Best-attempt Encoding

Best-attempt encoding is the choice a learner makes when first organizing a topic: encode each item narrowly, or encode it more widely. Narrow encoding holds one fact for one question type. Wider encoding holds the fact together with an application, a connection to other material, and the reason the fact holds. Best-attempt encoding is a stance inside Deep Processing.

A learner can start retrieval practice on material they only half understand. The half-understanding does not need to be fixed first. The learner builds the best organized structure of the topic they can currently make, then checks one thing: whether that structure can be tested. A structure that can be tested goes into retrieval practice at once. The practice shows where the structure is wrong, and the learner corrects it there.

Core takeaways

• Narrow encoding holds one fact for one question type. Wider encoding also holds an application, a connection, and the reason, and it costs more time up front.
• Two questions decide which items get wider encoding: whether the item attaches to a theme already known to matter, and whether it keeps coming up across question types. Asking both takes about twenty seconds an item.
• Testing shows a wrong sorting decision in both directions, so the decision does not have to be right.
• Five moves widen an item: a generated application, two or three connections, the mechanism question, a predicted variation, and a check of the chunk boundary.
• The target structure is one level above the learner’s current encoding level, not the ideal structure. Encoding skill takes months to years to improve.
• The structure is expected to be revised after retrieval. A gap found early costs much less to fix than a gap found after a large map has been built.

Narrow encoding and wider encoding

Every item can be encoded in two ways. The difference between them is how deeply the item is processed. With narrow encoding, the learner stores one fact for one question type. The item comes back when the question is asked the way the item was learned. It does not come back for adjacent questions, applications, or variations. Narrow encoding is fast. Wider encoding takes more time up front. The learner applies the item, integrates it with other material, and works with it across contexts. The item then comes back for the question it was learned for and for neighbouring questions on the same topic.

If wider encoding took the same time as narrow encoding, every item would get wider encoding, and there would be no decision to make. The extra time is the only reason the learner sorts items at all. This system’s own feel-test for the two widths is written as 1:1 against 1:1.2. That notation is not a hit rate and not a measured coverage number.

Wider encoding also gives later updates a place in the structure: a new regulation, a new finding, a new example. A narrowly encoded item has no place for the update, so the update becomes another isolated card.

The two filters

Two quick questions decide which items get the wider encoding. In this system’s own working model, asking both questions takes about twenty seconds an item. The decision itself takes little time. The encoding that follows the decision takes most of the time.

The first filter is conceptual relevance. It asks whether the item attaches to something the learner already knows to matter. A small change to smoking legislation is, on its own, one fact. It attaches to a large theme the learner already knows to matter, so it gets the wider encoding. A change to a speed limit appears to attach to nothing, so it gets narrow encoding.

The second filter is repeated testing relevance. It asks whether the item keeps coming up across question types and contexts. The speed-limit fact stays narrowly encoded until retrieval practice on city planning repeatedly calls for road incidents. The fact is then a usable example in several answers. The second filter shows a relevance that the first filter did not show. The two filters are two parts of one judgment.

When both filters are low, the learner makes a card for the item and moves on. When either filter is high, the learner spends the time on wider encoding. Flashcards are a class of tool, not a product. A card can still be made after the wider encoding, as a reminder. The card and the wider encoding are not alternatives. Wider encoding is a set of acts: the learner uses the fact inside a written answer, places it into an existing map as a worked example, and looks at what else it connects to.

Testing widens the margin of error on every sorting decision, in both directions. An item filed as isolated shows itself to be connected when retrieval practice keeps calling for it. An item given wider encoding that was not retained shows a gap during testing. The decision does not have to be right, because testing shows where it was wrong and the learner corrects it. The gain from wider encoding appears in the review queue, not in the encoding session. The card is answered correctly more often, its review interval grows, and it stops coming up daily. Narrowly encoded items keep coming up. Importance-Based Chunking is the same importance judgment made at the scale of a topic. The two filters are that judgment applied to a single item.

What the structure is for

A schema should make recall easier. It should not need to be memorized on its own. The test is whether the only way to remember how the material is organized is to memorize the organization itself. If the answer is yes, retention is at risk. The structure has become a second thing the learner must remember. It no longer helps the learner remember the material.

Five moves that widen an item

The first move is to generate an application. The learner uses the concept in one real example. One example is enough.

The second move is to make connections. An item with one connection is easily lost. An item with two or three connections is much harder to lose. The threshold of two or three is this system’s working number, not a measured finding.

The third move is the mechanism question: why does this work? This move most often produces a gain. A learner who knows the mechanism can rebuild the fact after forgetting it.

The fourth move is a predicted variation. The learner asks what would change if one condition were different.

The fifth move, done last, is to check the chunk boundary. The learner decides whether the item belongs inside an existing chunk or extends that chunk.

These five moves are the standard for any first encoding. They usually happen inside the refinement stage of the Bear Hunter System. That system is an encoding loop that ends with restructuring a map, not with taking notes. The five moves do not need a completed loop to run. When the restructuring is rushed, items are left shallowly encoded, and the poor retention shows up later.

Matching the structure to your current level

A good structure has three properties. It is specific enough that it fits only this topic. It is intuitive enough that remembering one element brings back the others. It is flexible enough to take an update. Before / during / after is an example that has all three. Remembering one member brings back the other two, so the learner only has to hold one part of the structure. Once a learner knows a domain well, the domain’s own groups of three work the same way.

