Aptitude skill lesson
Working memory: skill lesson
Working memory is the small, fast, easily disrupted workspace you use to hold a few items while you do something to them: reversing a digit string, keeping a running total, tracking which rule is currently in force. It is not general memory and it is not intelligence; it is capacity plus the ability to update and resist interference. Assessment batteries probe it with backward span, letter-number sequencing, n-back and running-memory tasks, and game-based screens lean on it heavily. At work it shows up whenever you hold a customer's account number while navigating three screens, or carry a partial calculation while someone talks at you. Everything you practise here is device-local: no account, and nothing leaves the device unless you export it.
Published · last reviewed
Applies to Novus Learn 0.1.0
What changed, and when
- , Replaced the body of all 27 non-judgement lessons with construct-specific material: six worked examples each, carrying the actual arithmetic, the actual inference or the actual procedure, plus expanded objectives, practice tips and glossary. Every 'Related Learn topics' link now points at a real page on this site rather than a generic search. The eight workplace-judgement lessons are unchanged.
- , Rebuilt the lesson page around a sticky contents rail, per-section links, previous/next lesson navigation, and two graded checkpoints drawn from the open practice bank.
- , Repaired the aptitude integrations behind the lessons so each one links to skill-specific practice instead of the unscoped fixture engine.
- , Published one lesson for each of the 35 aptitude skill constructs: objectives, worked examples, practice tips, glossary, Learn topic links, and sources.
These 35 skill lessons are authored and revised as one set, so they share one revision history rather than 35 identical dates.
Objectives
Copy link- Reverse and re-order a presented sequence without writing it down, using chunking to keep the load inside span rather than fighting it item by item.
- Track an n-back stream and correctly reject the 1-back lure that catches most first-time candidates.
- Hold and update a rule that changes mid-item, and state which stored value is now stale.
- Decompose multi-step mental arithmetic so that only one partial result is ever in play at a time.
- Explain why acoustically similar and multi-syllable material eats span faster than the item count alone predicts.
- State honestly what working-memory practice does and does not transfer to, so preparation targets task familiarity rather than a promised capacity increase.
Checkpoint: does the idea land?
Copy linkBefore the worked examples, check that the objectives above actually landed.
Two questions from the open practice bank, answered here and scored on this device. Untimed, ungraded, and not added to your practice history, the full bank is where attempts are recorded.
Examples
Copy linkBackward span: chunk first, then reverse the chunks
Read this once and answer without looking back: 7 2 9 4 1 6 3 8, reported backwards. Item-by-item reversal fails around six digits because you are re-reading a decaying trace eight times. Chunk instead. Pair them at input: 72, 94, 16, 38 - four items, comfortably inside span. Now reverse the chunk order and the pair inside each: 38 becomes 83, 16 becomes 61, 94 becomes 49, 72 becomes 27. Output 8 3 6 1 4 9 2 7. Check it against the original read left to right and it matches. The trap is a partial reversal: candidates reverse the chunks but forget to reverse inside them and answer 3 8 1 6 9 4 7 2, which looks right because every digit is present and the first digit is plausibly near the end of the original. That answer is tempting precisely because the gist check - same digits, roughly backwards - passes. Only a position check catches it, so verify the FIRST digit you output equals the LAST digit you heard, then stop worrying and recite.
N-back: the lure sits one position away
A 2-back task on the letter stream T, L, T, K, R, K, R, R, M. You press only when the current letter matches the one two positions earlier. Walk it. Position 3 is T against position 1, T - target. Position 4 is K against position 2, L - no. Position 5 is R against position 3, T - no. Position 6 is K against position 4, K - target. Position 7 is R against position 5, R - target. Position 8 is R against position 6, K - no, and this is the whole point of the item: position 8 matches position 7, an immediate repeat that feels overwhelmingly like a hit. It is a 1-back lure. Position 9 is M against position 7, R - no. Three targets: positions 3, 6 and 7. Familiarity is the wrong signal here, because the recent repeat is the most familiar thing in the stream; the task requires position, not familiarity. The practical defence is to run a strict two-slot window and physically discard the older slot as each new letter arrives, rather than letting a queue of five or six letters build up.
