How CAT DILR Tests Working Memory and Not Just Logic

DILR

Two candidates have the same reasoning ability. One finishes a set in six minutes, the other is still going at fourteen and makes an error they cannot locate. The difference is rarely that one of them is a better reasoner. It is usually how much each was holding in their head.

DILR loads working memory harder than anything else in the paper, and working memory is a small, fragile resource that degrades under exactly the conditions the exam creates. The useful part is that this is almost entirely fixable, because the fix is not to have a better memory but to stop relying on it. This piece explains where the load comes from and what to do about it.

Where the Load Actually Comes From

Name the sources and the fix becomes obvious, because each one has a different remedy.

Any one of those is manageable. Running all four simultaneously is what produces the sensation of a set being hard when its logic is not, and it is why the same candidate handles the same set comfortably when untimed.

Branching Multiplies Everything

The second source deserves separate treatment because it does not add load, it multiplies it.

When constraints force placements early, you carry one structure. When they do not, you carry several, and every subsequent condition has to be tested against each. Two open configurations is manageable. Four is usually fatal inside a hard 40 minute sectional limit, not because the reasoning is harder but because the bookkeeping is.

This is also why branching sets produce errors rather than just slowness. A misplacement in one branch propagates silently, and you discover it only when nothing fits, by which point you cannot tell where it entered.

The Fix Is to Stop Holding Things

Anything written down stops costing. It cannot decay, cannot be silently misremembered, and is free to consult. That single principle covers most of the remedy.

Load	Where it should live	What it looks like
Conditions not yet applicable	A short list to one side	Numbered, with a tick when used
Negative conditions	Marked in the structure	Crossed cells, not remembered exclusions
Open possibilities	Separate small grids	One per branch, labelled
Intermediate deductions	Placed immediately	In the grid, not in your head
The question being answered	Circled or underlined	Visible while you work

Negative Conditions Are the Worst Offenders

One category is consistently left in memory and it is the one that costs most.

A condition saying something is not in a particular position, or that two things are not adjacent, narrows as effectively as a positive placement. But it reads as weaker, so candidates note it and carry it rather than placing it, and then apply it at half strength when it matters.

Cross the excluded cells. A grid with exclusions marked behaves completely differently from one where they live in your head, and the difference shows up as sets closing that previously seemed underdetermined.

Put Branches on Paper Too

When a set genuinely branches, the instinct is to hold the alternatives mentally because drawing them feels wasteful. It is the opposite.

Draw two small grids rather than carrying two arrangements. Testing a new condition against both then becomes a visual check rather than a memory exercise, and when one branch dies you can see it die rather than losing track of which one you were in.

If you find yourself needing four grids, that is your signal to leave. Not because four is impossible to draw, but because a set requiring four open branches is going to be expensive regardless of how well you manage the bookkeeping.

Why the Third Section Is Harder

Working memory is not a constant, and that explains a pattern candidates notice and misattribute.

It degrades across a long paper. A candidate relying on memory finds sets getting harder as the 120 minutes progress, while a candidate who externalises onto paper finds them roughly as difficult as they were. The same set is genuinely harder at minute ninety than at minute ten for the first candidate and not for the second.

This is why writing more is the cheapest protection against attention decay across a paper. It converts a resource that depletes into one that does not, which matters most in exactly the section where you can least afford a drop.

Overload Errors Do Not Announce Themselves

The most expensive property of this failure is that it is silent, and that deserves its own note.

When you run out of working memory you do not notice. There is no sensation of forgetting, and a misremembered condition feels exactly like a correctly remembered one. You continue confidently on a structure that has quietly gone wrong.

Then the whole cluster is wrong together, since questions come off one structure. Under the marking scheme, where a correct answer is plus three and an incorrect MCQ is minus one, that is several penalties arriving at once from a single invisible slip.

The protection is procedural rather than perceptual. Check each condition off as you apply it, so an unapplied condition is visible rather than assumed. And when a placement feels obvious rather than forced, write down which condition forced it; if you cannot name one, you assumed it.

Training the Habit

Externalising is a habit rather than a technique, which means it needs practice under conditions that make the old habit costly.

Practise writing more than feels necessary, including on sets that seem easy. A routine that only activates on hard sets is not a routine, and the sets where you are most likely to skip the discipline are the ones where a silent error is least expected.

In review, when a set went wrong, work out whether the failure was logical or a bookkeeping slip. Most candidates find the second far more often than they expected, and the two need completely different responses: one wants more practice with the set type, the other wants a pencil.

Treat the Rough Sheet as a Layout

Since almost all of this comes down to using paper properly, it is worth being specific about how, because most candidates use their sheet as a scribble area rather than as a workspace.

Give it a layout before you start. The structure in the middle where there is room to extend it, the numbered condition list down one side where you can tick items off, and a small corner for intermediate values. That arrangement means you always know where to look, which removes a small search cost that repeats dozens of times in a section.

Keep one set per area of the sheet and do not write across old work. Candidates running short of space start squeezing a new structure into the gaps around an abandoned one, and then misread their own marks under time pressure. Turning the page costs a second and prevents an error class that is entirely self-inflicted.

Write larger than feels necessary. Small cramped marks are harder to scan, and scanning is what you will be doing repeatedly as you test conditions against the structure. The sheet is not a record to be kept tidy, it is a surface you will read from under pressure, and legibility to your own eye at minute thirty is the only standard that matters.

Recognition Reduces the Load at Source

There is an upstream fix worth mentioning because it reduces how much there is to hold in the first place.

A candidate who recognises a set's family arrives with a default structure in mind, so the representation decision is free and the conditions have somewhere to go immediately. A candidate working it out from scratch is holding the conditions while also deciding what shape the problem is, which is two loads instead of one.

That recognition is built by working type by type before mixing, across the recurring families: arrangements, games and tournaments, distribution and selection, tables and caselets, networks, scheduling. It is the same practice that makes cold starts cheap, and it happens to cut working memory load as a side effect.

The Summary

DILR loads working memory from four directions at once: unapplied conditions, open possibilities, intermediate deductions, and the question itself. Any one is manageable and running all four is what makes a logically simple set feel hard.

Branching is the worst because it multiplies rather than adds, and it produces silent errors that propagate through a branch you cannot audit. The fix throughout is to stop holding things: draw the structure before reading conditions, cross out excluded cells rather than remembering exclusions, draw branches as separate grids, and keep the question visible.

Working memory degrades across a long paper, which is why the later sections feel harder for candidates who rely on it and not for those who externalise. And because overload errors are invisible, the protection has to be procedural: tick conditions as you apply them, and name the condition that forced any placement that felt obvious.

If your DILR errors are ones you cannot locate afterwards, that is the signature of overload rather than of weak reasoning. A CAT preparation strategy review will separate the two from your set-level data, and a personalised CAT preparation plan sequences the type-wise work that cuts the load at source.