A structure can fail by being too specific or by being too generic. A structure that is too specific uses group names so particular that each group name has to be learned on its own, and remembering one group name says nothing about the next. A structure that is too generic uses the same split for every topic. For example, a home-design topic is split into interior and exterior, then into furniture against art and materials against color, and the next topic gets the same split, and the next topic after that. A structure that fits every topic does not tell the learner where any one item sits. The clinical template does this across a whole field: history, examination, investigation, management, applied to every disease in turn. Mechanism / presentation / treatment is the same kind of generic split.

What a learner finds intuitive depends on how much of the domain they already know. The learner sets the structure to their current knowledge, not to what a specialist would find obvious. The target is one level above the learner’s current encoding level, not the ideal level. Marginal Gains gives the reason one level above the current level is the better target: small gains kept up over a long period.

Encoding skill improves slowly. In this system’s own working model, going from a low starting level to the top takes ten months of consistent effort and often years. Most learners start in the middle of the scale and need to reach a good-enough level, which takes months. Study time and content volume do not wait for that improvement. A learner who aims every session at the best possible structure gets better quality per topic and covers fewer topics.

A small gain in retention reduces the work of every later review. In the same working model, a learner who retains around 40% of what they encode and improves the method to around 50% has not saved ten percent one time. Each item that would have needed roughly ten repetitions now needs fewer. That saving applies to every item, every week, for as long as the learner is still reviewing the material. For that reason a small encoding gain saves hours a month, not minutes a session. Those figures are the system’s illustration, not a finding.

When encoding skill is low, retrieval matters more than encoding. The learner makes up for weak encoding with more retrieval, not with a better first map. Explaining the topic from memory to someone who knows nothing about it forces the learner to put the material together, whether or not they can build a structure yet. When the explanation fails, the failure shows the gap. Later, once encoding quality decides how much retrieval is needed, encoding becomes the half where improvement pays more.

If daily card review takes more than about an hour to an hour and a half and the learner still falls behind, too many items have been encoded narrowly. Items are being added faster than the learner can retain them. The remedy is to spend more time integrating the items that keep coming up, not to spend more time on the queue. That threshold is the model’s own caution, not a measurement.

Revising the structure after testing

The structure does not come finished from the source. The learner’s first reading, even when its structure is wrong, is what lets the error be seen during testing. Testing turns a first reading into something the learner can rely on. Spaced Interleaved Retrieval is the retrieval half of that process.

Fixing a gap after retrieval costs a small fraction of the original build only when the gap is found early. A structural error found after a large map has been built means every later branch has to move. The same error found in the first half hour would have changed how the map was built. The cost of a revision depends on how much has been built on top of the error. That is why the first retrieval pass should be scheduled early.

The full cycle is: best-attempt encoding, then high-volume retrieval within one week, then diagnosis of the gaps by type, then re-encoding of the higher-order gaps, then targeted retrieval for the lower-order gaps, and cards for the items that stay isolated. How to prepare for ultra high-volume exams covers the schedule, the retrieval ordering, and the coverage strategy under exam volume.

Skilled learners and new learners find the same gaps. The difference between them is timing and follow-through. The skilled learner expects gaps, tests early, and rebuilds the structure quickly. The new learner delays the test, puts off the rebuild, and then rushes it. The accuracy of the first attempt is not what separates them.

Revision is a planned step of the cycle. The learner should expect it. The time spent rearranging a structure is what produces the better version of the knowledge. No version of the process skips that step.

How to practice this

1. Build the best structure of the topic you can currently make. If it can be tested, start retrieval practice on it at once. Notice where the practice shows the structure is wrong.
2. For each new item, ask whether it attaches to a theme you already know matters. Then ask whether it keeps coming up across question types and contexts. Notice that both questions together take about twenty seconds an item.
3. When either answer is yes, use the item inside a written answer. Then ask why the item works. Notice whether you can rebuild the fact from that reason after forgetting it.
4. Within one week, run high-volume retrieval on the structure. Sort the gaps you find by type. Notice how many later branches have to move when a gap is found late.
5. Re-encode the higher-order gaps. Give the lower-order gaps targeted retrieval, with cards for whatever stays isolated. Notice that the re-encoded items come up right more often in review and stop arriving daily.
6. Time your daily card review. Notice whether it runs past about an hour to an hour and a half while you still fall behind. If it does, spend more time integrating the items that keep arriving.

Related pages

• Deep Processing: the parent dimension this encoding stance sits inside
• Importance-Based Chunking: the same importance judgment at topic scale
• Bear Hunter System: the encoding loop whose refinement stage is where these moves usually run, and which they do not require
• Marginal Gains: why one level above baseline is a better target than the ideal level
• Spaced Interleaved Retrieval: the retrieval half that converts a best attempt into something reliable

1 more related pages

• How to prepare for ultra high-volume exams: the high-volume schedule, retrieval ordering, and coverage strategy; that page owns them

Sources

• Kapur, M. (2008, 2016). Productive failure: an initial structured attempt that is incomplete, then instruction, outperforms direct instruction on transfer.
• Tulving, E., & Thomson, D. M. (1973). Encoding specificity: a cue matches the original encoding, so narrow encoding predicts poor transfer.
• Barnett, S. M., & Ceci, S. J. (2002). Near transfer is common; far transfer is not automatic.
• Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing. Craik & Tulving (1975). Elaborative encoding.
• Pressley, M., et al. (1987). Elaborative interrogation. Chi, M. T. H., et al. (1989). Self-explanation. Dunlosky, J., et al. (2013). Moderate utility for both.
• Gentner, D., Loewenstein, J., & Thompson, L. (2003). Analogical encoding.
• Roediger, H. L., & Karpicke, J. D. (2006). The testing effect. Rowland, C. A. (2014). Psychological Bulletin meta-analysis.
• Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). Expertise reversal: structures that help novices can hurt experts, and vice versa.