Reviewing one chapter's cards in a row feels fluent. Mixing related cards feels harder — and builds the discrimination exams quietly require. Interleave on purpose so your deck trains choice, not only recognition inside a silo.

Blocked flashcard sessions are comfort food: today's tag, today's pile, today's sense of mastery. You flip through acids, rate them Easy, and leave convinced the topic is "done." Then a paper alternates acid-base with redox and kinetics, and the fluency evaporates. The cards were never wrong. The session composition was.

Interleaving — mixing cards from related topics within a review block — lowers that fluency and improves later ability to choose which idea applies. Robert Bjork's desirable difficulties and classroom interleaving research (including mathematics work associated with Rohrer, Taylor, and colleagues) keep making the same tradeoff visible: blocked practice boosts in-session performance; mixed practice often wins on delayed tests that require discrimination.

Exams rarely announce "this section is only chapter 4." Your deck should sometimes refuse to stay in one chapter too.

Why blocked reviews feel smarter than they are

When every card in a sitting shares a chapter label, your brain can lean on context instead of content. You know today's pile is "electrochemistry," so half the discrimination work is already done. That is cue dependence dressed as mastery: retrieval looks strong because the environment is doing part of the selecting.

Transfer-appropriate processing is blunt here. If the test mixes twins, practice that never mixes twins is misaligned — even with perfect spacing and honest ratings.

Blocked decks vs interleaved reviews

The middle rows feel worse. That worse feeling is often the signal that storage strength and discrimination are actually being trained.

How to interleave without breaking FSRS

You do not need to destroy your scheduler or abandon due dates:

• Allow due cards from multiple related tags in one sitting
• Add a short "mix pack" of comparison cards (A vs B on cue C)
• After due reviews, do 8-12 interleaved exam-style questions
• Occasionally free-recall a topic list without looking at prompts
• Keep FSRS ratings honest — mixing harder sessions is not a reason to mark Easy when recall was shaky

The algorithm spaces items. You still choose session composition. Those are different levers. Spacing without discrimination practice can leave you with a green silo and a surprised exam face.

When to keep blocking briefly

New cards right after first encoding can stay blocked for a few reps while the schema stabilizes. Working memory is already busy installing the model; immediate interleaving of three unfinished twins can create extraneous load rather than useful struggle.

A practical rule: block until teach-back of the core claim is clean, then mix. Permanent blocking is the trap — not short blocked warm-ups.

Discrimination cards beat more of the same

Interleaving shines when the deck includes cards that force a choice:

• "Given this observation, is this acid-base or redox — and why?"
• "How do A and B differ on the exam's favorite cue?"
• "Which rule fails in this boundary case?"

Ten more definition cards in the same silo will not fix a discrimination gap. One comparison cue, reviewed in mixed company, often will.

Pair cards with paper interleaving

Full mocks already mix. Use error logs to spot discrimination failures — the misses where you applied the wrong twin. Write comparison cards for those exact collisions. Re-test them in mixed reviews, then confirm with a short mixed question set.

That loop is interleaving with a purpose: exam error → comparison cue → mixed retrieval → remixed paper item. Without the middle steps, you only experience the mix once every mock and forget the repair.

Calendar it so comfort cannot erase it

Label a block Retrieval mix — topics A/B/C so the habit survives busy weeks. Comfort will try to revert you to single-tag evenings because they feel efficient. Efficiency during the session is not the metric; discrimination on delayed, mixed cues is.

Two to four related topics per mix is a strong default. Twenty unrelated subjects in a blender is fidgeting, not interleaving.

What interleaving does not replace

Interleaved reviews do not invent understanding. If you cannot teach the idea back, mixing fog with fog trains confusion. Encode and clarify first; then mix retrieval. Spacing, interleaving, and transfer practice are layers — not substitutes for a schema.

A worked micro-protocol (25 minutes)

1. Due cards from tags A and B only (related twins) — 12 minutes
2. Three comparison cards or free-recall "how do I tell A from B?" — 5 minutes
3. Six mixed exam-style items that refuse one chapter label — 8 minutes
4. One line in the error log if a twin collision appeared

Research snapshot: why mixing beats silos for discrimination

Classroom and lab studies on interleaved mathematics and category learning keep finding a pattern: blocked practice looks better during training; interleaved practice often looks better on delayed tests that require choosing a method. When every item shares a chapter label, your brain can outsource the "which concept?" decision to the environment. When items alternate near-twins, each trial forces a selection — and selection is what mixed exams quietly grade.

That is why a green chapter deck can coexist with a red paper. The deck trained fluency inside a silo. The paper demanded discrimination without the silo's hint. Interleaving moves some of that demand into practice, where errors are cheap.

Measuring whether the mix is working

Do not use "felt smooth" as the KPI. Use:

• Discrimination errors shrinking on mixed sets
• Fewer wrong-twin applications on papers
• Comparison cards moving from chronic Again toward Good without peeking

If mixed performance never improves, check relatedness of the mix, schema quality, and whether you are still peeking. Interleaving cannot fix a missing model; it can only force the model to choose under pressure.

FAQ

Won't mixing tank my retention stats? Short-term ratings may look harder; long-term transfer often improves. Don't optimize only for easy sessions or green streaks inside one tag.

How many topics per mix? Two to four related ones is a strong default. Relatedness matters more than count.

Is tag shuffle enough? Great start; still add non-card mixed questions and occasional free recall. Cards alone can stay cue-narrow.

When should I stop interleaving for a week? If teach-backs on the core topics are collapsing, pause the mix, repair schemas, then return to interleaved retrieval — don't use blocking forever as avoidance.

Related reading: Flashcards cluster

• When Flashcards Aren't Enough
• How to Write Flashcards That Transfer
• Interleaving on Your Calendar
• Interleaving vs Blocked Practice Exams
• Exam Transfer vs Recognition
• FSRS vs Cramming

Keep going across Sukrat

• Schedule the Struggle
• Full Papers Train Transfer
• Protect Retrieval Blocks