Reconstruction: Retrieval Beyond Recall

Reconstruction is a test. It takes pieces of knowledge the learner already holds and asks for a connection between them that nobody taught. Then it asks for a yes or a no on whether that connection exists.

Memory rebuilds an answer on every retrieval. It rebuilds from nodes, the pieces, and the relationships between those pieces. It does not pull out a stored copy. Because retrieval already rebuilds, the useful demand is a new valid relationship. The taught relationship is a weaker demand. Ruling a link impossible counts as an answer. The relationship asked for is one that has never existed yet.

Core takeaways

• Reconstruction takes pieces already known and asks for a connection nobody taught, then a judgment on whether that connection holds.
• Memory rebuilds on every retrieval from pieces and the relationships between them. A demand for a new relationship tests the operation memory already performs.
• Deciding two pieces cannot connect counts as much as connecting them. Either verdict needs each piece known well enough that its edges show, so a feeling of familiarity is a false cue.
• Recall can pass while use fails. The first failed attempt at a new join shows the knowledge is shallower than it felt.
• Reconstruction is a harder ask than recall. No finding shows it gives a larger memory gain than recall, so recall stays in the schedule and gets more weight later.
• Practising new shapes does not guarantee a pass on unfamiliar exam questions. Transfer to a new form needs elaborated retrieval, congruency, or high initial performance.

Why the demand matches how memory works

The reconstructive-memory literature says remembering is rebuilding. That literature does not prescribe this method. The method is an inference from that finding: if remembering is rebuilding, then ask for a rebuild nobody handed over. A method that demands a fresh shape tests the operation memory already performs.

Three ways to ask for a new shape

The rebuild nobody handed over has a few shapes a learner can run.

• Join one idea to a pair of ideas it was never paired with.
• Fold two separate explanations into a single account.
• Drop an equation into a setting that usually hosts a neighbouring equation.

Each attempt either stands or falls.

Deciding two pieces cannot connect

Deciding two pieces cannot connect is as useful as deciding they can. That verdict needs each piece known well enough that its edges are visible. A vague sense of having seen the material cannot show where its edges are. So the same exercise checks mastery of every piece it touches. A feeling of familiarity is a false cue for mastery.

How this compares with recall

The demand asks more than matching a prompt to something seen before. Matching asks almost no mastery, so it is a weak mastery test. A weak test still has some use. Active recall asks memory to rebuild something already known to be buildable. Reconstruction asks for more than that. No finding shows reconstruction gives a larger memory gain than recall. So recall stays.

Forming and judging new links works almost only at the top of the mastery ladder, the levels named Evaluate and Create. The lower levels get exercised at the same time, because the top level cannot be reached without them. That is a claim about what the exercise demands. No classroom experiment on that taxonomy of mastery levels tested it. The node-and-relationship model of memory is described in Memory Handling. The top of the mastery ladder is described in Knowledge Mastery From Recognition to Usable Knowledge.

What comes out of the exercise is new: guesses, conjectures, another way to explain the same idea, another way to apply it. Making those from pieces already in hand is itself a high mark of mastery. When someone describes the exercise as a higher gear of thinking, the checkable version of that claim is difficulty the learner can still meet. It is not a brain state.

What the test shows about unfamiliar exam questions

An exam curveball is a question that supplies pieces already known and demands a shape never practised. If revision only ever rebuilt the shapes that were taught, an unfamiliar arrangement looks impossible even when every piece is known. A learner who has practised this demand has already tried the shapes those pieces can take. A configuration never seen before is then one more arrangement. That is a reason the practice may help. It is not a promise that exams become easy. A new shape only transfers under conditions: elaborated retrieval, congruency, or high initial performance. Practising the demand does not guarantee a pass on curveball questions. A failed new connection is information about a gap. It does not show the method is broken.

Knowledge a person owns can be rearranged

Knowledge a person knows deeply can be put into new arrangements at will. A hobby known deeply lets a person put two favourite characters into a scene that was never written, and the person can say what happens without searching memory. Practice of this demand builds the same freedom on study material.

Recall can pass while use fails

A relationship can come back correctly on a recall test and still fail when asked to fit a new arrangement. Ordinary retrieval never asks for that new arrangement, so the weak point never shows. The first failed attempt at a brand-new join is the moment the knowledge proves shallower than it felt.

Past papers and reconstructive prompts

Past exam papers work because they test recall and application together. That is why they help even when used alone. Reconstructive prompts test the same territory on purpose. The learner attempts new connections around each concept, and the weak nodes show up. The learner does not have to work through every question type in the bank. That saving of time is an inference, not a measured finding. The gaps found this way feed the Test–Target–Teach loop in Revision.

How much recall to keep

Keep recall in the schedule. Each reconstruction item costs more minutes than a recall item, so a schedule of only reconstruction is hard to keep going. A schedule of only reconstruction is also not the aim. Working splits run from mostly reconstruction with a little recall down to an even split. The split is set by what the learner can manage. Those splits are defaults for manageability, not findings. Weight the early phase toward reconstruction while the structure of the topic is still forming. Shift toward recall later for detailed retrieval. That order is a default, not a tested rule.

Where this demand sits among the methods

Pieces the learner owns combine at will. A failed new connection is a gap that recall never posed. Spaced Interleaved Retrieval lists a menu of methods. The reconstructive end of that menu applies this demand.

How to practice this

1. Pick two ideas you know well that were never taught together. Try to join them. Notice whether the join stands or falls.
2. Take two separate explanations of one topic. Fold them into a single account. Notice where the two explanations do not fit together.
3. Take an equation and put it into a setting that usually hosts a neighbouring equation. Notice whether it still works there.
4. When two pieces will not connect, say so and give the reason. Notice whether you can see the edges of each piece. If you cannot, you know that piece less well than it felt.
5. Start with mostly reconstruction items and a little recall while the topic structure is still forming. Shift toward recall later for detailed retrieval. Notice how many more minutes each reconstruction item costs than a recall item.
6. Write down every failed new connection. Each one marks a gap that recall never showed. Feed those gaps into the Test–Target–Teach loop in Revision.

Related pages

• Spaced Interleaved Retrieval: the method menu whose reconstructive end is this demand.
• Interleaving for Complex Problem Solving: the applied extension, reconstruction run on problem solving and knowledge work.
• Revision: the Test–Target–Teach loop the gaps feed.
• Memory Handling: the node-and-relationship model of memory.
• Knowledge Mastery From Recognition to Usable Knowledge: the Evaluate and Create end of the mastery ladder.

Open questions

• Whether reconstructive prompts find gaps faster than an equivalent time on varied recall.
• Whether front-loading reconstruction then shifting to recall outperforms a mixed schedule.

Sources

• Bartlett, F. C. (1932). Remembering: A Study in Experimental and Social Psychology. Cambridge University Press. Remembering is reconstructive. It is not a stored copy pulled out intact.
• Rowland, C. A. (2014). The effect of testing versus restudy on retention: A meta-analytic review of the testing effect. Psychological Bulletin. Recall tests outperform recognition tests as practice. The review did not test reconstruction as a third condition.
• Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory. Generating a new structure is itself encoding.
• Pan, S. C., & Rickard, T. C. (2018). Transfer of test-enhanced learning: Meta-analytic review and synthesis. Psychological Bulletin. Transfer to a new form often fails without elaborated retrieval, response congruency, or high initial test performance.
• Kornell, N., & Bjork, R. A. (2008). Optimising self-regulated study: The benefits, and costs, of dropping flashcards. Memory, 16(2), 125–136. A feeling of knowing is a false cue for mastery.