Accuracy

Everything in Clinicoid can be wrong — a key, an explanation, an answer from the AI. This is what each one is built from, how far it is checked, and what we do when you tell us it is not right.

Where the content comes from

The bank is the faculty's own papers and question banks, because a question a department never asked tells you nothing about the exam that department is going to set. What we add is the ordering, the explanations, and the record of what you got wrong. Clinicoid also writes a separate set of practice exercises for Pro. They are labelled as Clinicoid's own wherever they appear, and they are never mixed in with your faculty's papers.

That distinction runs through the whole app, and it is why almost every screen can tell you where something came from. A question knows its paper and its year. An explanation knows its pages. An answer from Coid shows you both.

What	Where it comes from	How far it is checked
Questions	The faculty's own past papers and question banks — the things that were actually set.	Extracted from the paper itself, with the paper, the year and the page it came from recorded against the question.
Answer keys	The official key of the paper the question came from, where that paper had one.	Where it had none, the question is published with no key rather than with a guessed one. That is a normal state and the app says so on the question.
Explanations	The module's own books and notes — the ones your course uses.	Written against specific pages, then read and released by a person. Where part of an explanation goes beyond your book, the question says so.
Written-answer marking	The mark scheme of that question, point by point.	Marked against the scheme rather than against a model paragraph, and the marks you earned are shown inside your own text.
Coid's answers	The pages of the books loaded for the module you are in. Not the internet, and not the model's general knowledge.	The pages used are shown under the answer, and tapping one opens the book at that page.

How an answer key is decided

Four rules, and the first one is the one that matters: the department's answer is the answer. Everything else exists so that we can disagree with it in the open rather than quietly.

One — The paper decides, not us: A question is published with the key the department published it with. Where we think that key is wrong, we do not overrule it — see the next step.
Two — A question with no key is published anyway: Some papers were set without an official key. We would rather show you the question as it was set than invent an answer for it, so it is published keyless, it says so, and it does not count against you.
Three — A doubt is shown, never applied: When grounding an explanation raises a real doubt about the official answer, the question carries an advisory saying so — both readings, and why. The stored answer is never quietly changed underneath you. You are told there is a disagreement and shown enough to judge it.
Four — Out-of-syllabus is marked and taken out of the score: A question the current syllabus no longer covers is labelled, with the reason. It is still there to read, because past papers are worth reading whole, and it does not count towards your marks.

What the AI is allowed to say

Coid answers out of the books loaded for the module you are in, and it shows the pages it used. That is not a courtesy — it is the constraint the whole feature is built around. An answer with a page under it is an answer you can check. An answer without one is a paragraph that merely sounds right, and in this subject those are the dangerous kind.

So it is built to refuse. A question it cannot ground in the pages it has gets an honest the books do not cover this rather than a confident guess, and it will not answer about a real patient at all — including the person asking.

The same rule governs the explanations attached to questions, and when the book and the question do not line up the app says which of the two it is:

Not fully covered by your book — the book covers the topic but never states the detail this question asks, so part of the explanation came from outside your notes.
Differs from your book — the book covers this and says something else. Both readings are shown, and the question still counts.

Neither of those two labels is ever applied by the AI. Telling the book does not state this apart from the search missed it takes a person who has read the page, so a person sets them.

Coid is an AI, and it can be wrong — including when it is quoting a real page, because it can quote the wrong one. Check anything that matters against your book. The sources under every answer are there so that checking takes one tap.

What Coid will and will not answer

Reporting an error

Students find most of the errors in Clinicoid. That is not a failure of the review — it is thousands of people reading closely, which no internal pass can match. The report button exists because it is the fastest route we have to being right.

One — Report it from the question: There is a button on every question. It sends us the question itself, so you never have to describe which one you mean.
Two — A person reads it: Not a queue that sorts itself. The question is checked against the paper it came from and the pages the explanation was built on.
Three — What happens next depends on what was wrong: A wrong explanation is rewritten. A key we now believe is wrong gets an advisory rather than a silent edit. A question the department never keyed goes keyless. In every case the fix reaches the question itself, which is where you will meet it again.

What makes a report actionable

What you believe the answer should be.
Where you read it — the book and the page, or the lecture.
Whether the problem is the key, the explanation, or the question itself.
A screenshot, if something looked broken rather than merely wrong.

We do not publish a turnaround, because a number we published in a quiet week is a number we would miss in an exam week. What we will say is that a report with a source in it is the one that gets fixed first.

Corrections we have made

We do not publish a running list of them, and it is worth saying why rather than leaving the absence to be read as evasion.

A public log of corrections is a list of question numbers that were once wrong. To a student not sitting that paper it is worth nothing; to one who is, it is worse than nothing, because it reads as the complete set of our errors and it never is. It would also be the one page on this site that quietly encourages you to trust everything not on it.

A correction is only useful where you meet the question, so that is where it goes: rewritten explanations appear in place, a doubted key carries its advisory on the question, and a question under review says so where you are standing. You do not have to read a changelog to get the corrected version — you already have it.

Not medical advice

Clinicoid is a revision tool for students. Nothing in it is medical advice, a diagnosis, a treatment plan, or a substitute for your faculty, your textbooks or a qualified clinician. Never use it to make a decision about a real patient, including yourself.

That is the wording from the Terms of Use, repeated rather than paraphrased. It is also the line Coid is built to hold: it will not answer a question about a real patient, and it does not become willing to if you tell it the patient is you.