← Open the interactive adventure
A complete study guide with 18 chapters, six realistic missions and practical exercises for learning with AI while keeping your own thinking visible.
Use your browser’s Print command to print this guide or save a PDF. All mission situations and data are fictional. Work through decisions before reading their consequences. Written work is self-assessed; completion is not an externally assessed qualification.
Research / CHAPTER 1
Separate the decision you need to make from the evidence that would inform it.
A broad question such as “Which programme works best?” hides the outcome, audience and conditions. Name what success means and what comparison would matter. Then distinguish an answerable question from an assumption or preference. AI can help generate questions, but it cannot supply missing observations by writing confidently. If the evidence you need does not exist, plan a small way to collect it or make a more limited decision.
A club wants to choose a workshop format. The useful question is not whether online learning is universally better, but whether its own beginners can complete a defined task with the available support.
Write your decision, one measurable outcome and the evidence that could change your mind.
Gathering a large pile of interesting information that does not answer the decision question.
Research / CHAPTER 2
Compare sources by what they measured, not only by which conclusion sounds strongest.
Record the population, task, outcome, collection method and limitation for each source. Sources can disagree because they asked different questions. Before resolving a conflict, check whether there is a conflict at all. A source may be credible yet irrelevant to your precise claim. Keep your table short enough that you can explain why each source is included.
One fictional report counts attendance, while another tests independent performance. A high attendance rate does not contradict a low completion rate; they describe different outcomes.
Compare two supplied or independently verified sources using population, task, outcome and limitation.
Treating two percentages as directly comparable without checking their denominators or definitions.
Research / CHAPTER 3
Make the strength of your wording match the strength and scope of your evidence.
Distinguish observed results, possible explanations and recommendations. “In this group” is different from “for everyone.” “Associated with” is different from “caused by.” Specific limits make a statement more useful, not necessarily weaker. If evidence is missing, name the missing piece and explain whether it prevents a decision or merely suggests a reversible next step.
Eight of ten volunteers finished a task after receiving support. You can report completion in that group. You cannot infer that every future learner will finish, or that the support alone caused the result.
Write a conclusion in three sentences: observation, limitation and next step.
Using cautious-sounding words while keeping an unsupported universal claim underneath.
Diagnose / CHAPTER 4
A useful problem report states the input, expected result, actual result and conditions.
Replace “it does not work” with a small example someone else can repeat. Remove unrelated details, but keep conditions that might affect the failure. If AI suggests a cause, treat it as a hypothesis until a test distinguishes it from alternatives. Reproducibility turns troubleshooting from a stream of guesses into a sequence of informative checks.
A training handout works for facilitators but not new learners. The report identifies the exact step where beginners stop, the instruction they see and the action they think is expected.
Write a failure report with one concrete input, an expected outcome and what happened instead.
Changing several things before recording the original failure, then losing the ability to explain the fix.
Diagnose / CHAPTER 5
The best next test is one whose result would help you distinguish plausible causes.
List two or three hypotheses and predict what you would observe if each were true. Prefer a small, reversible test. A test that produces the same result under every hypothesis may be easy but uninformative. After the test, update your explanation rather than collecting only confirmations. Keep uncertainty where the result cannot separate the remaining causes.
If learners fail either because instructions are unclear or because a prerequisite is missing, ask them to explain the instructions before solving the task. This checks interpretation separately from performance.
Write two possible causes and a test with different predicted outcomes for each.
Running a familiar test repeatedly even though it cannot discriminate between your explanations.
Diagnose / CHAPTER 6
A repair should solve the original failure without introducing a new one nearby.
Use a small test set: the original case, a normal case and a boundary case. Explain what the change affects and what remains untested. Do not mistake one successful run for a general guarantee. If a new failure appears, inspect whether the fix changed a hidden assumption. Keep a simple fallback when the cost of a failed rollout is high.
A revised checklist handles missing answers, but it also rejects a valid answer of zero. A boundary test reveals that “blank” and “zero” were treated as the same value.
Name an original, normal and boundary case for a repair you propose. State the expected result of each.
Testing only the example that motivated the change and overlooking adjacent cases.
Decide / CHAPTER 7
A constraint determines whether an option is feasible. A preference helps compare feasible options.
Check non-negotiable requirements first. Only then compare convenience, appearance or other preferences. If every option fails a constraint, make the conflict visible instead of inventing a winner. Ask which requirement, resource or deadline can change. A decision can be useful even when its conclusion is “we need one more answer before proceeding.”
A venue with excellent equipment cannot meet a workshop requirement if it is unavailable at the required time. Its attractive features do not offset the feasibility failure.
List two constraints and three preferences for a real decision. Check whether you have accidentally mixed the categories.
Adding up preference points until an infeasible option appears to win.
