Layers of Learning

Part of Deep Processing.

Layers of Learning is an inside-out build order — first the few relations holding a topic together, then the ideas sitting on those relations, then the facts. Sources present those pieces mixed, so each new fact has to earn its place from scratch. Building the layers in that order makes every later piece cheaper to place.

The layers

Knowledge sits in layers. The layers are built by the questions that produce them, not by how large an idea looks.

The logic layer — the backbone — is the major relations, mechanisms, and organizing principles: why things connect, and in which direction. A backbone is not the first heading in the chapter. It is the handful of points carrying the most connections to everything else. Recalling from those heavily connected points brings the shape of the topic back, with the gaps sitting out at the edges. Recalling the same number of peripheral points brings individual facts and a hole in the middle. The two questions that build this layer are why this is important, and how it relates to everything else. The first cannot be answered without doing the second, which is why the pair produces relations instead of a list of headings.

The concept layer is the key ideas that populate that logic. Its questions are what the thing is, and how it happens. Logic is why-and-in-which-direction; concepts are what-and-how.

The detail layer is the specific facts, numbers, exceptions, and sub-concepts that fill those ideas. Some of those details earn early encoding: they are the concrete anchor that makes an abstract concept holdable. A concept given only in impressions stays unpicturable; one specific checkable fact snaps it into place. A painting described as bright, vibrant, an outdoor scene remains unpicturable until someone says how wide it is. The test is whether knowing this detail helps the concept it is attached to. If it does, it is encoded now. If it does not, it waits.

The leftovers are arbitrary details: pieces that do not yet help anything get understood, learned because the assessment asks for them. They are still learned. They are learned last. They are learned by repetition. And they stop being arbitrary as the structure grows.

Logic gives the frame. Concepts give the content. Details give the resolution.

Why the source order fights you

The resource does not mark these layers. Textbooks and lectures intersperse logic, concepts, and details. A heading is sometimes a logical grouping, sometimes a concept, sometimes a detail wearing a heading’s clothes.

The grouping a resource offers as its logic may not survive a later importance judgment. Importance-Based Chunking is the tool for that overruling — grouping the pieces of each layer by why they matter, not by the heading they sat under.

The layers have to be extracted and reconstructed. That is encoding, not discovery: the curriculum is not invented, it is rebuilt in a usable order. Following the source start to finish inherits the mix. Relevance then swings with no relation to what is already held, and difficulty stays high no matter how much has been covered.

When a piece arrives and the place for it is already visible, it is processed cheaply. When relevance swings, effort goes first to deciding whether the piece matters, and only then to what it means. That never-settling expensive feeling is what Cognitive Load & What Mental Effort Is Trying to Cue is trying to cue.

The layers move

The layers are not static. How relevant a detail is depends on whether the structure already exists to give it relevance. Logic generates relevance for concepts; concepts generate relevance for details. Difficulty of each pass falls as the network becomes robust enough to hold new pieces.

A large pile of facts that will not stick is a thin frame, not a hard subject.

The move when a fact will not stick is to repair the logic and concept layers, not to repeat the fact. A tree with an enormous number of leaves still has only a handful of main branches; working the branches is what makes the leaves come.

Those layers are layers of a Schema — the network a new piece attaches to, and the thing that generates its relevance.

Building the structure

Repairing the frame starts with finding the logic. The first pass skims for organizing principles: the main relations, the mechanisms, the causal chains. It is the expensive pass because it takes evaluation rather than reading — the work of Higher-Order Learning — and that expense is the point. While the map holds only a few relations, rearranging it is cheap. Once everything is on it, cleaning costs more than building did. A topic’s basic logic rarely changes much, so time spent settling it is not spent again.

That first pass names only the few largest relations and how they connect. Concepts go onto the frame next; each now has a location. Details go last. Far fewer of them then arrive unattached. Skipping to details before the logic is in place encodes information with no location for it to land.

This is self-regulated scaffolding: the topic is taken apart and rebuilt in the order that makes each step manageable. The named work that introduced scaffolding described someone else fading the support; the self-directed version is a later extension.

Order control is the mechanism that makes this possible while working sources: the next piece is chosen by what it can already attach to, not by the chapter order.

Where it fails, and how you would know

The first pass returns nothing that can be pointed at, which is why it is the pass that gets skipped. It costs more than reading and produces no visible notes.

Some material has no backbone to find. A vocabulary list, a procedure with no causal structure — the pieces barely interact, the investment does not return, and the leftover tier is the whole job.

A confidently wrong frame is more expensive than no frame, because everything gets attached to it. The bound is the same arithmetic: the frame is cheap to fix only while it is still small.

Layering does not remove the memorization tier. It shrinks it. Under a deadline the leftover details still have to be learned.

No published model matches this exact cut. What is supported outside is narrower: frames and elements before relations reduce load and improve later learning, and expertise is organized by principles rather than surface features.

A detail that would not stick before the frame was in place lands on first contact after it. A fact that still will not stick is a signal about the frame. The first pass ends when the few main relations can be said out loud with the source closed. If they cannot, that pass is not finished however much was read. Difficulty falls pass over pass. If every pass costs the same, the layers are not being built and the source order is still running the session.

Where the doing happens

The shape is a model, not a method. What to build first, and what is still cheap to fix, is the whole claim — not permission to skip the outer layers, which still have to be learned.

Prestudy is when: a short survey that finds the frame before the main event, spreading the cost of the big picture across sessions. Layering is what: which layer the session is building. Prestudy builds the logic layer; the main events fill concepts and details. Prestudy can also be a light concept pass.

Shoot is where the layers get built against real sources, question by question. Order control is the mechanism there: moving between resources for the most relevant next piece rather than following one source start to finish.

The Bear Hunter System is the four-pass loop that walks the layers in order.

Related

• Deep Processing — the parent dimension this model sits inside
• Prestudy — first-pass implementation at topic level: a short survey that finds the frame before the main event
• Shoot — where the layers get built against real sources, question by question
• Schema — the structure the layers are layers of
• Higher-Order Learning — evaluating and relating, which is the work the logic pass is made of
• Cognitive Load & What Mental Effort Is Trying to Cue — what the expensive, never-settling feeling under a mixed source order is trying to cue
• Importance-Based Chunking — how the pieces of each layer get grouped, and the tool for overruling a resource’s own grouping
• Bear Hunter System — the four-pass loop that walks the layers in order

Sources

• Ausubel, D. P. (1960). The use of advance organizers in the learning and retention of meaningful verbal material. Journal of Educational Psychology. A frame in place before the material arrives improves what is learned after.
• Mayer, R. E. Pre-training principle. Names and characteristics of the main elements, learned first, improve later learning of how those elements interact.
• Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12. Isolated elements before their relations lowers load and improves what is learned after.
• Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation of physics problems by experts and novices. Cognitive Science, 5(2). Expertise is organized by principles rather than surface features.
• Sweller, J., van Merriënboer, J. J. G., & Paas, F. (1998/2019). Cognitive architecture and instructional design. Mixed, highly interactive input keeps load high.
• Bransford, J. D., & Johnson, M. K. (1972). Contextual prerequisites for understanding. Journal of Verbal Learning and Verbal Behavior, 11. Material that is arbitrary without a frame becomes comprehensible with one.
• Wood, D., Bruner, J. S., & Ross, G. (1976). The role of tutoring in problem solving. Journal of Child Psychology and Psychiatry. Scaffolding as other-regulated fading support — the bound on the self-directed reading.
• Holton, D., & Clarke, D. (2006). Scaffolding and metacognition. International Journal of Mathematical Education in Science and Technology. The self-directed extension.