Turning revision notes into practice questions and self-tests converts stored information into something a learner must produce, explain, choose or apply. A useful self-test does more than hide a definition and reveal it again. It can ask for recall, relationships, method selection, interpretation, error diagnosis and transfer to a new context. The question should match the capability you are trying to strengthen.
Students searching for practice questions from notes, quiz from notes, self-testing for exams or how to create practice tests often meet tools that can generate questions instantly. Speed is useful, but question quality still matters. A large quiz can repeatedly test recognition while leaving explanation and application untouched. An answer key can look authoritative while containing an error. A self-test can become too familiar to reveal what the learner can do on a fresh task.
This guide shows Alicia, Tricia and Kai Kai how to convert checked revision material into a progression from recall to explanation, application, discrimination and transfer. It also shows how to verify answers, vary questions without changing the underlying concept, use AI as a drafting assistant rather than an examination authority, and interpret self-test results without pretending that a home-made quiz is an official mock examination.
50-second route
- You only make definition cards: add relationship and application questions.
- Your quiz scores are always high: reduce cues and use fresh variants.
- You cannot write good questions: begin with the learning claim and ask what output would demonstrate it.
- You use AI: verify scope, questions and answers before repeated practice.
- You need exam realism: hand off to matched past papers and simulations rather than pretending a generated quiz is official.
1. Start From a Learning Claim
The question-design job. This stage focuses on deciding what knowledge or decision the question should expose. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s note that final amount equals original times a multiplier. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid writing a question before knowing what it is meant to test. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
2. Write Recall Questions
The question-design job. This stage focuses on requiring production without supplying the answer. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s technical vocabulary note. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid turning every prompt into multiple choice. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
3. Write Relationship Questions
The question-design job. This stage focuses on asking how ideas connect rather than listing isolated facts. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s cause-and-effect process. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid rewarding a memorised sequence with no explanation. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
4. Write Contrast Questions
The question-design job. This stage focuses on forcing discrimination between neighbouring ideas. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s mean-versus-median notes. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid testing each concept separately forever. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
5. Write Why Questions
The question-design job. This stage focuses on making the learner justify a method or claim. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s percentage multiplier note. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid accepting a correct rule with no understanding of its condition. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
6. Write How Questions
The question-design job. This stage focuses on requiring a process to be reconstructed. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s science procedure. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid asking only for the final result. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
7. Write Example Questions
The question-design job. This stage focuses on making abstract ideas concrete. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s definition of inference. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid using the same memorised example every time. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
8. Write Counterexample Questions
The question-design job. This stage focuses on testing the boundary of broad rules. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s ‘always’ statement. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid assuming a slogan is universal because it worked once. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
9. Write Application Questions
The question-design job. This stage focuses on moving from statement to use. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s formula notes. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid changing only the numbers while leaving the decision cued. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
10. Write Representation-Change Questions
The question-design job. This stage focuses on moving among words, diagrams, tables and symbols. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s graph relationship. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid changing so many features that failure becomes uninterpretable. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
11. Write Method-Selection Questions
The question-design job. This stage focuses on requiring a choice among taught methods. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s algebra notes. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid putting the method name in the heading. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
12. Write Error-Diagnosis Questions
The question-design job. This stage focuses on asking what went wrong and why. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s sign-error example. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid showing the correction before the learner diagnoses the mechanism. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
13. Write Completion Questions
The question-design job. This stage focuses on fading part of a worked example. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s multi-step calculation. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid removing so much support that the task becomes first instruction again. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
14. Write Explanation Questions
The question-design job. This stage focuses on requiring a claim, evidence and reasoning. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s reading inference note. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid accepting a quotation with no link to the claim. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
15. Write Comparison Questions
The question-design job. This stage focuses on requiring a relationship across two cases. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s two passages. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid writing two separate summaries. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
16. Write Data Questions
The question-design job. This stage focuses on separating calculation, description and interpretation. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s small dataset. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid turning correlation or pattern into unsupported causation. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
17. Write Sequence Questions
The question-design job. This stage focuses on reconstructing ordered processes with reasons. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s process map. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid memorising labels while losing transitions. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
18. Write Unit Questions
The question-design job. This stage focuses on keeping quantities and units attached. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s speed notes. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid checking arithmetic while ignoring incompatible units. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