Decide / CHAPTER 8
A recommendation is more useful when you know what would make it change.
Identify the uncertain input that matters most. Compare the options under a plausible alternative value. If the preferred option changes easily, the decision is sensitive: seek better evidence, use a pilot or make a reversible choice. Do not report invented precision as certainty. Explain which assumptions carry the recommendation.
A team prefers a longer session only if most learners can attend the full slot. If attendance availability is unknown, a short availability check may be more valuable than refining the slide deck.
Name one uncertain assumption and rewrite your recommendation under the opposite plausible condition.
Keeping the same recommendation after changing the assumption that originally justified it.
Decide / CHAPTER 9
When evidence is limited, choose a small trial with a learning question and an exit rule.
A pilot is not simply a small launch. Specify what it will test, who will participate, how you will judge the result and what happens next. Include a stopping or revision condition. Keep the trial’s size proportional to the decision it can inform. A successful pilot may justify a larger test, not necessarily immediate full adoption.
A club trials one new workshop format with a small group, checks an independent task and asks about accessibility before scheduling the rest of the series.
Write a pilot question, success criterion, revision trigger and next decision.
Calling a trial successful because it ran, without measuring the outcome that motivated it.
Teach / CHAPTER 10
A clear explanation connects a concrete example to the general principle without confusing them.
Begin with a situation the learner can inspect. Then name the rule that explains it. Check which details belong to the example and which are essential to the principle. Ask the learner to explain the rule using a different example. AI can propose analogies, but you should test where the analogy stops matching the concept.
A recipe analogy can illustrate a procedure, but it may hide cases where a process requires feedback and revision. State that limitation rather than forcing every detail to fit.
Explain a concept first with a small example, then in two general sentences. Add one limitation of the example.
Assuming an entertaining analogy is automatically accurate or sufficient.
Teach / CHAPTER 11
A diagnostic question should expose reasoning, not only whether someone can recognise a familiar phrase.
Choose a tempting wrong answer that reflects a specific misconception. Ask for the reason, not just the option. A correct selection with an incorrect explanation may indicate guessing. Feedback should address the underlying rule and lead to a fresh attempt. Avoid trick wording that measures reading endurance rather than the concept.
To test understanding of percentage change, ask whether a 20% rise followed by a 20% fall returns to the start, and require an example. The reasoning reveals whether the learner tracks the base.
Write a diagnostic question, a plausible misconception and feedback that targets it.
Writing distractors so absurd that learners can select the right answer without understanding the topic.
Teach / CHAPTER 12
Feedback becomes useful when the learner knows what to change and gets another chance to try.
Identify one consequential gap, connect it to a criterion and suggest a small action. Preserve accurate parts of the attempt. Ask the learner to make the revision rather than simply accepting your replacement. Then use a new example to see whether the correction transferred. Tone matters, but encouragement without information cannot explain how to improve.
Instead of “unclear summary,” say “your conclusion names all attendees, but only respondents supplied views; revise the population phrase and explain the denominator.”
Write feedback on an imperfect example: one strength, one specific gap and one next attempt.
Correcting everything at once or rewriting the entire artifact so the learner cannot see the key change.
Review / CHAPTER 13
Treat feedback as proposals to inspect against your goal, evidence and constraints.
AI may improve wording while changing facts, removing uncertainty or adding requirements. Compare each important suggestion with the original brief. Adopt, adapt or reject it with a reason. If you cannot evaluate a technical suggestion, create a small test or consult a suitable original reference before relying on it. Record one rejected suggestion as evidence of your judgment.
An assistant changes “we recommend” to “we have approved.” The revised sentence is smoother but factually different, so the learner restores the decision status.
Take three feedback suggestions and label them adopt, adapt or reject, with a reason for each.
Assuming a more polished version is always a more accurate one.
Review / CHAPTER 14
Ask who or what is missing from your examples, data and proposed workflow.
A process can work for the typical case while failing for a legitimate variation. Identify cases that matter to the outcome: absent information, changed circumstances, different access needs or repeated requests. You do not need every imaginable exception, but you should explain the boundaries of your test set. Use fictional data for early checks where it is sufficient.
An exercise assumes every learner has a personal device. A paper option preserves the core task when that assumption does not hold.
List three relevant cases absent from your first draft and add a test or fallback for each.
Confusing a large set of similar examples with broad coverage of different failure modes.
Review / CHAPTER 15
A handoff needs enough context for someone to operate the result without guessing your intentions.
Include purpose, inputs, essential steps, expected outputs and known limits. Distinguish instructions from background explanation. Ask a reader to perform a small task using only the handoff, then observe where they hesitate. Those gaps are more informative than a vague request to say whether the document looks clear.
A checklist says “confirm the result” but does not define confirmation. A better version names the expected total, how to calculate it and where to record a discrepancy.
