Schema
Schema
Part of Deep Processing.
A schema is a network of concepts and routes: each piece connects to others and can be reached from more than one entry. Facts sitting outside that network stay isolated. Facts inside it can be reached, reasoned through, and used on a new case.
The network is built on purpose, in three passes. The first is a rough sketch of how the names in a topic might connect. The second grows that sketch only by attaching new pieces to it. The third stops input often and tidies what is there. Construction, assimilation, and reorganization are this system’s names for those three jobs: a draft of the network, an add-and-connect pass, and a cleanup that takes no new input. Every new piece gets one test on the way in: whether it solves a problem or fits a pattern. What sits in the network is kept, and what does not is ordinary forgetting. The work sits in Deep Processing, encoding that builds relations rather than copies.
Core takeaways
- A fact by itself answers only the question that named it. A fact wired to its neighbours can be reached from a different question, used to explain a third idea, and applied to a case the material never showed.
- The slow step in learning is joining each new piece to the pieces already held, not reading. Going back over pages and unstructured repetition activate connections that already exist and build none.
- The network is formed in three passes: a first draft from a single keyword list and what is already known, an add-and-connect pass in which each piece has to say how it relates, and a cleanup with no new input, run about every ten to fifteen minutes.
- A new piece stays when it is attached in two directions: to something already known, and to another new piece. Memory and understanding are symptoms of that structure, not of intending to remember.
- Looking over a finished map tests familiarity. Rebuilding the map from memory with the map closed, then comparing, is formation. Gaps that surprise during that rebuild were never connected.
- When nothing will attach, the material is too detailed for the current level. A surface pass comes first, and the piece comes back after.
What a schema is
The structure knowledge sits in decides what can be done with it. A fact by itself answers only the question that named it. A fact wired to its neighbours can be reached from a different question, used to explain a third idea, and applied to a case the material never showed. That is the difference Knowledge Mastery: From Recognition to Usable Knowledge draws: recognized is not the same as usable.
When the network is strong, a question that touches any node activates the surrounding structure. The path does not have to start at the official entrance. It can start from a symptom, a counter-example, an exception, or a neighbouring topic, and still arrive. When the network is thin, the same few entry points work and the others do not. A question asked from an entry that does not work looks like forgotten material. Often the material was never connected.
One way to picture this is a transport network. Each concept is a station. Each relationship is a route. A dense network reaches almost any destination from almost anywhere. A thin network keeps offering the same few stations, and fails when those are not the ones the question used. The failure is not always a blank. Sometimes the material is there, and there is no route to it from the question that was asked.
Two qualities of the network can be told apart. Retention is how much of it stays accessible over time: whether last week’s stations are still on the map. Mastery is how well-connected, accurate, and usable the map is: whether a new question can ride the existing routes. A network can score high on one and low on the other. A well-rehearsed list of terms lasts a week and still cannot answer a sideways question. A richly connected map that was never retrieved fades, and the routes go with it.
Both qualities rise by the same process: building the network on purpose. Neither is reliably improved by going back over the same pages, by repeating items that were never joined, or by sitting through material without placing it. The pages feel familiar. Familiarity is retention of the surface, not mastery of the routes. Faced with a complex problem, a well-built schema lets the problem come apart into the concepts it is made of, with a place to start.
The conversion bottleneck
The slow step is not intake. Pages turn quickly. Lectures run at speaking speed. What takes the time is converting what just arrived into a network that can be entered later. Reading costs little. Joining each new piece to the pieces already held is the step that costs.
A piece that arrives with nowhere to sit is processed at full cognitive cost every time. It is stored poorly. It is gone in hours or days. A piece that arrives into a network already waiting is placed, joined to what it belongs with, and lasts. The extra effort goes on deciding whether the piece matters, and where, before any of the meaning is processed. That full-cost feeling is what Cognitive Load & What Mental Effort Is Trying to Cue is trying to cue.