19. Write Condition Questions
The question-design job. This stage focuses on asking when a rule applies. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s formula sheet. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid retrieving the formula with no idea when to use it. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
20. Write Mixed Sets
The question-design job. This stage focuses on combining taught alternatives so selection matters. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s three question families. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid mixing untaught material and calling the confusion desirable difficulty. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
21. Build a Question Ladder
The question-design job. This stage focuses on moving from close retrieval to wider transfer. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s five-stage practice set. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid making every later question harder in several dimensions at once. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
22. Build Answer Keys
The question-design job. This stage focuses on making the checking route explicit. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s worked solution. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid writing only the final answer for a reasoning task. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
23. Build Rubrics for Open Responses
The question-design job. This stage focuses on naming the features a response should demonstrate. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s comparison paragraph. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid inventing an unofficial mark scheme and treating it as authoritative. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
24. Verify Numerical Answers
The question-design job. This stage focuses on checking by substitution, alternate method or source. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s equation set. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid trusting a generated key because it is formatted confidently. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
25. Verify Factual Answers
The question-design job. This stage focuses on checking claims against trustworthy course sources. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s science quiz. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid repeating an unverified summary across many questions. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
26. Use Distractors Carefully
The question-design job. This stage focuses on making wrong options represent plausible distinctions. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s multiple-choice set. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid creating silly distractors that turn the question into recognition by elimination. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
27. Reduce Cueing
The question-design job. This stage focuses on removing hints as the learner becomes more independent. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s scaffolded questions. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid counting heavily prompted success as independent retrieval. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
28. Use Fresh Variants
The question-design job. This stage focuses on changing surface features while preserving the concept. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s percentage questions. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid changing the underlying method without noticing. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
29. Retest Repairs
The question-design job. This stage focuses on using a new question after feedback. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s corrected inference. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid repeating the identical corrected answer immediately. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
30. Track Question Familiarity
The question-design job. This stage focuses on distinguishing fresh, seen and repeated items. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s question bank. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid calling a repeated set an unseen test. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
31. Interpret Self-Test Scores
The question-design job. This stage focuses on keeping denominator, conditions and scope visible. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s 8-of-10 quiz. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid turning a small home quiz into a predicted exam grade. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
32. Diagnose Wrong Answers
The question-design job. This stage focuses on separating knowledge, selection, interpretation and execution failures. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s mixed set. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid putting every error into one ‘weak topic’ bucket. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
33. Diagnose Right Answers
The question-design job. This stage focuses on checking whether success came from knowledge, cues or guessing. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s multiple-choice result. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid assuming every correct option proves durable understanding. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
34. Schedule Returns
The question-design job. This stage focuses on bringing failed or fragile questions back after useful delay. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s revisit queue. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid retesting everything immediately until it feels easy. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
35. Use AI to Draft Questions
The question-design job. This stage focuses on asking for variants, contrasts and explanations within a checked scope. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s AI-generated quiz. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid treating generated material as official or automatically correct. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
36. Audit AI Answers
The question-design job. This stage focuses on verifying generated keys before repeated practice. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s generated algebra set. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid letting one fluent error propagate into dozens of cards. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
37. Protect Official Materials
The question-design job. This stage focuses on keeping generated and teacher-made practice distinct from real papers. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s paper folder. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid labelling home-made questions as past-paper questions. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
38. Build a Weekly Self-Test
The question-design job. This stage focuses on sampling old and current learning without consuming the whole week. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Tricia’s Friday check. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid testing only what was studied most recently. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
39. Build a Final-Month Question Bank
The question-design job. This stage focuses on prioritising recurring errors, mixed demands and fresh items. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Kai Kai’s final-month bank. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid collecting hundreds of questions with no selection logic. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
40. Hand Off to Exam Simulation
The question-design job. This stage focuses on knowing when self-tests should yield to matched papers and realistic conditions. Begin with the learning claim, not with a fashionable question format. Ask what a learner would have to produce, choose, explain or apply if the underlying knowledge were genuinely available. The question should make that capability visible while withholding only the support that the learner is ready to do without.