Write a five-line handoff and ask what a new user could misunderstand in each line.
Leaving crucial decisions in your head because the workflow feels obvious to its creator.
Independence / CHAPTER 16
An improvement claim needs to distinguish your own ability from the help you received.
Record task difficulty, time, hints and access to reference material. A faster result with extensive assistance may be useful productivity evidence, but it is not the same as unaided skill. Choose a comparable fresh task for an independent check. Report both what you can do and where support still helps. This gives a more actionable picture than a single completion score.
A learner produces a strong report with AI editing, then writes a shorter report on new data without editing support. The two attempts answer different questions about productivity and independence.
Compare two attempts and list the support available in each. State one conclusion the comparison does and does not justify.
Calling every improvement in an assisted artifact an equal improvement in independent capability.
Independence / CHAPTER 17
Look for evidence that could change your current view, not only examples that reinforce it.
Write your best explanation and one plausible alternative. Identify an observation that would be difficult for your favourite explanation to account for. Try to obtain that observation through a relevant test or source. If the result is mixed, narrow your claim. This is an ongoing habit of revision, not a requirement to distrust every useful conclusion.
A facilitator thinks shorter instructions caused better task completion, but the second group also had more prior experience. A new comparison should consider both explanations.
Write your preferred explanation, a rival explanation and a discriminating observation.
Collecting more examples from the same narrow situation and calling the repetition independent confirmation.
Independence / CHAPTER 18
Demonstrate the skill across a small set of meaningful variations and explain its limits.
Choose a near example, a changed constraint and a case where the usual method should not be applied directly. Record your reasoning and any help you needed. The portfolio should show decisions, corrections and boundaries, not just finished outputs. Use it to choose the next practice target rather than to claim expertise across an entire field.
A learner interprets one survey, compares two surveys with different response rates and declines an unsupported claim about people who never responded. The boundary case is part of the evidence.
Choose three variations for your capstone and define what a sound response would include.
Treating one successful familiar task as proof that every harder or different task will be easy.
MISSION 1 / Research synthesiser
Two fictional reports inform a training decision. One measures attendance and the other measures task completion. Build a conclusion without pretending they answer the same question.
A comparison table and a three-sentence recommendation.
Name each outcome and population. Avoid a causal or universal claim. Identify the next useful evidence.
Report A: 24 of 30 invitees attended an online session. Report B: 12 of 20 attendees at a separate in-person session completed an independent task.
Which comparison is valid?
Write your choice and reason before reading the consequences below.
Practice: Calculate both percentages and write what each does and does not measure.
The organiser wants beginners to complete a specific task without help. Time and access conditions differ between the two sessions.
What should you seek next?
Write your choice and reason before reading the consequences below.
Practice: Write a fair comparison plan with one outcome and three conditions to record.
Only one small pilot can be run before the next planning meeting.
What is the strongest recommendation?
Write your choice and reason before reading the consequences below.
Practice: Write the observation, limitation and pilot recommendation in three sentences.
A pilot reports 9 completions out of 12 beginners, but four used extra hints. What additional breakdown would help interpret independent performance?
Report 75% total completion without labelling it fully independent. Ask which completions used hints; total hints and total completions alone do not identify the unaided completion count.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 2 / Troubleshooting partner
A fictional worksheet works for its designer, but new learners stop at the first calculation. Diagnose the difficulty before rewriting everything.
A reproducible failure report, a discriminating test and a revised instruction.
Distinguish understanding the instruction from knowing the calculation. Test more than one case.
The worksheet says “normalise the values” without defining the operation. Beginners interpret that phrase differently.
What should you record first?
Write your choice and reason before reading the consequences below.
Practice: Write a failure report for fictional values 2, 3 and 5, where the intended operation is dividing each by their total.
You suspect either unclear wording or unfamiliarity with division.
Which test is most informative?
Write your choice and reason before reading the consequences below.
Practice: Predict what you would see under each of the two explanations.
Your new instruction says to divide every value by the total. A new test uses 0, 0 and 0.
What should the worksheet do?
Write your choice and reason before reading the consequences below.
Practice: Write the revised instruction and expected results for a normal and zero-total case.
A later input is -2, 1 and 1. Its total is zero although not every value is zero. Revise any check that looked only for all-zero inputs.
Test the actual total, not merely whether every value equals zero. Also ask whether negative values are meaningful in the intended application.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 3 / Decision designer
A fictional team must choose a venue for a 20-person workshop on Tuesday. Compare feasibility, preferences and uncertain demand.
An options brief with constraints, a conditional recommendation and a pilot question.
Check feasibility before preferences. Make assumptions explicit. Avoid inventing availability.
