This article is adapted from a post in Dr Koshy Jacob's "Building Revise Radiology in Public" series on LinkedIn.

We hold something in the region of 23,000 cases. That is the figure a company in our position would normally lead with, and having thought about it for a while I have concluded it is close to the least useful thing I could tell you about Core.

The number that matters is not how much material exists somewhere in our collection. It is how much of it has been checked, found genuinely useful, and put in front of you in a form you can work through on an ordinary Tuesday evening after a long list. Those are very different figures, and I would rather explain the second one honestly than advertise the first.

So this article sets out what Core is, how material gets into it, what it does, and where I think it genuinely helps. One thing to say plainly before I start: what follows describes a prototype and a direction of travel. Some of it is live, some of it is weeks away, and some of it is further out than that. I will come back to this at the end, but I would rather you knew it in the first paragraph than the last.

What Core is meant to be

Core is a library rather than a course. There is no start date, no end date, and no cohort moving at a pace that is not yours. You subscribe, it does not expire, and you use it at whatever depth the evening allows.

It exists because the thing I could not get when I was preparing, and still could not honestly claim even after I passed, was any way to know that I had seen what I needed to see. That is what a library organised around coverage is for, and it is why Core is built the way it is.

Let me be honest about the numbers

So let me do the arithmetic properly, because the honest answer is more complicated than a headline and I think more reassuring.

Core does not contain all 23,000 of those cases and is not going to. A substantial portion is allocated to the Mastery courses and to our live courses, where it is curated for a specific exam and a specific cohort. More importantly, nothing enters Core until it has been checked and found to be genuinely useful. Not merely correct, and not merely present. Useful.

The question banks work the same way. For FRCR 2A we hold around 8,000 questions. About 2,500 of those are going into the Mastery course, and perhaps another 500 into the live course. The remainder will be reviewed before any of it is added to Core. The same principle applies across every content type we hold.

What this means practically is that Core grows rather than arrives complete, and that the number of cases in it on the day you join will be smaller than 23,000 by a considerable margin. I would rather tell you that plainly than advertise a headline figure that includes material we have not yet stood behind.

It is also, I think, the right way round. Anyone who has followed this series knows we spent months cutting our short case packets from 40 to 30 rather than padding them out, and that we did so because coverage matters more than volume and because a candidate's time is the scarcest thing in this entire equation. It would be strange to apply that discipline to the courses and then tip everything unchecked into the subscription. A candidate reading carefully at midnight will find the weak case, and they will be right to.

One pool of cases, answered at any depth

This is the design decision I am most pleased with, and it is unusual enough to be worth explaining.

Most platforms separate material by exam format: here are the short cases, here are the long cases, here are the rapid reporting sets. We have deliberately not done that. Core holds one pool of reportable cases and you decide how deeply to answer each one. The same case can be a one-line spotter call, modality and diagnosis and nothing further, or a findings-and-differential answer of two or three sentences, or a full structured report across observations, interpretation, principal diagnosis, differential and management.

Why this matters on an ordinary Tuesday: with 20 minutes after a long list, you can make 30 rapid gestalt calls. With an unhurried hour on a Saturday, you can write four full reports properly. It is the same library either way, and nothing needs to be re-bought or reorganised. That is the only arrangement I have found that survives contact with a working radiologist's actual week.

There is a rapid reporting mode for high-volume normal-versus-abnormal calls at pace, and a viva mode for the more complex cases that suit being presented aloud.

You answer the way you actually work

You can type your report, dictate it, or record yourself presenting it. The recording option matters because speaking a case is a different skill from writing it, and most people discover this uncomfortably late. Recording a viva presentation alone at your desk and then listening back against the model answer is not the same as sitting in front of an examiner, but it is a great deal better than never having heard yourself do it once.

A proper viewer, not a picture of a scan

Cases open in a real image viewer rather than as flat pictures. Where a case has them you get the reconstructions, and the full sequences, so that you are scrolling through a stack rather than studying a single slice that somebody else has already chosen for you. There are window and level presets and magnification tools.

That sounds like a technical detail and is really a learning one. Practising on one static, pre-selected image trains something the examination does not ask of you. Scrolling through a study to find the abnormality yourself, and deciding for yourself which slice matters, is a large part of the actual skill, and it ought to be part of the practice.

The AI tutor, described precisely

Two AI features sit inside each case, and I would rather describe them accurately than enthusiastically.

The first is answer marking. You write or dictate your report, submit it, and the tutor reads it against the model answer and returns a structured assessment: a mark, where you were strong, where the gaps were, and a verdict. It is not a substitute for a consultant's judgement, and any marking of free-text radiology reporting carries a degree of subjectivity that I would rather acknowledge than paper over. What it does reliably is tell you when you have missed something you should have said, which is most of what a candidate needs and is available at 11 at night when no consultant is.

The second is a conversational tutor which has the full context of the case and the model answer and which you can simply ask things. Why is that the principal diagnosis and not the second one. What would change the management. What is that structure. I suspect this is the feature that will matter most over years rather than months, because the question a learner most needs answered is usually the one they are too embarrassed to ask a colleague twice.

