordion.
The copy on this page is written for a general adult reader with no clinical vocabulary, which is a different register from the rest of this system — Voice & tone's claim ladder assumes a clinical reader. Comprehension here is a measurable outcome, not a stylistic preference, and this system has not measured it. See Known gaps.
States
| State | Rendering |
|---|---|
| Issued — negative | Finding, scope, expiry, route back. No alarm hue: a negative result is not a confirmation banner and does not get celebratory styling. |
| Issued — referable | What was found, who will contact them, by when, and what to do if that does not happen. The referral's due date is on the result. |
| Awaiting human grader | Stated as pending with an expected date — never as a result. See Key–value pair. |
| Superseded | A re-analysis or a grader's decision does not overwrite the issued result. Both remain visible with their dates. |
| Expired | Past its recall date, the result renders as historical and says so. It never presents as current simply because nothing newer exists. |
| Delivery unconfirmed | An issued result the person has not received is an open item for the programme, not a closed encounter. |
Do's and don'ts
No signs of diabetic retinopathy needing treatment were found in either eye.
Glaucoma, cataract, macular degeneration and your glasses prescription were not assessed.
Scope stated at the same weight as the finding, in the same breath. The reader cannot take one without the other.
Your eye screening was normal. No further action needed.
Two overclaims in nine words. "Normal" covers everything the device never examined, and "no further action" contradicts the recall it just created.
Photographs of both eyes were checked by software. A person did not look at them.
The recipient assumes human review unless told otherwise. Saying it plainly is the only way that assumption gets corrected.
Your images have been assessed by our screening service.
True, and engineered to leave the wrong impression. "Assessed by our service" is what a person hears as "a clinician looked".
Next screening due by 14 August 2027. You will be invited — booked automatically today.
A date, and a commitment the system has already honoured. Nothing here depends on the person or a coordinator remembering.
Please arrange your next screening in approximately 12 months.
An interval instead of a date, and the follow-up delegated to the person least equipped to own it. This is how people leave a programme without anyone noticing.
Right eye: no referable disease. Left eye: could not be assessed — referred to a grader, who will contact you within 5 working days.
Two eyes, two outcomes, two next steps. Neither result is allowed to stand in for the other.
Screening complete. No referable disease detected. Some images were of limited quality.
A referral has quietly become a caveat. The person reads a negative result and will not expect anyone to contact them.
Accessibility
- This is the least expert reader in the system. Plain language, short sentences, and no clinical vocabulary that is not immediately explained — a different register from every other page here.
- The finding is a heading, not emphasised body text, so it is reachable by heading navigation and is the first thing announced.
- Scope and route back are in the reading order immediately after the finding, never visually adjacent but sequentially distant.
- Dates are written out — "14 August 2027", not
14/08/27, which is ambiguous across locales and unreliably announced. See International design. - The route back is never colour-only. It is an inline advisory with text, and it must survive being printed in greyscale — which is how a result handed over on paper is actually read.
- No alarm hue on a negative result, and no confirmation green either. See Colour.
- Reflow to 320 px (SC 1.4.10) — most recipients read this on a phone, which makes the narrow viewport the primary case rather than the edge case.
- Targets ≥ 24 px (SC 2.5.8), assuming an older reader on a small screen, possibly with tremor and without their reading glasses.
- Text, not an image of text (WCAG 2.2 SC 1.4.5). A result exported as a page image — a scanned letter, a flattened PDF — is unreadable to assistive technology and cannot be resized. Screening results are exported and posted more often than they are read on screen, which makes this the failure mode most likely to reach a recipient.
Outcomes of use
What this contributes to, in the terms of Usability & context of use. These are attributes believed to contribute to an outcome; the outcome itself is settled by observing real use in a specified context, not by this page.
- Effectiveness — the recipient's belief about what was assessed matches what was assessed, and they act on a change in symptoms rather than waiting for the next invitation. Both are measurable, and neither has been measured here.
- Efficiency — no coordinator effort is required to create a follow-up, because the system created it. The expendable resource in a screening programme is administrative attention, and it is spent on the people who did not attend.
- Satisfaction — trust that survives finding out a machine made the decision. Concealing autonomy buys short-term comfort and spends long-term credibility; the disclosure is what makes the programme defensible when someone asks.
Clinical safety notes
Trace these in your risk file (ISO 14971) and usability engineering file (IEC 62366-1).
- Scope of assessment stated adjacent to the finding, at equal weight. Mitigates: a condition-specific negative read as general eye health.
- Autonomy disclosed explicitly. Mitigates: a recipient believing a clinician reviewed their images.
- Recall created before the result is issued. Mitigates: a person leaving the programme because no follow-up was ever booked.
- Result withheld if the recall cannot be created. Mitigates: an issued result with no safety net behind it.
- Expiry expressed as a date. Mitigates: interval arithmetic performed wrongly, or not at all, months later.
- Symptom-based route back with a named contact and urgency. Mitigates: interval cancers and interval sight loss — deterioration between screening rounds that the person did not recognise as actionable.
- Route back present on negative results. Mitigates: the reasonable inference that a negative result means nothing can change.
- Per-eye outcomes never merged. Mitigates: one eye's referral disappearing into the other eye's negative result.
- Issued results never overwritten by re-analysis. Mitigates: a decision becoming unreconstructable after a model update.
- Undelivered results tracked as open. Mitigates: an encounter closed in the system that never reached the person.
Related
- Capture & quality gate — the other half of this arc, and the reason a result can be trusted to have had adequate input.
- Confidence disclosure — scope disclosure for a reader who can weigh it themselves.
- Voice & tone — the clinical register this page deliberately departs from.
- Key–value pair — pending is not negative, and absence is never blank.
- Alert Banner — why the route back is an advisory and not an alarm.
- Badge & pagination — counts and denominators, which a screening programme lives on.
- FieldGrade — the reference application and its position on the autonomy and harm-latency axes.