Self-explanation is one of the best-supported study techniques in the literature — and one of the most misused. Students hear "explain it to yourself" and produce vague narration: "okay, so this makes sense because... yeah, that works." It feels like engagement and it's the same fluency problem as rereading, because nothing checks whether the explanation is right.

The fix is making self-explanation scoreable. After each worked step, write one sentence a TA could mark wrong: why this move is legal, what illegal move was tempting, what fails if you skip it. If your sentence can't be scored — if a stranger couldn't say "that's wrong" — it's journaling, not self-explanation.

The distinction is between narration and justification. "This step combines like terms" is narration — it names what happened. "This step is legal because the coefficients are on the same variable" is justification — it names the condition that makes the move work. The exam asks for the justification; the narration is what the worked example already shows.

Why vague self-explanation fails

Saying "makes sense" at each line is the self-explanation version of highlighting — it produces the feeling of engagement while training nothing. You're reading the step, confirming it looks right, and moving on. What's missing is the check: can you state the condition that makes the step legal, and would a wrong version of the same sentence sound plausible? If "you can always multiply both sides" sounds right to you, the self-explanation isn't working.

Self-explanation works when it's specific to a step — Chi's research showed the benefit comes from explaining the why of individual moves, not summarizing the general approach. Vague summaries recreate the fluency problem: they feel like understanding because they paraphrase the content, but they don't test whether you know the hidden condition. A scorable sentence forces the specificity: "legal because n is positive" is checkable; "it's algebra" is not.

The scorable-explanation workflow

Scope to one worked example. A proof or derivation from lecture — the worked steps your professor showed, not a summary of the topic. The specific steps are where the hidden assumptions live; the topic-level version skips them.

One sentence per non-obvious step. Most lines in a worked example are mechanical — combine terms, substitute, simplify. The interesting ones are where a condition does the work: multiplying by n assumes n is positive, dividing assumes nonzero, swapping assumes the relation is symmetric. Those are the steps worth a sentence.

Three sentence shapes. Why is this move legal (names the hidden assumption)? What illegal move was tempting (names the trap)? What fails if you skip this step (names the dependency)? Each shape produces a checkable sentence — a TA could say "no, n could be negative" and mark it wrong.

Name the fake-progress version. For each item, say what "makes sense" would have missed: "nodding at the multiply-by-n step doesn't tell me why it's legal." The contrast makes the scoreability visible — a sentence that can't be wrong isn't an explanation.

Retest with a different step. Same worked example, different non-obvious line — the 1/3/7-day spacing holds, and the new step prevents memorizing the explanation instead of learning the justification pattern.

A 40-minute session on the clock

The weak answer is the trap — "you can always multiply both sides" is the kind of claim that sounds like algebra until you try to score it. It names no condition, so it can't be wrong, so it explains nothing. The strong answer produces the condition: n is positive in this lemma, which preserves the inequality direction. And it names the failure: zero or negative flips or breaks the step. That's a sentence a TA could mark — and a condition the exam will test.

Why scoreability is the point

The value of self-explanation isn't the explaining — it's the check on the explaining. When you write a sentence that could be wrong, you commit to a claim about the step's justification, and the worked example can confirm or correct it. When you narrate vaguely, there's nothing to check — "makes sense" is always true, which is why it teaches nothing.

This is why the TA-gradable test is the right standard. "Would a TA mark this wrong?" is the same as "does this sentence make a claim?" — and claims are what exams score. The student who produces justifications can answer "why does this work"; the student who produces narration can only say "it does." The exam asks for the first.