Encoded knowledge decays unless it is reconstructed, spaced, varied, and repaired over time. Spaced Interleaved Retrieval is the user’s main system for that work.

Summary

SIR combines three principles:

  • Retrieval: recall or use knowledge before checking the source.
  • Spacing: revisit material after delays that gradually widen.
  • Interleaving: mix topics, formats, difficulty levels, and angles of use.

The goal is to prove that knowledge can be reconstructed and used under varied conditions.

In the user’s system, SIR sits under Retrieval and is coordinated by Self-Regulation.

For complex problem solving and knowledge work, interleaving should be treated as reconstruction practice, not only mixed scheduling. Interleaving for Complex Problem Solving is the more general operating note for using interleaving to expose approach gaps, variable relationships, and weak problem frames.

What It Tests

SIR should test multiple levels of mastery:

  • Facts and labels.
  • Explanations of individual concepts.
  • Relationships between concepts.
  • Judgments about which relationships matter most.
  • Application in unfamiliar cases.
  • Procedure execution under realistic constraints.

If retrieval only tests recognition or isolated recall, it will miss many of the gaps that matter for real performance.

Default Retrieval Menu

Use different methods depending on the target knowledge.

NeedUseful Retrieval Format
Isolated factsNarrow flashcards, image occlusion, quick written recall.
Concept explanationTeach the concept without notes, then check.
RelationshipsAnswer relational questions or create a map from memory.
Big-picture structureClosed-book mindmap or chunkmap brain dump.
EvaluationCompare competing explanations and justify which relationship matters more.
Procedural skillSolve varied problems, execute the skill, or simulate the target task.
Exam readinessTimed mixed questions, full answer generation, and error review.

SIR Methods

This list is a working menu of retrieval methods. The goal is to choose methods by knowledge type and mastery target, not to use every method for every subject.

Declarative Methods

Use these when the target knowledge is mostly conceptual, factual, explanatory, or relational.

MethodBest UseWatch For
3CsBasic labels, diagrams, isolated terms, and fill-in-style details.Too low-order for complex understanding.
Simple flashcardsNarrow facts, definitions, constants, labels, and small details.Exploding card counts usually indicate weak encoding.
Simple relational flashcardsSpecific relationships between a small number of concepts.Do not force every relationship into cards.
Evaluative flashcardsHigh-value questions that require judgment about relationships.Better created from a strong map or question set.
Isolated generated questionsSingle-concept recall or explanation.Can become too low-order if overused.
Relational generated questionsCompare, contrast, connect, or explain interaction between concepts.Needs full answers, not mental familiarity.
Evaluative generated questionsJudge importance, relevance, priority, or conditions across concepts.Time-consuming but high value.
Linear brain dumpRapid recall of facts, terms, and broad coverage.Weak at representing relationships.
Mindmap brain dumpReconstruct the topic structure and relationships from memory.May miss tiny details unless followed by detail checks.
Isolated teachingExplain one concept or process out loud.Often less efficient than narrower tools for pure facts.
Relational teachingExplain how concepts interact.Strong early-to-mid retrieval method.
WPWHigh-volume, multi-order reteaching across whole and parts.Requires strong encoding and can expose many gaps.
Direct practice questionsAnswer from memory, then check.Can hide uncertain reasoning if checking is too fast.
Extended practice questionsAnswer, mark uncertainty, build a better answer, then check.More time-intensive but better for gap finding.
Advanced group practiceCreate, exchange, answer, compare, and discuss challenging questions.High setup cost; best for important topics.
Evaluative discussionArgue and judge relationships, importance, and edge cases.Must be directed; undirected discussion is usually weak.
Chunkmap reconstructionRebuild or improve the chunk structure from memory.If hard, the original encoding may have been shallow.
Feynman-style explanationExplain simply, find gaps, refine the explanation.Useful as micro-retrieval, not a full system by itself.
Memory systemsOrdered lists, labels, vocabulary, or narrow rote demands.Avoid using them as a substitute for encoding.

Procedural Methods

Use these when the target knowledge is mostly skill execution, problem solving, production, language use, coding, calculation, lab work, or performance.

MethodBest UseWatch For
Retrieved execution, simplePerform a basic procedure from memory.Stop once it becomes repetitive beyond mastery.
Retrieved execution, integrativeCombine multiple component skills into a larger function.Move here early once simple execution is stable.
Retrieved execution, appliedStart from a target output and work backward to execute.Best for realistic transfer.
Simple challengesSolve prompted basic problems.Good entry point, but not enough for mastery.
Integrative challengesCombine multiple processes to solve a less predictable task.Requires both procedural skill and declarative understanding.
Edge-case challengesHandle atypical, unfamiliar, or boundary-condition problems.Excellent for high-level transfer.
Variable modificationChange values or constraints while keeping the same basic problem type.Good for early variation.
Variable additionAdd new conditions that change the problem structure.Requires stronger conceptual judgment.

Methods To Avoid As Defaults

  • Passive rereading.
  • Passive rewriting.
  • Passive relistening.
  • Practice questions answered only mentally.
  • Checking answers before fully generating an attempt.
  • Undirected group discussion.
  • Repeating already-mastered procedures without variation.

