How to diagnose and fix exam mistakes

Every error left over from a paper is written into one of three rows before anything else is done: higher-order, lower-order, or procedural. Each row has its own cause and its own fix. Treating every error as carelessness leads to more hours of the same practice. The score does not improve.

The sort has three columns: what the error looks like on the paper, what caused it, and what the next practice has to be.

Core takeaways

• Every leftover error goes into one of three rows: higher-order, lower-order, or procedural. Each row has a different cause and a different fix.
• A higher-order error comes from a conceptual gap. The fix is to build the structure again, on purpose, with why, how, and relationship. Re-reading does not fix it.
• A lower-order error is a slip at one step while the structure is intact. The fix is to slow that step and add a review pass. A missing fact gets a targeted retrieve or a card.
• A procedural error is a wrong or skipped step in a sequence the learner already knows. The fix is to practise the sequence as a sequence.
• Calling an error “silly” produces two moves: ignoring the error, and trying harder. Neither puts a check at the step where the slip happens.
• When a mistake could be conceptual or procedural, two tests decide the row before any fix: double generation and the two-attempt check.

Higher-order errors

A higher-order error is an error in how ideas connect, in which approach to take, or in what the question is asking. On the paper it looks like the wrong approach, a misread question, or a concept applied in the wrong domain.

The cause is a conceptual gap. The schema is the organised structure the learner currently holds for a topic, including what hangs off what. In a higher-order error, that schema does not connect at the level of understanding. On review, the official answer reads as obvious, and the learner cannot produce the chain of reasoning with the source closed. The feeling that the answer is obvious is the sign of this kind of error. It is not proof that the knowledge was there.

A conversion error can look like a slip. It often stops happening once the learner understands the unit as a quantity. Before that, the unit is a label to swap. After that, a wrong conversion fails two tests: it breaks the memorised rule, and it does not make sense.

The fix is to re-encode the concept: build the structure again, on purpose, with why, how, and relationship. Re-reading does not do this. That rebuild at the learner’s current skill level is Best-attempt Encoding. The four-pass workflow that the re-encode instruction points at is the Bear Hunter System.

Lower-order errors

A lower-order error is an error in a specific fact, a unit, a sign, or a copied number. On the paper it looks like an arithmetic slip, a sign error, the wrong word, a missed unit, or a mis-copy.

The schema is intact. The slip happened at a specific step. The cause at that step is load, haste, a missing check, or a detail that was never made automatic. Sometimes the drop comes from one night’s sleep or from acute stress. More often the slip is a habit: the label “silly” has let the same slip run for years.

The fix is to slow that step and add a review pass for mechanical output. If the hole is a missing fact, the fix is a targeted retrieve or a card. A slower line of working does not supply a missing fact.

The same three names appear at preparation scale on How to prepare for ultra high-volume exams. On that page, lower-order means missing facts. On this page, lower-order may mean a missing fact or a slipped step.

Procedural errors

A procedural error looks like this: the concept was known, the approach was right, and the sequence was wrong or a step was skipped. The cause is that the sequence is not yet automatic. The fix is to practise the sequence as a sequence. Practising the components alone does not make the sequence automatic.

Errors that do not fit the three rows

The three rows do not cover every exam error. Two kinds sit outside the table. The first is pure anxiety: errors that are random, linked to the stakes, and show no content pattern. The second is questions on material that was never studied.

The three rows will mis-sort the random, stakes-linked pattern into lower-order, and the learner will slow steps that are not the problem. A slower arithmetic line does not fix that pattern. The pattern, and the five-letter protocol for it, are on Silly Mistake Syndrome.

Why the label “silly” misleads

The label “silly” is appealing because it frames the error as a lapse of attention. A lapse of attention seems to have a simple fix: be more careful. The label produces two moves, ignoring the error and trying harder. Neither move installs a check at the step where the slip happens. “Be more careful” as a general intention does not install a check either. A habit changes under monitoring. A vow on its own does not change a habit. The way “be more careful” turns into the appearance of care is described on The Shortcut Problem.

A higher-order error treated as carelessness will recur, because the conceptual gap remains. A lower-order error treated as a conceptual gap wastes encoding effort on something already understood. These two mismatches are the reason for the three-way sort.

Diagnose before fixing

Diagnosis comes before any fix. Extra practice at the wrong level has a cost: it reinforces the wrong schema or the wrong process. Practice at the wrong level makes the error harder to fix later. It does not make the error impossible to fix. Gaps in applying the concept are closed first. Habits for sitting the paper come second. Exam technique applied over a knowledge gap does not fill the gap. The sort exists so that the next hour of practice is spent at the right level. A longer hour at the wrong level does not help.

Two tests for an unclear error

When a mistake could be conceptual or procedural, one of two tests runs before the row is chosen.

Double generation has three steps. The learner produces the answer. The learner then produces a model answer from notes. The learner then compares both to the official key, if one exists. The answer is written out. Answering in the head is not enough. Low confidence counts as a gap even when the answer was right. Producing the model answer is itself the learning, because the model sheet shows gaps the question did not test. Using an official answer sheet alone skips that step. When no official key exists, the learner compares two model sheets built independently. A partner’s questions find holes that one’s own paper will not.