Worked case. Consider Alicia’s final transition. Turn the note into a prompt that requires the learner to reconstruct or use the important relationship. Then write the expected answer before giving the question to the learner. This exposes ambiguity early: if two different answers could both satisfy the wording, either clarify the prompt or decide explicitly that multiple defensible responses are allowed.
Alicia, Tricia and Kai Kai vary the demand. One version asks for direct recall. A second asks why the idea works or what condition it depends on. A third changes the surface context and asks the learner to select the idea without being told its name. These versions are not automatically ordered from easy to hard for every learner. They sample different parts of usable knowledge.
Build the checking route. A self-test needs a trustworthy way to inspect the response. For a numerical problem, verify the calculation and reasoning. For a factual claim, use the checked course source. For an open response, identify the features that matter and preserve room for valid alternative wording. Do not invent an official mark allocation unless you are actually using the applicable official scheme.
Interpret the result carefully. A wrong answer may reflect unavailable knowledge, misreading, method selection, an incomplete explanation or an execution slip. A correct answer may reflect genuine retrieval, a visible cue, answer familiarity or guessing. Record enough context to choose the next task. The purpose of the self-test is not to generate a dramatic score; it is to improve the next learning decision.
Failure mode. Avoid pretending a custom quiz can replace the actual assessment format. Repair the design before adding more volume. A hundred questions with the same cue can create impressive completion totals while testing one narrow response repeatedly. A smaller set that changes representation, removes cues and asks for reasoning can reveal more about what the learner can actually do.
Progression. After feedback, use a fresh but interpretable variant. If the learner succeeds, widen delay, context or mixing one dimension at a time. If the learner fails, simplify until the first unstable decision becomes visible. The question bank should therefore evolve with evidence: explanation questions where understanding is weak, discrimination where methods are confused, and transfer where familiar forms are already stable.
41. Worked laboratory: turn one percentage note into six different questions
Start with the checked note: “After an increase of 20 percent, final amount = original amount × 1.20.” A direct recall question asks for the multiplier. An explanation question asks why 1.20 represents the new amount. A calculation question gives an original amount of 75 and asks for the final amount, which is 90. These three questions already demand different outputs from one short note.
Now add a reverse question: an amount becomes 90 after a 20 percent increase; find the original. The learner should recognise that 90 represents 120 percent of the original and calculate 90 ÷ 1.20 = 75. A contrast question asks why subtracting 20 percent of 90 does not reverse the original increase. Twenty percent of 90 is 18, producing 72, because the second percentage uses a different base.
A transfer question changes the surface context: a subscription price after a 20 percent increase is 54 units. What was the original price? The relationship remains final = original × 1.20, so the original is 45. The new nouns do not create a new mathematical method. This is a useful kind of variation because it asks whether the learner follows the relationship rather than memorising the earlier numbers.
Notice what has not happened. The self-test has not become an official examination paper, and six correct answers do not prove mastery of every percentage problem. The set samples recall, explanation, forward application, reverse application, discrimination and contextual transfer. Its strength comes from the deliberate changes in demand, not from the number six.
If Kai Kai misses only the reverse and contrast questions, the repair should target the base relationship rather than assigning another twenty forward-percentage calculations. If he misses the arithmetic but explains the relationship correctly, the next task is different. The question set becomes diagnostic because its items were designed to separate plausible mechanisms.
42. Worked laboratory: turn a reading note into evidence questions
Use this original note: “An inference should be supported by details from the text, and the strength of the claim should not exceed the evidence.” Now read: “Aisha placed the finished poster against the wall, checked the clock twice and moved the presentation cards beneath her notebook when the door opened.” A recall question about the note would ask what an inference needs. A better application question asks what the passage suggests and which details support that interpretation.