Letter-number sequencing: two sorts, one pass
Presented: R 4 B 9 K 1 M 7. The instruction is to report the numbers in ascending order first, then the letters in alphabetical order. Splitting at input is what makes this tractable: as each character arrives, drop it into one of two mental piles instead of storing the presented order at all. Numbers pile: 4, 9, 1, 7, sorted to 1 4 7 9. Letters pile: R, B, K, M, sorted to B K M R. Answer: 1 4 7 9 B K M R. Candidates who store all eight characters in the order given then try to sort afterwards are holding eight items and running two sorts on a decaying trace, which is roughly double the load for no benefit. The common wrong answer preserves presented order inside each pile - 4 9 1 7 B K M R - because the sort step was dropped under load while the split step survived. If you can only do one thing well under pressure, sort at input and let presentation order go.
Sequential arithmetic: never hold two partial results
Instructions given aloud, one line at a time: start at 17, multiply by 4, subtract 12, divide by 8, add 19. Run it strictly in order. 17 x 4 = 68. 68 - 12 = 56. 56 / 8 = 7. 7 + 19 = 26. The answer is 26. Two traps live here. First, order of operations: this is a sequence of instructions, not an algebraic expression, so nobody does the division before the subtraction, and a candidate who silently applies precedence rules gets a different and confidently wrong number. Second, load: the discipline is that after each line, only ONE number exists. The moment you are carrying both 68 and 12 as separate live values you have doubled the load for no reason. The same discipline turns awkward multiplications into easy ones by rounding to a friendly anchor and correcting once: 68 x 7 becomes 70 x 7 = 490, minus 2 x 7 = 14, giving 476. One anchor, one correction, one live value at a time.
Rule updating: the stale value is what gets marked wrong
The standing rule is to sort a support queue by priority, High before Low. An exception is added: if the queue name begins with a vowel, sort by age instead, oldest first. The queue is Onboarding, holding ticket A (High, 2 days old), ticket B (Low, 9 days) and ticket C (High, 5 days). Onboarding begins with O, so the exception is live and priority is now irrelevant. Sorted by age, oldest first: B, C, A. The tempting answer is A, C, B - the two High tickets first, oldest of them leading - because the priority labels are the most visually salient thing on the screen and the exception was a single clause read several seconds ago. This is what working-memory items actually test: not whether you can sort, but whether the currently active rule successfully overwrote the default. The habit that fixes it is to say the live rule out loud, in one short phrase, immediately before you act - 'vowel, so age, oldest first' - which forces a retrieval of the rule rather than a retrieval of the habit.
Why some lists are heavier than others of the same length
Span is not measured in items alone. Hold B, C, D, G, P, T, V - seven letters that all rhyme in English. Now hold F, K, L, Q, R, Y, W - also seven letters, but acoustically distinct. Almost everyone recalls the second set more accurately, and the errors on the first set are confusions within the rhyming set rather than lost items. That is the phonological similarity effect, reported by Conrad and Hull in 1964, and it tells you the verbal store holds something sound-like. Length matters the same way: five short country names - Chad, Burma, Greece, Cuba, Malta - are easier to hold than five long ones such as Nicaragua, Afghanistan, Venezuela, Madagascar and Yugoslavia, an effect Baddeley, Thomson and Buchanan reported in 1975 and tied to how much can be rehearsed in about two seconds. The practical consequence for a test: when material is confusable, stop rehearsing it as sound and re-encode it visually or spatially - imagine the digits on a keypad, or the letters at clock positions - because that moves the load to a different store instead of stacking it in the one that is already saturated.
Checkpoint: can you apply it?
