Multipass System

The Multipass System is a way of covering a block of material — a whole unit, all five weeks of it — where every pass crosses the entire block and the early passes stay shallow on purpose. Each sweep goes a little deeper than the last, so the shape of the subject and the reasons inside it are in place before any leftover fact is touched.

One pass is the whole block

A pass is one sweep of everything in scope: cross all of a unit at the first depth, then cross all of it again, deeper. The early sweeps are not skims — a skim samples, while a first pass here is complete in coverage and shallow in depth, touching every part of the block at the logic layer before any part gets detail. That is the mechanism. A deep pass that never reaches the back of the block leaves a hole nothing later covers; shallow complete sweeps miss different things each time, so what one drops the next catches. Three passes is the usual count, four or five when one depth needs a couple of days, and the passes belong on separate days where the calendar allows — a schedule that spaces the encoding by construction. Scope runs against instinct: take the largest block a sweep can honestly cross — studying biology, sweep the whole of biology at once — up to a sizing limit: a subject so large that one sweep would take weeks gets split into topics and swept topic by topic. Its natural home is large, dense material where the relationships carry the marks, and the compressed window it was built to run inside is Cramming.

Logic first, details last

The order inside the sweeps is the strategy. Logic and concepts go down first; details come last, and only if time remains — because details do not retain without the logic under them, so the conventional order ends up paying for its details twice. The time budget flips with it: details routinely swallow half the study time or more while making up the minority of what a subject is, and the marks lean the other way — the bulk derives from logic and concepts, with pure detail-recall questions tending to score low — so the flip hands logic and concepts the larger share of the hours and leaves details the remainder, an allocation closer to how the knowledge itself is structured. The layers are not assigned to the passes; they emerge — as the chunks shrink, the questions change on their own — and a pass is not tidily one layer; the third usually straddles two. The progression is a tendency to ride, never a schedule to enforce. The layer model itself is Layers of Learning‘s subject.

All of which leaves the reader’s actual method untouched. Multipass is not a way of studying — it decides what is being studied and when, and changes nothing else. Sessions run exactly as they already run, on Bear Hunter System in most of this vault’s cases or on whatever encoding method is already in hand; this is Order Control raised from the session to the block.

Inquiry-led: the questions make the layers

When no deadline is pressing, the sweeps steer by inquiry: follow what is currently worth asking, pass by pass. The first pass runs the biggest chunks on the basic questions — what is this, what does it mean, why does it matter, how does it fit together — and builds the framework. On the next pass the questions get harder on their own, because the framework makes sub-concepts visible that were invisible before — this is the pass where the concept layer takes shape — and nothing needs escalating by hand. A third pass chases sub-concepts and specific processes, feeding them into the map or onto cards; a fourth, if time exists, goes after genuinely fine detail. The steering is Aim — writing what the material must answer before opening it. Of the two steerings this is the quicker one and the more pleasant to run, its encoding is the deepest, and the skill it demands is the highest — which also makes it the steering for material that future learning has to stand on. A reader with no questioning habit cannot run it yet, and the closing section says what that costs.

Objectives-led: easiest first, then join the spikes

Under a hard deadline, or for a topic that will not matter much past the exam, the sweeps steer by the syllabus instead. Each pass takes whichever objectives currently rank easiest — as many as one session holds: given six hours, however many objectives six hours can cover — and takes them from everywhere in the block at once, so what the early passes leave behind is knowledge in scattered spikes, fragmented on purpose. Later passes join them: the spikes get worked into a structural map, the gaps met while joining become the next pass’s aims, and what was hard last pass re-ranks as easy once its neighbours are known. The map-building is also the repair for what this ordering skips — the usual chunk-finding never happens up front, and joining the spikes forces it back in. Objectives are the right list because a curriculum is legally obliged to publish them and the assessment is obliged to build on them — interpreted loosely enough that curveballs survive — and the ordering’s price is real: it spends encoding depth and curveball resilience to be easier under pressure. When nothing is left on the list, the run is finished, often by the fourth pass; if not, it runs again.