A thin network produces a recognizable cycle. Retention is poor, so the material fades in hours or days. Relearning is required. The same process runs again: more pages, more highlights, the same isolated pieces. The same thin network comes back out. Repeating that process harder produces the same network faster. Changing the volume does not address the cause. Changing the process, forming the network rather than adding to the pile, does. The hours spent on the second and third pass through the same chapter feel like progress because more of the chapter has been seen. Having seen a page does not connect it. The feeling of progress rises while the structure stays the same.
Why more volume does not fix it
More volume does not repair the network. The conversion step does. Going back over notes and running unstructured repetition do not raise retention or mastery, however many times they run. They activate what is already connected. They do not build the connections that were never made.
A well-organised pile is still a pile. A tidy outline that cannot be entered from a new question is not a schema. The lookalike is the thing to watch for: notes that look complete, headings that look like structure, a map that can be recognized and cannot be used. The outline can be recited. A question asked from the side still gets no answer.
Retrieval practice sits next to schema building and does not replace it. Retrieval tests a network. It cannot invent the routes that were never built.
What makes a piece stay
Each new piece gets one test: whether it solves a problem or fits a pattern. Every method for attaching a piece is one of two kinds of relationship.
An analogous relationship links the new piece to something already known: an accurate, comprehensive analogy, a tie to prior learning, or a simpler label that makes the term usable. A decorative analogy does not transfer; the structure has to match. An in-context relationship links the new piece to another new piece: comparing them, grouping them, drawing how they affect each other, placing them by what they connect to. Grouping by shared features is the weak form; Importance Based Chunking is how the groups reduce load. Applying the piece to a problem immediately is the problem half of the test: retrieval plus use. Comparing and evaluating is the depth these links are for, the work Higher Order Learning names.
One kind can carry a session; both is the target. Either kind can be primed before the session starts. Practice questions ahead of time, and a Prestudy pass that finds the main ideas and how they relate, surface the points that make both kinds easier to see.
Memory and understanding are symptoms of structure, not of intending to remember. Trying to hold an isolated item by effort is the move the attachment test replaces. Cave Theory is the frame: isolated school facts slip; connected, purpose-tied ones stay. Running the test on every piece is that frame as a method.
Construction: the first draft
This system forms a schema in three passes. Skipping any one leaves the network incomplete or inaccessible. The three are a procedure, not a published research triad.
Construction is the first draft of a schema for a new topic: a rough but usable foundation, not a deep pass. The aim is a network that can take new pieces later, without filling the workbench with details that have nowhere to sit. Accuracy is not the point of that draft. A draft that exists is the point. Three moves get one.
The first move is one page of keywords, pulled from every source that will later be used, onto a single list. Headings, objectives, lecture slides, and diagrams come off slides, a textbook, and videos onto that one list. The same term in three open windows is three workbenches for one name; attention splits across the windows instead of landing on how the names relate. One list is the repair. That setup is a cousin of Prestudy, which is the frame-finding pass before an event, not a substitute for this draft. The same draft-first move, run at the scale of a whole set of dense sources rather than one topic, is Syntopical Reading - Learning from Multiple Dense Resources.
The second move is to begin the draft with what is already known, even if that knowledge is thin. Familiar ideas become the first anchors, and the new names hang off them. Even a wrong scaffold is a place to test the next piece against. A name with no anchor has nowhere to go; a name attached to something already held has a place to wait.
The third move is to guess, from those anchors, how the concepts might relate. Educated guesses are allowed and expected. Relations are guessed before they are fully understood. Accuracy is not the goal of this pass. The goal is something that can later be corrected and expanded: a wrong link that can be fixed beats an empty page that cannot. Permission to be wrong is what keeps the draft revisable.
The output is a rough map: imprecise, incomplete, revisable. Waiting for a correct map before starting is how the conversion step never starts. This pass should feel relatively light. Going deep into details too early works against the draft: isolated elements have nowhere to sit, and the session fills with pieces that cannot yet be placed. Depth arrives later, once there is a structure that can receive it.