REBIM (Repetitive Execution Beyond Initial Mastery) is the specific failure mode for procedural practice. REBIM is repeating a skill or process that can already be performed to high proficiency — coding the same simple scripts, solving the same difficulty of problems, or running the same conversation pattern in a language already well-handled. The benefit is negligible compared to what could be achieved with variation, edge cases, or applied challenges in the same time. REBIM typically feels productive because it produces smooth performance, which is indistinguishable from progress to the learner.

Encoding and Retrieval: The Tradeoff

Encoding and retrieval are not separate processes — the research treats them as fundamentally intertwined. In practice they are addressed separately because the techniques differ, but they influence each other directly: stronger encoding produces higher retention, which means less retrieval is required to maintain it. Weaker encoding produces more knowledge decay, which means retrieval sessions must be more frequent and more extensive.

As encoding skill develops, retrieval sessions become shorter and find fewer gaps. In the early stages — when encoding quality is still low — retrieval sessions will find many gaps, take longer, and feel laborious. This is the expected state, not a sign the system is failing.

Finding many gaps during retrieval is a sign the session is working, not that it is ineffective. A retrieval session that finds few gaps when mastery is still low is the sign of a session that is not probing deeply enough.

Concrete Spacing Schedule

The exact schedule matters less than session quality and consistency. A practical default that works across most academic contexts:

WhenSession type
Same dayEncode and do a light retrieval pass on new material.
Next day (or within 2 days)First gap-finding pass for yesterday’s content.
End of weekInterleaved retrieval for all material from the last two weeks.
End of monthInterleaved retrieval for all material from the last four weeks.
3–4 weeks before assessmentComprehensive retrieval pass for all relevant content.

With this schedule, any given piece of material is encoded and re-encoded approximately six times in a month — through prestudy, first encoding, same-day retrieval, mid-week retrieval, end-of-week retrieval, and end-of-month retrieval. At strong encoding levels, the end-of-week pass for a full week’s material takes a few hours; the end-of-month pass for four weeks takes roughly half a day.

The schedule can be followed loosely. Exact timing matters less than ensuring each layer happens.

Scheduling Heuristic

The exact schedule is less important than session quality. A useful default for declarative-heavy material:

TimingMain Job
Same dayMicro-retrieval while encoding.
2-5 daysFirst gap-finding pass.
1-2 weeksRelationship and question-based pass.
1 monthLarger mixed challenge.
2 months and beyondMaintenance, transfer, and higher-order challenge.

For procedural-heavy topics, move sooner toward varied execution practice. Declarative setup may still matter, but the system should quickly shift toward doing.

Session Rules

  • Retrieve before looking.
  • Write, speak, draw, or perform the answer fully.
  • Mark uncertainty as a gap even if the final answer is correct.
  • Check against the source, answer sheet, or worked example.
  • Decide whether the gap needs retrieval repetition or re-encoding.
  • Update the next session based on what failed.

Gap Types

GapSymptomRepair
Fact gapMissing term, label, number, definition, or stepNarrow flashcard or quick recall drill.
Relationship gapCannot explain how ideas influence each otherRebuild local chunk map.
Structure gapCannot reconstruct the topic’s big pictureRe-Skin the BHS map.
Transfer gapCan answer familiar questions but fails variationsAdd interleaving and examples.
Confidence gapCorrect answer but shaky reasoningGenerate full answer sheet and compare.
Execution gapConcept is known but performance failsPractice the target procedure with feedback.

Relationship To Encoding

SIR improves the encoding system because retrieval failures reveal whether Bear Hunter System produced a usable structure.

If retrieval finds many lower-order gaps, do not automatically add hundreds of flashcards. First ask whether the encoding structure was weak. Better chunking can reduce the amount of isolated memorization needed.

LLM Use

LLMs can help SIR by:

  • Generating question variations.
  • Creating mixed practice sets.
  • Producing counterexamples.
  • Checking answer completeness.
  • Asking follow-up questions after a weak answer.
  • Converting a map into retrieval prompts.
  • Generating questions at varied difficulty levels across mastery levels.

Use AI for: saving time on low-cognitive tasks (collecting keywords, big-picture orientation); consolidating multiple sources into a single source to reduce split attention; generating test questions; asking questions to explore a topic like a tutor.

Avoid using AI for: checking the truth of nuanced or complex claims (LLMs cannot reliably arbitrate on nuanced topics); bypassing the mental modeling work of chunking, evaluating relationships, and judging importance — these are the processes that create encoding. If AI does the structural thinking, the learner coasts into passive understanding mode without building the schema. For beginners especially, avoid using AI to chunk and organize information until there is a reliable internal sense of when cognitive load is passive versus active.

The practical test: retrieve first, fully generate an answer, then use the model for feedback and calibration. When AI becomes the primary thinker and the learner becomes the editor, the benefit of retrieval practice disappears.

How It Should Feel

SIR should feel like controlled strain. The point is to expose what can and cannot be reconstructed under spacing, variation, and discrimination.

Good signs:

  • recall feels effortful but possible;
  • gaps become specific;
  • interleaving makes similar ideas easier to distinguish;
  • spacing reveals what is actually stable;
  • and failures point back to either retrieval practice or re-encoding.

Warning sign: SIR has become recognition when checking notes feels easier than reconstructing first.

Open Questions

  • What is the user’s real weekly retrieval capacity?
  • Which current subjects are declarative-heavy, procedural-heavy, or mixed?
  • Should this vault include an SIR scheduler template?