ENFR
Home  /  The gray-zone decision study
Study

Same case, same numbers, different decisions

A study in how stroke physicians decide when the evidence does not decide for them — and a way to see where your own judgement sits.

Put the same acute stroke case in front of ten physicians — the same age, the same deficit, the same imaging, the same clock — and you will not get ten copies of one decision. Somewhere between the numbers and the needle, the answers spread. Everyone who takes stroke calls has seen this happen in a room; almost no one has measured it on themselves.

Daniel Kahneman gave the spread its grammar. Bias is being wrong in the same direction every time. Noise is a group being wrong in different directions at once. Medicine has spent forty years auditing the first and almost none measuring the second.

The two hide in different places. Bias can be found from inside: audit your own decisions and the systematic lean eventually shows itself. Noise cannot — a single judge, examined alone, looks consistent. The scatter only appears when many judgements of the same case are laid side by side.

Kahneman's four targets Four ringed targets in a row. Accurate: shots tightly clustered on the centre. Biased: the same tight cluster, moved up and to the right of centre. Noisy: shots scattered widely around the centre. Biased and noisy: shots scattered widely and moved up and to the right. Accurate Biased Noisy Biased and noisy Kahneman's four targets Four ringed targets in a two-by-two grid. Accurate: shots tightly clustered on the centre. Biased: the same tight cluster, moved up and to the right of centre. Noisy: shots scattered widely around the centre. Biased and noisy: shots scattered widely and moved up and to the right. Accurate Biased Noisy Biased and noisy
Kahneman's four targets: bias moves every shot the same way; noise spreads them — and only the pattern, never a single shot, reveals which is which.

γνῶθι σεαυτόν — know thyself, the Delphic instruction — names the one measurement a clinician cannot take alone. You can know your own reasoning perfectly and still not know whether you are the outlier in the room. The only way to find out is to put the same case to everybody at once.

That is what this study does. Constructed acute stroke cases, each sitting deliberately where the trials went quiet, are answered independently by physicians who make these decisions for a living. Laid side by side, the answers show each of us something no audit of our own files can: where we actually stand.

you The spread on one case A horizontal strip running from would not treat on the left to would treat on the right. Twelve invented answers sit as dots along it, most of them massed left of centre with two far to the right, and one marked position, labelled you, sits just left of the middle.ILLUSTRATION — INVENTED VALUES, NOT DATA
would not treatwould treat

12 invented answers on the strip.

Nothing you add here is stored or sent.

An illustration with invented values — the real distribution, one strip per case, replaces it on this page on the evening of 16 September 2026.

Taking part

Who it is for. Physicians who make, or contribute to, acute stroke treatment decisions. No patients are involved and no patient data is collected — every case is constructed, written as fiction from the start.

What it asks. Six short cases, about ten minutes, answered once. Consent and data handling live on the survey platform, which presents its own information screen before any question; this page collects nothing and stores nothing.

Answered alone. Please answer without conferring — independent judgement is the measurement, and an answer agreed with a colleague measures the room instead of you.

What happens next. The aggregate distribution appears on this page the same evening, and each participant receives their own position against it by email within 24 hours.

The link to take part will appear here when the survey opens, before the meeting on 16 September.

Questions first: [email protected]