Adding and connecting
Adding and connecting is the second pass. Each new piece is integrated into the existing structure, not accumulated beside it. The field word assimilation is often used for this step, and in this system it means adding-and-connecting: expanding an existing draft by integrating new information without creating chaos. It is not the published developmental process that shares the name. The published pair is two processes, one of which changes the structure; this pass is the narrower job of hanging a new piece on what is already drawn. When a piece will not fit, the structure is changed to take it, which is why “add and connect” is the plainer name.
Every new piece answers one required question: how does this relate to what is already on the draft? Three questions do the work: how does this connect? What does it flow from and to? Which existing group does it expand, correct, or subdivide? A piece that cannot answer is not added yet. The question is the pass, not a check after it. Without it, adding is piling under another name. Importance-Based Chunking is how those nodes get grouped once the relations have been judged.
The first additions stay simple. Approachable concepts go on first; dense or technical sections get flagged and waited for. A piece that cannot be placed yet is flagged and moved past, and returned to after the surrounding structure has developed. Trying to finish every piece before moving on stalls the pass. A technical paragraph dumped onto a thin sketch recreates the problem construction was meant to prevent: pieces with nowhere to sit, only now they look like progress because they came from a harder source.
The work proceeds in layers: approachable concepts first, then secondary ones, then detail. A foundation first, then the next layer of detail, rather than full depth immediately. Adding-and-connecting becomes easier once a rough draft exists, because each new piece has slots to try and a question to answer. That is why construction comes first. A fact that will not sit is often a frame problem, not a memory problem: the same rule as layered learning, applied to a single session. The build order lives in Layers of Learning.
The gatekeeper
Without a regular cleanup, the first two passes produce increasingly messy maps. Reorganization is the gatekeeper: it converts raw learning into knowledge that can still be used. Construction and assimilation without it eventually overload: more names, more guessed links, more flagged sections, and no simpler structure underneath. The more consistently the cleanup runs, the faster building can continue. The session that feels slower in the moment is the conversion, not a stall.
A usable approach to this work supports all three jobs: draft, add, clean. If one is missing, formation gets slow, messy, or shallow. That is a design check for this procedure, not a trial of every method that builds knowledge. When it is working, the session feels more like building a map than consuming content.
Reorganization
Reorganization stops new input and cleans, simplifies, and restructures what is already there, so the schema becomes more usable and more memorable. Nothing new enters during this pass. A new fact arriving mid-cleanup restarts the mess the pass exists to reduce. That is a law, the same law Skin owns in full.
The pass runs often. This system’s own default during dense university-level study is about every ten to fifteen minutes. “Often” is the load-bearing half; the number is a working heuristic, not a finding. Waiting until the map is already overwhelming is waiting too long: the cleanup then has to do construction’s job as well as its own.
Four verbs do the work: group related ideas, remove unnecessary elements, straighten relationships, and make the overall structure simpler to follow. Grouping puts the same-kind pieces under one heading and groups more accurately. Removing cuts what does not earn a place, which includes noise and irrelevant or premature detail. Straightening turns a guessed link into a named relation, or deletes it, and corrects misplacements. Simplifying is the test: a structure that can be followed without the material open. Four questions run the pass: can groupings be simpler? Are the arrows real relations or merely plausible? Is there irrelevant or premature detail? Does the backbone stand out? The grouping skill those verbs rely on is Chunking as a Technique.
The pass feels slower because no new content is covered. That felt slowness is the step that converts accumulation into a durable schema. Skipping it to cover more is how the pile grows and the network does not.
Formation is not review
Schema review is not schema formation. Reviewing revisits an existing network to recognize what is there. It activates what is already connected. Formation builds the connections. The two sessions can look identical from the outside, the same map, the same desk, the same hour, and they do opposite work. One confirms that the stations are still labelled. The other finds which routes were never laid.
