Why DILR Often Peaks Only in Your Second CAT Attempt

DILR

It is a pattern repeaters describe often enough to be worth examining: Quant and VARC improved steadily through the first year, while DILR stayed flat and then jumped in the second attempt. The usual explanation offered is experience, which is true and useless, because experience is not something you can go and get.

The useful version names what specifically changed. DILR rewards a different kind of learning from the other two sections, and that kind is slower to accumulate and harder to feel while it is happening. This piece explains the four mechanisms behind the second-attempt jump, and, more importantly, which of them a first-time candidate can install deliberately rather than waiting a year for.

Why DILR Improves Differently From the Other Two

The comparison with Quant and VARC is where the explanation starts.

Quant improvement is largely additive. You learn a chapter, your accuracy in that chapter rises, and the gain is visible because it is attached to identifiable content. VARC improvement is slower but still fairly continuous, because reading speed and structural tracking improve gradually with volume.

DILR does not work that way. It is not chapter-based, set types and counts swing between years, and the skill being built is pattern recognition across a wide and loosely defined space plus judgement about cost. You can practise for two months, feel no different, and then find that sets which used to take twelve minutes now take six, because enough exposure accumulated to cross a threshold.

That lumpiness is the first half of the explanation. A flat first year is not evidence that nothing was learned, and candidates who conclude it is frequently abandon the approach that was about to pay.

What Actually Changes in the Second Year

Recognition Breadth Is Just Volume

Of the four, this one is the most mechanical and the easiest to compress.

Recognition means seeing a set and knowing its family within seconds, which brings a default structure with it and removes the expensive cold start. The recurring families are stable enough to learn: arrangements, games and tournaments, distribution and selection, tables and caselets, networks, scheduling.

A repeater has simply seen more instances of each. There is nothing in that a first-time candidate cannot replicate by working type by type deliberately rather than doing mixed practice from the start. Type-wise volume builds the recognition, and mixed practice afterwards trains identifying the type under pressure. Doing only mixed practice from the beginning is the commonest reason recognition builds slowly.

Cost Judgement Needs Feedback, Not Time

The second mechanism is where most of the gain lives, and it is the one a first-timer most often fails to build.

Repeaters have a calibrated sense of what a set will cost because they have repeatedly estimated, been wrong, and noticed. That feedback loop is what produces calibration, and it is available to anyone who records the data.

The obstacle is that most candidates do not record time taken per set, so the loop never closes. A set solved in six minutes and a set solved in fifteen look identical in a log that only marks right and wrong, and no amount of practice calibrates an estimate against data that was never collected.

Selection Is Usually Learned Through Loss

The third mechanism is the least pleasant and the most decisive.

A first-time candidate's instinct is to attempt what they can solve, which sounds sensible and ignores cost entirely. It takes losing a section to a single set, in a way that actually matters, for most people to internalise that choosing is the primary skill and solving is secondary.

DILR is more unpredictable than Quant, which is precisely why this matters more here. In Quant, weak selection costs some questions. In DILR, where one set can absorb a third of the section, weak selection costs the section. A repeater knows that as a lived fact rather than as advice.

The installable version is to treat a mock's DILR section as a selection exercise rather than a solving exercise for a few attempts. Sweep all the sets, rank them by cost, start with the cheapest you are confident about, and accept a lower attempt count while the habit forms.

Mechanism	What it really is	Can a first-timer install it
Recognition breadth	Volume, type by type	Yes, directly, with sequenced practice
Cost judgement	A closed feedback loop	Yes, by predicting then recording
Selection discipline	Usually learned through loss	Partly, with deliberate selection mocks
Lower panic	Familiarity with the conditions	Partly, through conditioned full mocks

Compressing a Year Into a Quarter

If three of the four mechanisms are installable, it is worth laying out what installing them actually looks like across a few months, because the order matters.

The first block is type-wise volume. Work through the recurring families one at a time, staying with each long enough that the shape becomes familiar rather than moving on after a handful. The aim is not to solve a large number of sets, it is to reach the point where a new instance of that family reads as familiar. That is the recognition component and it is pure exposure.

The second block is mixed practice with prediction attached. Now the families are no longer announced, so you are training identification under uncertainty, and every set carries a written prediction of attempt-or-leave plus a minutes estimate before you start. The predictions are the point. Solving is how you check them.

The third block is sectional practice where selection is the object. Sweep every set, rank by cost, start with the cheapest you are confident about, and accept a lower attempt count while the habit settles. Expect these sections to feel worse than usual, because you are spending minutes on reading rather than solving and the return arrives later.

Run in that order, the three blocks reproduce most of what a repeater picks up over a year, and they do it deliberately rather than by accumulation. What they will not do is feel productive in the way solving sets feels productive, which is the main reason candidates skip them.

The Part That Genuinely Needs Having Sat It

Honesty requires admitting that one mechanism does not fully compress.

Sitting the actual exam once removes a layer of novelty that no mock entirely reproduces. DILR is where that matters most, because panic degrades judgement before it degrades computation, and judgement is what the section runs on. A candidate who can still do arithmetic while rattled may have lost exactly the capacity the section was testing.

What narrows the gap is conditioning rather than repetition of the real thing. Full length mocks in one sitting, sectional limits enforced, no pauses, and ideally at your allotted time of day. That does not fully replicate exam day and it removes most of the avoidable novelty, which is the portion you can actually control.

Do Not Treat This as a Plan

There is a reading of all this that is worth heading off directly.

Noticing that DILR often peaks in a second attempt is not a reason to treat the first attempt as practice. The mechanisms behind the jump are mostly installable now, the cost of a deliberate extra year is a year, and a candidate who mentally files the first attempt as a rehearsal will prepare like someone rehearsing.

Use the observation the other way. It tells you exactly which components repeaters accumulate accidentally, which means you can build them on purpose and arrive at the first attempt with a section that would otherwise have taken two.

The Summary

DILR peaks late for many candidates because it improves in steps rather than slopes, and because the skills it rewards accumulate quietly and are hard to feel building.

Four things change by the second attempt. Recognition breadth, which is volume seen type by type. Cost judgement, which is a feedback loop between estimates and reality. Selection discipline, which most people learn by losing a section to one set. And lower panic, which is the only one that genuinely benefits from having sat the exam.

Three of those four are installable now. Practise type by type before mixed, predict cost and record actuals so the loop closes, and run a few mock sections as selection exercises rather than solving exercises. Do not plan a second attempt, because the observation about repeaters is a list of what to build early rather than a reason to wait.

If your DILR has been flat and you cannot tell whether the approach is failing or just has not landed yet, that is worth settling with data rather than instinct. A CAT preparation strategy review will read your set-level pattern, and a personalised CAT preparation plan sequences the type-wise work before the mixed practice.