Room A seats 24 and is available Tuesday but lacks a projector. Room B seats 30 and has a projector but is available only Wednesday. Tuesday is fixed; paper activities can meet the goal.
Which conclusion follows?
Write your choice and reason before reading the consequences below.
Practice: Separate the fixed constraints from preferences and write a provisional recommendation.
The team now expects between 18 and 28 people. Room A seats 24.
What must change in your reasoning?
Write your choice and reason before reading the consequences below.
Practice: Write the condition under which A works and one fallback if demand exceeds it.
The team wants to repeat the workshop monthly but has not measured learning or demand.
What should the first session test?
Write your choice and reason before reading the consequences below.
Practice: Write a pilot success criterion and a reason you would revise rather than repeat the same plan.
The pilot fills all 24 seats but only 10 learners attempt the exit task. Explain what the attendance result establishes and what remains unknown about learning.
A full room supports a demand observation for that event. It does not establish skill acquisition for all attendees; investigate why only ten attempted the check.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 4 / Peer teacher
A fictional peer session looks successful because everyone agrees with the explanation. Design a check that reveals reasoning and supports another attempt.
A two-level explanation, a diagnostic question and targeted feedback.
Use an accurate example. Ask for reasoning. Follow feedback with a fresh task.
You explain that increasing and then decreasing by the same percentage need not return to the starting value. Learners nod.
What is the useful next action?
Write your choice and reason before reading the consequences below.
Practice: Calculate a 10% rise and then a 10% fall from 200. Explain the result.
A learner says the result must be 200 because “plus ten minus ten is zero.”
What feedback targets the misconception?
Write your choice and reason before reading the consequences below.
Practice: Write a short feedback response that preserves the learner’s correct first step and targets the second step.
The learner now repeats your explanation accurately.
Which follow-up provides stronger evidence?
Write your choice and reason before reading the consequences below.
Practice: Create and solve a fresh rise-then-fall example; state what the check can and cannot show.
A learner asks whether a 10% decrease followed by a 10% increase also ends at 198 when starting from 200. Check the claim and distinguish final value from intermediate values.
200 becomes 180 and then 198. Multiplying by 0.9 and 1.1 gives the same final factor here, although the intermediate amounts differ.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 5 / Critical reviewer
An AI assistant improves a fictional team handoff but introduces unsupported facts and overlooks an edge case. Review the revision rather than accepting its polish.
A reviewed handoff, a feedback decision log and a small case checklist.
Preserve factual status. Explain adopted and rejected changes. Make instructions usable for a new reader.
Original: “We recommend testing the new checklist with three volunteers.” AI revision: “The new checklist has been validated by the team.”
What should you do?
Write your choice and reason before reading the consequences below.
Practice: Write the corrected sentence and a reason for rejecting the AI change.
The checklist says “reject any entry without a score.” In this exercise zero is a valid score and an empty field means not yet assessed.
Which test matters?
Write your choice and reason before reading the consequences below.
Practice: Write the expected handling of zero, five and blank.
The handoff ends with “check everything carefully.” A new volunteer asks what to do.
What is a useful replacement?
Write your choice and reason before reading the consequences below.
Practice: Write a short handoff for entering and reviewing scores.
A new field allows the text “not applicable.” Adapt the instructions without silently treating it as zero or as an unassessed blank.
Keep measured zero, unassessed blank and not-applicable status distinct. Define whether and how each belongs in any summary calculation before calculating averages.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 6 / Independent practitioner
A fictional team sees better performance after changing instructions, but the second group had more experience. Build a defensible interpretation and a transfer plan.
A bounded conclusion, a rival explanation and a three-case independent portfolio plan.
Record support and task conditions. Consider a rival explanation. Demonstrate boundaries as well as success.
Group A: 6 of 10 complete a task after the old instructions. Group B: 9 of 10 complete after revised instructions, but B has more prior experience.
What follows?
Write your choice and reason before reading the consequences below.
Practice: Write the rates, their difference in percentage points and a causal limitation.
The facilitator believes the instructions explain the whole difference.
What would improve the investigation?
Write your choice and reason before reading the consequences below.
Practice: State the preferred and rival explanations, then propose an informative next observation.
You want to demonstrate your own skill in evaluating comparisons.
Which portfolio is most informative?
Write your choice and reason before reading the consequences below.
Practice: Design three short assessment cases and the criteria for a sound answer to each.
One group has a higher completion rate but received unlimited hints. The other had no hints. Write a conclusion that recognises useful assisted performance without equating it with unaided capability.
Report the measured results and the difference in support. Explain that the comparison does not directly rank independent skill; propose a comparable unaided check if that is the intended outcome.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
Choose a useful artifact from your own work, study or hobby. Record these nine parts:
Review purpose, accuracy, reasoning, transfer and ownership. The interactive tool can save this evidence and export a portfolio.