Reviewing an intact schema tests familiarity, not usability. Retrieving from memory without the schema in view, then comparing to find what is missing or wrong, is formation under retrieval conditions. The distinction is the same one as going back over notes versus testing recall. Gaps that surprise during a closed reconstruction were never connected, however many times the map was looked at. That is why the split matters for Spaced Interleaved Retrieval: a reconstruction test is formation. A look back at the finished map is not. A session that never closes the artifact has not formed anything, however long it ran.
Usable knowledge is a network that can be entered from many points. Reviewing a finished network is not how that network got built.
When nothing attaches
If no relationship will form at all, the material is too detailed for the current level. The move is not to push harder. A surface-level foundation through Prestudy comes first, then a return. A fact held alone, or a concept understood in isolation, is forgotten faster; that is ordinary forgetting, not a nightly sweep.
The misuses of the attachment test are a decorative analogy, a push through detail that will not attach, and treating the methods as a garnish on memorization. The price is both directions, on purpose, every piece; that is the work, not an extra. The quit signal is still-isolated facts after a session, or a can’t-relate answered by trying harder. The checkable expectation: for a new piece, one problem it helps with or one pattern it fits can be named.
What stays is what passed that test on purpose, attached to what was already known and among the new pieces. Isolated pieces still slip. The work is running the test before they do.
Signs and limits
When the three passes are working, new information is easier to place, and covering a lot produces less overwhelm, because each piece arrives into a structure rather than onto a pile. Reorganization sessions feel slower in the moment but create clarity; that slowness is the conversion step, felt from the inside. Concepts can be explained in ordinary words with the relations visible. Two more signs: forgetting feels less catastrophic, because the structure remains even when a detail has gone; and related new topics come faster, because a model already built can take a neighbour.
When it is not working, studying feels like accumulating disconnected facts, and the session falls back on re-reading or passive review. Two sessions in a row that still feel like that are a quit signal.
The method belongs where relationships matter: dense university or professional content, problem-solving subjects, a lot to learn in limited time. It is the wrong tool for loosely related item lists; there is no network to draft. It is useless if input never stops. The tidy feels slower, and that slowness is the price.
The checkable tests: a new piece has a place to sit, and the relation can be said in ordinary words. A tidy outline that cannot be entered from a new question is not a schema. A review session that never closes the artifact is not formation. A question that hits any node should reach the surrounding structure; if it cannot, the network is thin in that region.
A better schema means less relearning. That direction is real. No measured claim is made that the three passes produce significantly more content in less time.
Reorganization builds the artifact. Spaced Interleaved Retrieval tests it. A reader who tidies forever and never retrieves has built a cleaner map and not checked whether it can be rebuilt. The usable picture exists because the tidy ran all along, not because more was piled on. Forgetting still happens; the structure that remains is what the next session stands on.
In the study system
Schema formation is the central outcome of the encoding loop. BHS is that loop. A short survey before the main event drafts the network: construction. Questions then sharpen the draft: what has to be true, what relates to what. Working the material adds and connects. The last pass takes no new input and reorganizes. Retrieval practice tests reconstruction. Prestudy, Aim, Shoot, and Skin are those jobs under their house names. Construction, adding and connecting, and reorganization are the same three moves as Aim, Shoot, and Skin in another vocabulary, not a fourth loop.
The external form of the schema is a mindmap. GRINDE is the standard for schema quality in final form; that standard is set on the mindmap page. Best-attempt Encoding is the same formation decision at the level of one concept: build slightly wider than the minimum so the structure is usable under unseen conditions. The margin itself is set on that page.
Skill and procedural schemas exist and look different. A coder or a language learner still needs a network. It may be a set of productions rather than a concept map: if-this-then-that, a motor pattern, a sentence frame. The method does not become a pile either way. The nodes and routes take a different shape.
How to practice this
- Before reading in depth, pull the keywords from every source you will use onto one list. Notice that your attention stops jumping between windows to find the same term.
- Start the draft from what you already know and guess how the new names relate. Notice that a wrong link can be fixed later, and an empty page cannot.