The two-attempt check is the second test. The learner produces the answer twice from memory, ten minutes apart, without looking at the first attempt. Ten minutes is a fixed number this system uses. The gap is not there to let recency fade. Not looking at the first attempt does more of the work than the length of the gap. The step where the two attempts differ is the unstable step. If both attempts fail at the same place, the cause is a gap. Carelessness is not the cause. If the attempts fail at different places, the cause is execution variance. Execution variance is the one thing this test can see and double generation cannot. Generating an answer once can feel fluent even when the knowledge is shaky. That is why more than one generation is needed.

The review loop

After each practice session or exam review, the learner sorts every error into one of the three rows before choosing the next action. The loop starts with the cluster of errors that is costing the most. The learner names the largest root cause, tries one targeted repair, looks again, then repeats. A complete list of last week’s errors followed by “be more careful” is the outcome the sort exists to prevent.

A higher-order error flags the concept for re-encoding. Re-reading is not re-encoding. A lower-order error identifies the step that produces the slip, and the learner slows that step. If a fact is missing, the learner runs a targeted retrieve instead. An ambiguous error runs one of the two tests above before the row is chosen. The learner tracks across sessions which row the errors cluster in. A cluster that persists points to a systemic fix.

The retrieval routine that surfaces the errors this sort then takes is Spaced Interleaved Retrieval. The steering layer that makes sort, then fix, then track clusters a habit rather than a one-off is Self-Regulation.

What the loop costs and when to stop

The price of the loop is writing the answer out, then writing a model answer, every time a paper is reviewed. This is slower than glancing at a key. The signal to stop is two review cycles that produce only “be more careful” as the next action. The loop is working when the next paper’s errors land in a named row and that row’s fix is the only extra practice run. Once every paper is sorted into rows and given each row’s fix, the label “silly” is no longer used.

How to practice this

1. After each practice session or exam review, write every leftover error into a higher-order, lower-order, or procedural row. Do this before choosing any next action. Notice any error with no content pattern; a random, stakes-linked error does not belong in a row.
2. For an error that could be conceptual or procedural, write the answer out, then a model answer from notes. Compare both to the official key, or, if there is no key, compare two independently built model sheets. Notice any gap the question did not test, and any right answer given with low confidence.
3. For an unclear error, write the answer from memory twice, ten minutes apart, without looking at the first attempt. Notice whether both attempts fail at the same step or at different steps. The same step means a gap; different steps mean execution variance.
4. For a higher-order error, build the concept’s structure again, on purpose, with why, how, and relationship. Notice whether you can then produce the chain of reasoning with the source closed.
5. For a lower-order error, slow the step that slips and add a review pass, or retrieve the missing fact. For a procedural error, practise the sequence as a sequence, not the parts alone. Notice whether the same slip has repeated across papers; a repeat is a habit, not a one-off.
6. Start with the cluster of errors that is costing the most. Name its largest root cause and try one targeted repair, then look again and repeat. Track across sessions which row the errors cluster in; a cluster that persists points to a systemic fix.

Related pages

• How to prepare for ultra high-volume exams: the same three gap types at preparation scale; lower-order there means missing facts
• Best-attempt Encoding: what “re-encode” looks like at the current skill level
• Spaced Interleaved Retrieval: the retrieval routine that surfaces the errors this sort then takes
• Bear Hunter System: the four-pass encoding workflow the re-encode instruction points at
• Self-Regulation: the steering layer that makes sort, then fix, then track clusters a habit rather than a one-off

2 more related pages

• The Shortcut Problem: why “be more careful” turns into the appearance of care
• Silly Mistake Syndrome: the random, stakes-linked pattern and the five-letter protocol that sit outside the three rows

Open questions

• Which error category accounts for most of the current exam losses?
• Do higher-order errors cluster around specific concept types or domains?
• At what point in a long practice session do lower-order errors start increasing?

Sources

• Callender, A. A., & McDaniel, M. A. (2009). The limited benefits of rereading educational texts. Contemporary Educational Psychology, 34(1), 30–41. Rereading is a weak repair for a conceptual gap.
• Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students’ learning with effective learning techniques. Psychological Science in the Public Interest, 14(1), 4–58. Rereading rated low utility; elaborative interrogation and self-explanation rated moderate. The why / how / relationship prompt is that interrogation.
• Metcalfe, J. (2017). Learning from errors. Annual Review of Psychology, 68, 465–489. The benefit of an error depends on it being identified and corrected, not merely repeated past.
• Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science: An illusion of explanatory depth. Cognitive Science, 26(5), 521–562. Generating once can feel fluent even when the knowledge is shaky. Answering in the head does not break the illusion.
• Fitts, P. M., & Posner, M. I. (1967). Human Performance. Brooks/Cole. Cognitive → associative → autonomous skill acquisition. Automaticity is the target of sequence practice.