A defensible answer might say that Aisha appears uneasy or hesitant about presenting despite having prepared materials. The finished poster supports preparation; checking the clock and hiding the cards can support hesitation. A question-design error would be to ask, “Why is Aisha terrified of public speaking?” The passage does not establish terror or a general condition. The wording has already overclaimed before the learner answers.
Write a discrimination question: which claim is better supported—“Aisha has done no work” or “Aisha seems reluctant to begin presenting”? Ask the learner to reject the weaker claim using evidence. This tests not only selection of evidence but control of claim strength. Then create a fresh passage where preparation and willingness point in different directions again.
For the answer key, list the reasoning features rather than one compulsory sentence: a plausible inference, relevant textual details and an explanation connecting the details to the claim. If the course has an applicable official marking scheme, use that for official scoring. The home-made rubric here is a learning aid for inspecting reasoning, not an invented examination authority.
43. Worked laboratory: turn a science note into a data self-test
Start with the note: “A pattern in observations can support a description, but a causal explanation requires evidence that addresses alternative causes.” Create an original dataset: four equal time intervals show totals of 10, 18, 25 and 31 units. Ask the learner to calculate the increases: 8, 7 and 6. Then ask what can be described without inventing a cause.
A sound description is that the total rises at each recorded interval while the amount added between intervals decreases across these observations. Now offer the claim, “The total rises more slowly because the equipment is wearing out.” Ask whether the numbers establish that explanation. They do not. No evidence about equipment condition has been supplied.
A fresh variant changes the context but preserves the reasoning: cumulative pages read are 5, 11, 16 and 20. The additions are 6, 5 and 4. Ask the learner to distinguish the rising cumulative total from the falling additions. Then ask for one causal claim that would require additional evidence. The self-test now checks transfer of the data relationship rather than memory of the first numbers.
Do not make every science question a scepticism exercise where the correct response is always “cannot conclude.” Include tasks where the supplied design and evidence do support the requested comparison at the level being studied. Question banks need contrasts so that learners practise judging evidence, not merely repeating a safe phrase.
44. Build a 20-question self-test without making twenty versions of one question
Choose four important learning claims. For each, write one direct retrieval question, one explanation or contrast question, one application question, one changed-representation question and one fresh transfer question. The resulting twenty items have a visible design: five different demands applied across four pieces of learning. This is more informative than generating twenty near-identical definition prompts.
Shuffle the order only after the questions are checked. During first learning, grouping can help the learner understand a method. During later selection practice, mixing can make the learner decide which idea applies. The same bank can therefore serve different purposes depending on how it is presented, but the result should be labelled accordingly.
Before use, answer every item or obtain a trustworthy answer source. Flag open questions that permit several defensible responses. Remove accidental cues such as chapter names, repeated wording or an answer embedded in a later question. Decide whether calculators, notes or other supports are allowed for this particular self-test.
After the attempt, do not simply calculate 17/20 and move on. Identify which demands produced the three errors. Three missed transfer questions tell a different story from three arithmetic slips across otherwise sound reasoning. The score is a summary; the pattern is what designs the next practice.
45. Final operating rule: questions should reveal the next learning decision
A self-test is successful when it gives the learner useful evidence. Sometimes that evidence is encouraging: an idea remains available after delay and survives a new context. Sometimes it is inconvenient: a definition is remembered but cannot be applied, or a familiar method is chosen whenever a surface cue appears. Both outcomes are valuable when they lead to a more precise next task.
Alicia’s question bank becomes smaller and more varied. Tricia stops treating one model answer as the only acceptable wording and writes feature-based checking guides. Kai Kai records which items are fresh, which are repeated and which required support. Their scores become easier to interpret because the conditions and question jobs are visible.
Use How to Organise Revision Notes for Exams Without Rewriting the Textbook when the source material itself is still unwieldy. Use How to Check Revision Resources Against Your Exam Syllabus when the accuracy or course fit of the source is uncertain. When realistic examination conditions become the main job, hand the learner to matched papers and the existing examination-performance system.