- For each new piece, say how it relates to what is already on the draft. Name one problem it helps with or one pattern it fits. A piece that cannot answer is not added yet.
- About every ten to fifteen minutes, stop taking in new material and tidy: group, remove, straighten, simplify. Notice that the session feels slower and that the next pieces are easier to place.
- Close the map, rebuild it from memory, then compare. Notice which gaps surprise you; those were never connected.
- When nothing will attach, do a surface pass first and come back later. Notice that pushing harder does not attach the piece.
Related pages
- Deep Processing: the parent dimension and hub, encoding that builds relations, not copies.
- Bear Hunter System: the encoding loop whose outcome is a schema; Aim, Shoot, and Skin are the same three moves in another vocabulary, so the three passes are not a fourth loop.
- Prestudy: construction’s cousin, the setup pass before an event; the short pass that finds the main ideas and how they relate, the prime that surfaces relevance points, and the repair when nothing will attach.
- Mindmaps: the external form of the schema.
- Best-attempt Encoding: the width decision at the level of one concept.
Open questions
- Do skill schemas look different enough to change the method, or is a network of productions still built by the same three passes?
- When is a schema detailed enough to generate applications the material never showed?
- How the useful cleanup cadence actually shifts by subject, once “often” is already the rule.
Sources
- Bartlett, F. C. (1932). Remembering: A Study in Experimental and Social Psychology. Remembering is reconstruction from an organized mass, not a trace replay.
- Piaget, J. Assimilation and accommodation. Field backdrop for adding new material into a structure, and for the structure changing, not the three-pass procedure above.
- Sweller, J. (1988). Cognitive load during problem solving. Cognitive Science, 12(2). Learning as schema construction and automation; working memory is relieved as schemas form.
- Sweller, J., van Merriënboer, J. J. G., & Paas, F. (2019). Cognitive architecture and instructional design: 20 years later. Educational Psychology Review.
- Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation of physics problems by experts and novices. Cognitive Science, 5(2). Expert/novice differences are schema differences: principle versus surface.
- Chandler, P., & Sweller, J. (1991). Cognitive load theory and the format of instruction. Cognition and Instruction. Split attention: the same terms across several open sources waste the workbench; one keyword list first.
- Chandler, P., & Sweller, J. (1992). The split-attention effect as a factor in the design of instruction. British Journal of Educational Psychology.
- Ausubel, D. P. (1960). The use of advance organizers in the learning and retention of meaningful verbal material. Journal of Educational Psychology. Draft before depth.
- Mayer, R. E., Mathias, A., & Wetzell, K. (2002). Fostering understanding of multimedia messages through pre-training: Evidence for two types of pre-training. Journal of Experimental Psychology: Applied. Pre-training: a simple model first, then the dense material.
- Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students’ learning with effective learning techniques. Psychological Science in the Public Interest. Rereading is low utility. The connect-back question is elaborative interrogation / self-explanation, rated moderate utility; the relationships are the learning, not a garnish.
- Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning. Psychological Science. Retrieving without the artifact in view is formation, not review. Immediate application is retrieval plus use.
- Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior. Deeper, more connected encoding produces a more durable trace.
- Craik, F. I. M., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General.
- Anderson, J. R., & Reder, L. M. (1979). An elaborative processing explanation of depth of processing. In L. S. Cermak & F. I. M. Craik (Eds.), Levels of Processing in Human Memory. Elaboration is the working account of depth.
- Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive Science. Alfieri, L., Nokes-Malach, T. J., & Schunn, C. D. (2013). Learning through case comparisons. Educational Psychologist. Schwartz, D. L., Chase, C. C., Oppezzo, M. A., & Chin, D. B. (2011). Practicing versus inventing with contrasting cases. Journal of Educational Psychology. Analogies and comparison help when they are structural.
- Watch This To Force Your Brain To Study FASTER (2024-11-01). Public video. Transport-network picture; retention versus mastery; house source for the three-pass procedure. The video title is listed, not a program; the speaker’s paid program is not a source.