One limitation worth stating: the marking works on written and dictated answers, not on recorded audio. For recorded vivas you listen back against the model and mark yourself.

Spaced review, which is the engine underneath

Cases you find difficult come back to you. Core keeps a review queue and returns hard cases at spaced intervals, and you can add any case to that queue yourself when you know you have not really got it.

I have written before about why this works, so I will not repeat the evidence beyond the essential point: material retrieved after a delay, once it has begun to fade, holds in a way that material re-read immediately does not. Spaced review is the mechanism that turns a library from something you have looked at into something you know.

Knowing what you have not seen

Core tracks coverage and retention separately, and the distinction is more useful than it sounds. Progress on any topic shows as two figures: what you have mastered, and what you have attempted but not yet retained, against the total available. The remaining gap is material you have not opened at all.

This is the thing that did not exist when I was preparing. Not a feeling of having worked hard, which is unreliable, but an actual map of where you stand.

Case of the Day, free forever

One case a day, the same case for everybody, free whether you subscribe or not. There is a streak counter and a shield for the days when life makes it impossible.

Why free: partly because a daily ritual builds the habit that matters more than any single long session. Partly because it is the most honest advertisement available to us, since you can judge our material by working through a real case rather than reading our claims about it. And partly because a radiologist who cannot currently pay for anything should still have something.

The forum, and a quality loop that genuinely works

Every case and question carries its own discussion, and there is a wider forum with channels for general discussion, announcements, feature requests and problems. You can comment on cases and you can flag one to our team, with a reason, when something looks wrong.

That flagging mechanism matters more to us than it appears. A large library will contain errors, and the fastest way to find them is a candidate reading closely who notices that a differential does not belong. Some of the most useful corrections we have made this year arrived exactly that way, including a few that were uncomfortable to receive.

Points and badges, with a caveat

Core awards points for the work you actually do, and those points go towards the cost of our courses. There are badges at various milestones along the way. I am deliberately not setting out the arithmetic here, partly because the detail will change as we learn what works, and partly because the numbers are not the interesting part.

The caveat is that I am wary of gamification and would rather say so. Points are a nudge for the evenings when you are tired and need one, not the reason to do the work. Ignore them entirely, report cases, and you will get the same benefit without doing it wrong. They exist because streaks genuinely help some people sustain a habit, not because radiology should be a game.

A prototype, not a finished product

I want to return to this properly, because it is the most important caveat in the article and it would be easy to skim past. Everything above describes a prototype and an intention. A good deal of it works today. Some of it arrives in the coming weeks. Some sits further out than that, and a few parts will no doubt change shape entirely once real radiologists start using them and tell us what is wrong.

Please do not read the preceding pages as a feature list you can tick off this afternoon. Read them as where we are going, and as a commitment we are making in public, which is rather harder to quietly abandon than one made internally. If a particular feature matters to your decision about whether to subscribe, ask us and we will tell you plainly whether it exists yet. I would much rather field that question than have somebody join expecting something that has not been built.

And yes, someone could copy this

There is an obvious objection to publishing all of the above, which is that I have just handed a detailed specification to anybody who fancies building the same thing. I have thought about that and I am comfortable with it.

Partly because we are not really in the business of defending a moat. We exist to serve the Learning Radiologist, and if a competitor reads this and builds something that helps radiologists learn better than our version does, then radiologists are better off, which was the point. I would rather be part of a field that is improving than the sole owner of a mediocre idea.

Partly, too, because the features are the easy part. Anybody with a good developer can build a viewer with window presets and a spaced review queue. What takes years is the case collection sitting underneath it: something in the region of 23,000 cases, gathered over more than a decade, which is probably one of the largest of its kind anywhere, and every one of which has to be found, prepared, written up and then checked by a consultant who knows the subject. That is the part nobody can shortcut, and it is the part I would still be proud of if every other feature were copied tomorrow.

Why I think it is genuinely useful

Three reasons, in ascending order of importance.

First, it fits a real life. 20 minutes counts. Nothing expires. No cohort leaves you behind. The commonest reason people stop preparing is not weak will but a resource demanding more continuous time than a clinical job permits.

Second, it answers the coverage question. You can see what you have seen, what you have retained, and what you have never opened, which removes a good deal of low-grade anxiety and points your effort at what is actually weak rather than what is comfortable to revise.

Third, and this is the one I care most about, it is built for a career rather than for the 18 months before an examination. There is no date in Core because most of a radiologist's learning happens without one. The registrar with no sitting booked, the consultant moving into an unfamiliar subspecialty, the radiologist who simply wants to be better at chest films than last year, all of them are doing the same thing a candidate is doing, and none of them are served by something that ends.

That is what Core is becoming. A curated and growing library where nothing arrives unchecked, a proper viewer, a tutor available when nobody else is, a mechanism that returns what you nearly forgot, and an honest account of where you stand. It is priced low deliberately, for the reasons I set out in the last article. Everything after that is up to you, which is rather the point.

Originally shared by Dr Koshy Jacob on LinkedIn. Follow the series there