The best practice question is not the one that looks most like an exam question. It is the one whose purpose is clear, whose answer can be checked, whose difficulty can be interpreted and whose result tells you what the learner should do next.
46. A seven-day question-building workshop
On day one, select five checked notes and write the learning claim behind each one. Do not write questions yet. Ask what the learner should eventually be able to retrieve, explain, distinguish or apply. If the note itself is ambiguous or unsupported, return to the source before converting the ambiguity into repeated practice.
On day two, write one direct retrieval question for each claim and answer it. On day three, convert each into a relationship, contrast or “why” question. On day four, create one application or changed-representation version. The purpose of the sequence is not to make fifteen questions compulsory; it is to practise seeing that one piece of knowledge can support several different kinds of cognitive work.
On day five, audit the answers. Check calculations independently, verify factual claims against the appropriate course source and identify where open responses permit more than one defensible wording. Remove accidental clues. If a later question reveals an earlier answer, reorder or rewrite the set. If a multiple-choice distractor is obviously absurd, replace it with an option representing a plausible misunderstanding.
On day six, attempt a selection of the questions without the notes. Record which items were fresh, what support was allowed and which errors occurred. Do not immediately retest every missed item until the answer becomes familiar. Repair the mechanism, then create or reserve a fresh variant for a later check. On day seven, review which question types produced useful evidence and which merely consumed time.
Alicia may discover that direct recall is already strong and needs fewer cards. Tricia may find that explanation questions reveal unsupported leaps that definition quizzes never exposed. Kai Kai may discover that mixed method-selection questions are more informative than another blocked worksheet. Their future banks should therefore diverge. A question system is adaptive when the evidence changes what gets written next.
The workshop is an illustrative onboarding sequence, not a validated seven-day protocol. Compress it into one session or spread it over several weeks as appropriate. The governing principle is stable: write from a checked learning claim, design a question whose output matches that claim, verify the checking route, observe the learner’s response and let the result determine the next question.
47. What a self-test can and cannot prove
A self-test can show what happened on its selected items under its stated conditions. It can reveal that a learner retrieved a definition after a delay, selected a method among alternatives, explained a relationship, or failed at a particular transfer. It can help locate recurring errors and decide what to practise. These are valuable uses precisely because the claims remain close to the evidence.
A self-test cannot automatically establish an official grade, predict the exact examination, prove complete syllabus coverage or reproduce the psychometric properties of a formal assessment. A bank created from the learner’s own notes may overrepresent what the learner chose to record. Generated questions may repeat patterns. Familiar items may inflate apparent fluency. These limits do not make self-testing useless; they define its responsible role.
When the learner needs broader evidence, widen the sample and conditions. When the learner needs examination-format evidence, use appropriately matched official or authorised materials. When the learner needs repair, return to targeted questions. Preparation works best when each tool is allowed to answer the question it was designed to answer rather than being promoted into a universal measure of readiness.
48. The final question before adding another question
Before expanding the bank, ask what uncertainty the new item will reduce. If ten existing questions already show that a learner can recall a definition, an eleventh near-duplicate may add little. A question that asks for a contrast, a changed representation or an application may reveal a more important boundary. Volume should follow a reason.
Also ask whether the learner needs a question at all. A persistent misunderstanding may need explanation. A writing weakness may need feedback. A timing problem may need realistic execution. A missing prerequisite may need repair before another mixed test. Self-testing is powerful because it makes knowledge observable, but observation alone does not teach every missing capability.
Keep the bank lean enough that fresh items remain available and rich enough that important decisions can be sampled. Retire questions whose answers have become memorised when freshness matters, while retaining them for ordinary retrieval if repetition still serves a useful purpose. Label the role instead of discarding useful history.
The final discipline is simple: every question should have a job, every answer should have a checking route, and every result should have a possible consequence for what happens next. That is how revision notes stop being stored information and become a controlled practice system.