Error Checking Test: Speed, Accuracy and Preparation

Error checking is the aptitude test where being clever helps least of all. Any single item is trivial: does this string match that string, or does it not. The difficulty lives entirely in doing that judgement two hundred times in a row, at speed, without your attention sliding away around item forty. Candidates who score strongly on numerical and verbal reasoning often land in the middle here, because what is measured is sustained attention under time pressure, not reasoning.

Two mechanical facts make the format harsher than it looks. Error checking is very commonly negatively marked, so a wrong answer costs you a mark rather than nothing. And it is frequently delivered as an indefinite bank against a clock, holding far more items than anybody finishes, so there is no final question to reach. Between them, those two facts wreck the instinct every candidate arrives with, which is to go as fast as humanly possible.

These ten questions are in Cappfinity format. The Cappfinity Style Practice Pack runs these question types at full length — 5 timed tests, 315 questions, a worked explanation on every one.

The formats inside this test type

The label covers at least five distinct question formats. They look alike but differ sharply in what they demand and in what each item is worth in time.

Paired-record comparison

Two versions of the same data string sit side by side and you decide whether they are identical. The response is usually binary, with a "no error" option standing in for "identical". Some stack four or five pairs into one item and ask how many contain a mismatch.

Mechanically this is the lightest format: you never have to say where the fault is, so the moment you find any mismatch you can stop and answer. The characteristic faults are the transposition (4471 against 4417) and the near-identical glyph, 0 against O, 1 against l, 5 against S, rn against m. Reading a string as a word invites your eye to normalise those away, so chunk both into groups of three or four characters and compare chunk against chunk.

Multi-field record checking against a source

A source record carrying several fields, typically a name, a date, a reference number and an amount, is set against a transcribed copy, and you report which field differs. The response is a field label, not a yes or no.

That single change makes it a different cognitive job. You cannot answer from a general impression; you must localise the fault and map it to a label. Nor can you always stop at the first mismatch, since some versions allow two fields to differ. Each field fails in its own way too: dates by component swap, references by transposition, amounts by decimal place. The per-item budget is several times that of a binary pair.

Text proofreading with a count response

A short passage seeded with spelling, punctuation and grammar faults, and you report how many it contains. There is no source text, which is the mechanical difference that matters: the fault is defined by a rule in your head rather than by a mismatch on the page. Count responses are unforgiving, since being one out scores the same as being ten out, and reading for meaning suppresses surface faults. Sweep against the grain instead, once for sentence-final punctuation, once for agreement, once for homophones.

Coded-string verification

Here a rule is stated instead of a source given. A reference might have to open with two letters matching a branch code, carry the date as four digits in DDMM order, and close with a check letter fixed by the sum of those digits. You mark each string valid or invalid. No comparison happens at all: this is rule application with a small calculation inside it, so working memory load is high, the per-item budget is the longest of the five, and the failure mode shifts from attention slips to arithmetic slips and dropped rule clauses.

Table and form transcription checking

A table or completed form on one side, a data-entry rendering on the other, and you check whether values landed in the right place. The signature fault is not a wrong character but a right value in the wrong row, since one row shift makes every later value wrong while each looks plausible alone. That rewards checking structure before content: row labels, column headers and totals first, values second.

Scoring and timing, and what they mean for strategy

Most error checking tests use formula scoring: one mark for a correct answer, a deduction for a wrong answer, and nothing at all for an item left blank. A minority use raw scoring, where wrong and blank are both simply zero. The two models pull in opposite directions, so establish which one you face before you start.

Under formula scoring the rule is blunt. On a paper that takes marks off for errors but not for blanks, a rushed guess in the closing seconds actively lowers the score you had. There is a break-even point you can work out in advance: if a wrong answer costs d marks and a correct one gains 1, attempting an item only pays when your accuracy on that kind of item beats d divided by (1 + d). Full-mark deductions therefore need accuracy above one half; quarter-mark deductions need only one fifth. Your buzzer guess is also nowhere near random, because hurried candidates cluster on "no error" and on whichever option they used last, which pushes real guessing accuracy below chance.

Timing models vary just as much. An indefinite bank against a clock has no finish line, so raw speed buys nothing on its own; only accuracy per attempted item moves the score. A fixed question count with time left over means the opposite, that a careful second pass is free. Per-section timers let you bank saved time nowhere, so pacing has to be per section. Per-question timers remove skipping. Where difficulty escalates as you progress, an early comfortable rhythm is no guide to later items.

A worked example

Source record: Marguerite Ashdown-Reilly, 03/11/1987, BR-4471-092K, £1,340.75.

Entered record: Marguerite Ashdown-Reilly, 03/11/1987, BR-4417-092K, £1,340.75.

Which field differs? A Name. B Date of birth. C Reference. D Amount. E No error.

The answer is C. Work the fields in a fixed order and never vary it. Name: compare in chunks, "Margue", "rite", "Ashdown", hyphen, "Reilly"; identical. Date: compare as three separate components, 03 against 03, 11 against 11, 1987 against 1987; identical. Reference: "BR-" matches, then 4471 against 4417, the middle two digits transposed. Confirm the tail, 092K against 092K. Amount: identical to the penny.

The wrong options are each built from a specific misapplied rule. A is chosen by candidates comparing overall shape rather than characters, who flag the visually heaviest field on the assumption that long strings hide faults. B comes from silently reformatting 03/11/1987 into month-first order on one side and day-first on the other, a conversion nobody asked for. D comes from comparing amounts by value, stripping the separator and reading the digits as a quantity, which loses the decimal position. E is the most instructive: it catches anyone whose scan stops once a field matches for its first few characters. "BR-44" agrees in both records, so a prefix check clears the reference and returns "no error".

How to prepare for this type specifically

Settle the scoring model and the bank size before you sit down, because they decide whether you answer the last three items or leave them blank. Then compute your break-even accuracy from the deduction and practise finding the pace at which you sit just above it, rather than practising raw speed, which is what almost everyone trains by default and what costs marks.

Train the confusion sets directly: 0 and O, 1 and l and I, 5 and S, 2 and Z, rn and m, cl and d, plus digit transpositions, the most common planted fault and the one your eye most willingly smooths over. Practise reading reference numbers right to left, which breaks the habit of reading strings for meaning.

Fix a field order for multi-field items and keep it identical every time, so losing your place becomes impossible. Log your faults by kind rather than by count: misses and false alarms need opposite corrections, since one says slow down and the other says trust the first pass. Work in blocks longer than the real section so the attention decline happens while you can still study it. Finally, check whether the format lets you skip and return, because on an indefinite bank it usually does not, which makes every hard item a decision about abandoning it.

Frequently asked questions

TestSolve is independent and not affiliated with any test provider or employer named on this page. All product names and trademarks belong to their respective owners.