The research Struxels is built on.

Miller's 1956 paper proposed that human short-term memory could hold around seven items at once. Forty-five years later, Cowan revisited the data with stricter methods that controlled for rehearsal and chunking strategies, and arrived at a tighter estimate: about four independent chunks. Both numbers describe the same constraint - a finite buffer through which incoming information has to pass before it can be encoded into long-term memory.

Struxels respects this ceiling in two places. Each column on a board groups material that belongs to one concept, so each item the reader holds in working memory is a single column rather than a scattered set of sentences. And the visible card density per column stays low enough that scanning a column does not itself exhaust the buffer.

Sweller's cognitive load theory distinguishes between the intrinsic load of the material itself, the extraneous load imposed by how the material is presented, and the germane load consumed by building durable mental schemas. Extraneous load - the cost of parsing the format - competes with germane load for the same finite pool of attention. Paas and van Merriënboer extended this with an instrument that measures perceived mental effort directly, giving the field a way to test whether a format change actually reduces cognitive overhead.

Scrolling, paging, and re-finding the previous sentence are pure extraneous load. They do not contribute to comprehension; they subtract from it. A reading surface that holds the whole conceptual map in view eliminates that cost - the reader spends their finite budget on the ideas, not on navigation.

Larkin and Simon's 1987 analysis showed that diagrams and equivalent prose can contain the same information, yet differ enormously in how quickly a reader can locate any particular element. Their argument turned on search cost: a diagram places related elements next to each other in space, so finding a related piece is a glance rather than a scan.

The same logic governs a structured reading board. When the original prose has been spatially organised - definition next to example next to counter-example - answering a question becomes a spatial lookup rather than a linear re-read. The information is identical; the search cost is not.

Bower and colleagues demonstrated in 1969 that words organised into a hierarchy of categories were recalled at roughly three times the rate of the same words presented in a random list. Rosch's 1976 work on basic-level categories extended this by showing that the brain has preferred levels of categorisation - neither too abstract nor too granular - at which recognition and recall both peak.

Pre-categorising the source removes a step the reader would otherwise have to do under load. The categories are visible before the reader begins; comprehension begins at the level of structured ideas rather than at the level of undifferentiated text.

Paivio's dual coding theory holds that verbal and visual information are processed by partially independent channels and that information encoded in both channels is recalled more reliably than information encoded in only one. Mayer's cognitive theory of multimedia learning operationalises this for instructional design - the strongest gains come when text and imagery are temporally and spatially aligned so the working-memory system can integrate them without extra effort.

A reading surface that pairs text with a stable spatial layout gives the visual channel a job to do. The position of a card on the board becomes a second handle on the same idea - one verbal, one spatial - and the resulting memory trace is more durable than either channel alone would produce.

Roediger and Karpicke's 2006 study compared two groups studying the same passage: one re-read it; the other tested themselves on it. The testing group showed dramatically better long-term retention even though the re-reading group rated their own learning as higher. Bjork's broader work on desirable difficulties places this in context - the conditions that feel most effortful at study time often produce the strongest retention later.

Reading a board once is not the end of the encoding process. A revisit cadence that prompts active retrieval - the reader reconstructing what was in a column, rather than rereading it - is what converts the initial structure into durable memory.

The hippocampus evolved as a spatial-navigation system. O'Keefe and Nadel proposed in 1978 that it functions as a cognitive map encoding the locations of objects in an environment. Maguire's 2000 study of London taxi drivers found measurable hippocampal differences correlated with years of navigational expertise. Moser and Moser's grid-cell and place-cell work provided the cellular machinery underlying this map.

Dresler and colleagues studied participants trained in the method of loci - the ancient memory-palace technique - and found that after training, their brain connectivity patterns came to resemble those of world-class memory athletes, with corresponding gains in recall. Binding non-spatial information to remembered places exploits hardware the brain has already paid the evolutionary cost of building.

Columns respect the chunking ceiling. Nested cards give visual hierarchy. The matrix layout pre-categorises before the reader begins. Synchronised text and audio with live highlighting feed two channels at once. A revisit cadence converts initial encoding into active recall. And the palace view binds the whole structure to a spatial scene the hippocampus is already built to remember. Six findings from six different decades of research, each implemented in a single product surface, because each finding describes a separate constraint that the format of reading has to respect.