CAT DILR Practice: 75% Accuracy at the 35th Percentile

Across 23 full mocks between 10 May and 24 August, my DILR raw score has run from 2 to 30. Same section, same person, a 3.5-month window. That's not a claim about how volatile DILR is in the abstract - it's the actual range in my own log, and it's wider than either VARC's ceiling swing or QUANT's. If you're newer to CAT and DILR feels uniquely unpredictable, this is why: it is.

The good news is that the one real sectional test I've logged for DILR points at something specific and fixable, not at a vague need to "practice more."

One sectional, and what it actually cost me

On 17 May I sat an IMS DILR sectional. Score: 18 out of 66. Attempted: 8 questions. Correct: 6 of those 8. Percentile: 35.

Read those numbers in order and the story is not "I'm bad at DILR." Six correct out of eight attempted is 75% accuracy - genuinely good. The problem was volume: I only attempted 8 questions out of a much larger set. A second review pass, logged as R2, left this note against the same test: "could have gotten 39 marks - Q10, Gold Standard for Games and Tournaments Questions." Thirty-nine against an actual score of 18 is more than double, and none of it required learning anything new - it required attempting more of what I could already solve at 75% accuracy.

That's the single highest-leverage fact in this whole post: in DILR, the ceiling on your score is usually set by how much you attempt, not by how much you know. A 35th-percentile result sitting on top of 75% accuracy is a selection problem wearing a knowledge problem's clothes.

Set selection: what I do with the first two minutes

You get roughly 40 minutes and 4-5 sets. The 17 May sectional is exactly what happens when set selection goes wrong: time went into fewer sets than it should have, and the ones attempted were solved well. The fix isn't "solve faster" - it's spending the first two minutes scanning every set before committing to any of them:

• Read only the introduction paragraph of each set, and count the variables - columns in a table, people in an arrangement, number of constraints.
• Classify each as Easy, Medium, or Avoid before solving anything. Easy: a familiar structure, few variables, deterministic constraints. Avoid: open-ended "at least / at most" language with many unknowns.
• Solve Easy sets fully rather than partially - a half-solved set wastes the time spent reading its constraints without banking marks.
• Only then look at Medium sets, with whatever time is left.

Two minutes feels like a lot to spend before writing anything down. It is not, against the alternative of spending 15 minutes inside a set you should have skipped.

Why the range is 2 to 30, not just "sometimes hard"

The 2 came on 30 July, and it wasn't isolated - VARC was 9 and QUANT was 29 that same mock, so all three sections dropped together, which points at a bad day rather than a DILR-specific gap. Two days later the same three sections read 33, 29, 32. The lesson isn't about DILR technique at all: one mock, taken alone, tells you very little. What told me something was the run of scores around it - 96 the mock before the crash, 94 the mock after. For the fuller argument on why a single data point misleads and a trend doesn't, see how I analyze a mock end to end.

Here's the full run of DILR raw scores behind that 2-to-30 range, in date order:

Two things stand out beyond the 2-to-30 extremes. First, the score never sat still for long - four or five consecutive mocks in the high teens or low twenties is the closest thing to a plateau in the whole set, and even that broke both up and down. Second, the most recent five mocks (3 Aug through 24 Aug) sit in a tighter 20-28 band than the May-June stretch did, which is the shape an improving-but-still-noisy section actually takes: the floor rises before the ceiling does, and the ceiling doesn't stop moving.

The two 30s in the set (16 May and 20 June) sit at opposite ends of the percentile scale for reasons that have nothing to do with skill - a different cohort sat each paper. That's a section-specific instance of a pattern covered with the full percentile table in CAT percentile vs score; worth reading if a percentile drop is making you second-guess a score that didn't actually get worse.

Games and Tournaments: the set type behind my own missed marks

The 39-mark note above named Games and Tournaments specifically, so it's worth being exact about the one formula that set type rewards: with n teams in a single round-robin, the number of games played is n(n-1)/2. That comes from each team playing every other team exactly once - n(n-1) ordered pairs, halved because "A plays B" and "B plays A" are the same game. Knowing this cold means you can check a partially-solved bracket against the total game count instead of re-deriving it under time pressure, which is exactly the kind of arithmetic tax that eats into attempt volume.

Most Games and Tournaments sets give you partial results and ask you to work backward from point or win totals. Write every deduced result on paper the moment you have it - cognitive load, not the puzzle's actual difficulty, is what causes errors here.

The DI half: where the marks are the easiest to bank

Since 2023, IIM Ahmedabad's mixed-set format means a single DILR set can require reading a table, extracting a value, and then applying a logical constraint - there's no longer a clean line between "DI" and "LR" sets. But the DI-flavoured extraction work inside a mixed set is still the most reliably bankable part of it, because it has no ambiguity once you've read the number correctly:

• Tables: practice reading a row/column intersection in one eye movement rather than tracing across with a finger - the habit that costs the most time is re-finding a cell you already looked at once.
• Bar charts and line graphs: write down intermediate values as you extract them. Holding two numbers in your head while reaching for a third is where transcription errors creep in, and a transcription error here reads as a careless error in the Step 1 classification from the mock analysis framework, not a logic error.
• Caselets (data embedded in a paragraph): extract every number into a clean table before answering anything. Skipping this step is the single most common reason a caselet's second and third questions go wrong even when the first one was right - a misread value early on quietly propagates.

None of this is about being cleverer. It's about not losing marks to a transcription slip on a set you'd already correctly solved in your head.

A 12-week structure, if you're building from scratch

Randomised practice is inefficient. A staged progression works better:

• Weeks 1-4: two easy sets a day, untimed. The goal is a clean, repeatable table-construction process, not speed.
• Weeks 5-8: introduce a 12-minute timer per set. After each one, classify every error as a constraint misread, a deduction error, or a calculation slip - each needs a different fix.
• Weeks 9-12: full 40-minute section simulations, 4-5 sets, with the two-minute scan run at the start every time. Review the selection decision itself afterward, not just the accuracy.

Two sets a day for 84 days is 168 sets of exposure - more than most people get by cramming in the final weeks, and spread out enough that each review actually sticks.

What to log, and why the R2 note mattered more than the R1

The 39-versus-18 gap wasn't visible on the day of the sectional. It surfaced on a second pass, days later, once the emotional attachment to the original score had faded. Karma Yogi's revision rounds exist for exactly this: log the analysis session, tag it R1 through R7 against the test, and the mock card shows an overdue reminder if a round is due. Nothing here required new content - it required coming back to the same test with fresh eyes on a schedule, instead of trusting the first read as final.

Ready to start tracking your own DILR sectionals and mocks the same way? See how Karma Yogi tracks CAT preparation, or start free.