Copy linkNow apply it. These come from a later section of the bank, so they are not more of the same.
Two questions from the open practice bank, answered here and scored on this device. Untimed, ungraded, and not added to your practice history, the full bank is where attempts are recorded.
Practice tips
Copy link- Chunk at input, not at output. Pairing digits or grouping letters as they arrive costs nothing and turns an eight-item load into a four-item one; trying to chunk a trace you are already struggling to hold does not work.
- Train updating separately from holding. Backward span trains holding; n-back and running-memory tasks train the discard step, and it is the discard step that fails first under time pressure.
- Deliberately practise with a distractor running - count backwards by threes, or say 'the, the, the' while you rehearse. That is articulatory suppression, and it is a fair simulation of an assessment centre where an invigilator is talking.
- When a rule can change mid-item, re-state the live rule in three or four words immediately before answering. Most lost marks on rule-updating items are the previous rule applied fluently, not confusion about what to do.
- Do not chase a promised capacity increase. Meta-analytic reviews of working-memory training, including Melby-Lervag and Hulme (2013), have found reliable improvement on the trained task and its close relatives but little evidence of transfer to general reasoning - so practise for format familiarity and lower anxiety, which are real gains, and treat anything else as a bonus.
- Run untimed until the strategy is automatic, then add the clock. Attempts are stored on this device only, so an ugly first pass through n-back costs you nothing.
Glossary
Copy link- Phonological loop
- The component of Baddeley and Hitch's 1974 model that holds speech-like material for a couple of seconds and refreshes it by silent rehearsal. Its limits produce the phonological similarity and word-length effects.
- Visuospatial sketchpad
- The parallel store for visual and spatial material - shapes, layouts, routes. Because it is separate from the verbal store, re-encoding a confusable list as a spatial image genuinely offloads it.
- Central executive
- The control component that allocates attention, switches between tasks and keeps the currently active rule in force. Rule-updating items load this far more than they load storage.
- Chunk
- A group of items bound into a single unit by meaning or familiarity. Capacity is counted in chunks, not raw items, which is why 1-9-4-5 costs one slot to someone who reads it as a year.
- Span
- The longest sequence reproduced reliably. Forward span measures storage; backward and re-ordered span add manipulation and are consistently shorter for the same person.
- N-back
- A continuous task where you respond when the current item matches the one n positions earlier. It stresses updating and discarding rather than storage, and its characteristic error is the immediate-repeat lure.
- Articulatory suppression
- Speaking irrelevant sounds while trying to hold verbal material. It blocks rehearsal and sharply cuts verbal span, which is why noisy rooms and talkative invigilators are not a trivial complaint.
- Proactive interference
- Older material intruding on new material of the same kind. It is why the fifth practice trial of a digit-span block is harder than the first, and why varying the material between blocks helps.
Sources
Copy link- Worked sequences, n-back streams and rule-updating items written for Novus Learn. Every string and number above is original; no published test item is reproduced.
- Model terminology follows the public account of Baddeley and Hitch's multicomponent working-memory model (1974) and the capacity debate between Miller (1956) and Cowan (2001). Concepts only.
- The phonological similarity and word-length effects are attributed to Conrad and Hull (1964) and Baddeley, Thomson and Buchanan (1975); the transfer caveat reflects the meta-analytic review by Melby-Lervag and Hulme (2013). No effect sizes are quoted.
- Novus Learn aptitude construct registry (catalog seed) for construct scope and suite mapping.
- Public educational framing only - not affiliated with any official exam board, publisher or employer, and not a clinical or diagnostic measure.
Your notebook for this lesson
Saved sections and notes are kept in this browser, on this device. No account is required, and nothing is sent to us unless you create one and switch on backup. They travel with your backup file from My Learning, and they are lost if you clear this site’s data.
Saved sections
None yet. Use Save beside any section heading above to keep it here.
4000 characters left
Your own words, so you can find this again from any other lesson. Up to 12.