The routing rule between the two is the system’s own default: no exam emergency, inquiry-led; strong pressure or low long-term relevance, objectives-led. Both share everything that matters — every pass covers the block, and the logic goes down first. One system, two steerings.

How the run ends

A finished run closes with a clean sweep in either steering: one final pass, checked against the objectives — the list that keeps it off low-yield material — to find what every earlier sweep missed. The sweep itself is not optional; the follow-ups are. Facts that will not derive go to Flashcards or a scene from Method of Loci; if time allows, an interleaving pass sits on top; and what the sweep finds feeds ordinary Revision.

An unfinished run is the other case, and it is where the ordering pays hardest. Reaching only the second pass still leaves logic and concepts banked across the whole block, and the bet behind that position has four parts. What the exam will ask is unknowable either way — but whatever it asks, it will be probing understanding of the topic, so that much is certain. Understanding is built mostly out of logic and concepts — the system puts the share at 80 percent. Then the arithmetic: against a paper of a hundred questions drawn from a thousand or ten thousand askable things, memorising a few hundred wins about four in ten even assuming perfect recall of every one, while deriving answers from understood logic wins closer to seven in ten and stretches to questions nobody planned for. And the last part rides along free: some fine details stick without ever being memorised. The figures are the system’s own accounting of the bet, not measured results.

What it costs, where it fails

An advanced move, by its own material’s admission: without order control, a questioning habit, and a working way to take notes, the passes have nothing to run — the prerequisites live with the encoding method, not here. The evidence line is equally plain: the layer model and the odds above are the system teaching itself; the outside record supports spreading encoding episodes across days over one massed effort, and conceptual structure helping knowledge travel, neither of which is a trial of this protocol. The felt price: every pass costs a full traverse of the block, so the first two feel like nothing is being learned — deliberate shallowness reads as failure to anyone who needs a finished chapter for reassurance. The quit signals are structural. A second sweep no faster than the first means the first was not shallow — sections got finished instead of the block covered. No questions available at the end of pass one means inquiry-led is not available yet; objectives-led is the route. A clean sweep returning the same gaps means the passes are not deepening. And working is checkable: pass two runs faster over the same material than pass one did, its questions are about how parts connect rather than what things are, and by the clean sweep most remaining gaps are arbitrary facts rather than mechanisms.

Which returns to the opening’s two facts — the whole block, every pass, shallow first — now carrying what they buy: the run is safe wherever it stops. Because every sweep crossed everything, an interrupted run leaves a working structure over the whole subject rather than a finished front and an untouched back — and a completed one pays out the same way: a high count of high-yield points covered, retention that keeps recall high, and application that comes out correct because the structure of how the knowledge fits together went in first. When the sweeps are done, Spaced Interleaved Retrieval takes over as the top-up.

Links into the knowledge base

• Cramming — the compressed window this operator was built to run inside; it owns the when, the skip-and-sleep guidelines, and the early finish.
• Bear Hunter System — the encoding method most sweeps in this vault run on; each session is unchanged.
• Layers of Learning — the depth layers the passes move through.
• Order Control — the steering skill this system operates at block scale.
• Spaced Interleaved Retrieval — the retrieval top-up once the sweeps are done.
• Revision — where the clean sweep’s findings get worked.

Open questions

• How shallow is shallow enough on a first pass — where does a layer stop being a layer and become a skim?
• Where is the block-size line past which sweeping a whole subject costs more than sweeping it topic by topic?
• Does the question ladder still form for a reader with no prior structure in the domain, or does the first pass stall?
• Does objectives-led’s encoding cost show up on a delayed test, or only on curveballs?

Sources

• Cepeda, Pashler, Vul, Wixted & Rohrer (2006); Cepeda, Vul, Rohrer, Wixted & Pashler (2008) — several separated encoding episodes beat one massed effort; the sweeps-across-days schedule is spacing by construction.
• Pan & Rickard (2018) — transfer benefits of elaborated, structured knowledge; compatible with logic-first ordering, not a trial of it.
• The pass structure, both steerings, and the odds are the learning system this wiki organises, speaking for itself; its figures are its own model throughout, never measurements.