Primary 6 advanced vocabulary for probability, uncertainty, risk and decision-making helps learners explain what can happen, how likely it is, what remains unknown and how a sensible choice changes when the consequences matter. This Grade 6 probability vocabulary manual develops words for outcomes, events, sample spaces, likelihood, experimental probability, expected results, scenarios, risk, trade-offs and decisions under uncertainty. The aim is not to make chance sound mysterious. It is to give learners enough language to distinguish possible from probable, one surprising outcome from a bad model, and a sensible precaution from the impossible demand for zero uncertainty.

Readers searching for 6th grade probability vocabulary, chance and likelihood words, sample space and outcomes, experimental probability, risk vocabulary for students and decision-making words often meet isolated definitions. This collection connects the vocabulary to actual reasoning. A coin can land heads even when heads has only a one-half probability; a rare outcome can occur without making the earlier probability wrong; and a decision can be reasonable even when it does not produce the best outcome every time. Those distinctions matter in mathematics, science, reading comprehension, planning and everyday judgement.

This world-facing Primary 6 / Grade 6 uncertainty and decision vocabulary lane is deliberately narrower than eduKateSingapore’s general statistics and reasoning owners. Volume 6 owns charts, tables and data interpretation. Volume 7 owns experimental design and scientific conclusions. Volume 8 owns the language of uncertain outcomes, probability models, risk, scenarios, expected consequences and defensible choices. Alicia, Tricia and Kai Kai are fictional learners; all games, forecasts, decision packets and classroom cases are original teaching material rather than reports of actual pupils or advice for financial, medical or safety-critical decisions.

Primary 6 Advanced Vocabulary Collection · Volume 8. The article uses simple mathematical probability alongside careful English. Fractions, percentages and ratios appear where they clarify likelihood, but the deeper objective is explanation: what is the event, what is the reference set, which assumptions make the model work, and what does the probability statement allow a reader to expect? The collection is an optional extension rather than an official prescribed list or examination prediction.

Choose the uncertainty problem before choosing the vocabulary

Use entries 1–10 for outcomes, events and sample spaces. Use entries 11–20 for certain, impossible, likely and unlikely. Later groups develop probability measures, experiments and simulation, combined events and dependence, variation and forecasting, risk and consequence, decision language, judgement under uncertainty and probability writing actions.

For complete application, use the later probability and decision laboratories. Each packet states its rules fully in text, so no missing spinner, deck or graph has to be imagined. The original story shows how a decision can be sensible even when the realised outcome is disappointing. The teaching sequence then moves from vocabulary recognition to explanation, comparison and revision.

Probability describes uncertainty; it does not promise one result

Suppose a fair fictional bag contains three blue counters and one red counter. The probability of drawing blue is three quarters. That does not mean every group of four draws must contain exactly three blue results. Nor does one red result prove the model was wrong. Probability describes the structure of uncertain outcomes under stated assumptions. Individual outcomes remain uncertain unless their probability is zero or one.

A decision uses that uncertainty rather than pretending it disappears. If one option has a small chance of a serious loss and another has a larger chance of a mild inconvenience, the relevant words include probability, consequence, risk, trade-off and criterion. A sensible choice depends not only on which outcome is most likely, but also on what the outcomes mean. This article teaches that vocabulary through fictional classroom cases, not through recommendations about real investments, medical choices or hazardous activities.

Entries 1–10: name what can happen before discussing how likely it is

1. Outcome

An outcome is one possible result of a chance process or trial. When a fair six-sided die is rolled once, obtaining a four is an outcome. In a spinner with sectors labelled A, B and C, landing on B is an outcome. The word names the result of one trial, not the whole set of possibilities.

An outcome can be simple or detailed. “Blue” may be an outcome if colour alone is recorded; “blue and number 3” may be an outcome in a two-feature experiment. The recording rule determines what counts as distinct. A learner should not count several descriptions of the same result as separate outcomes unless the model actually distinguishes them.

Use possible outcome, observed outcome and one outcome of the trial. Alicia writes the recording rule first. For practice, compare a die-roll experiment that records only parity with one that records the exact face. Explain why the first has two recorded outcomes while the second has six.

2. Event

An event is a set of one or more outcomes that satisfy a stated condition. On a six-sided die, “roll an even number” is the event containing outcomes 2, 4 and 6. “Roll a six” is an event containing one outcome. Events are useful because many questions concern conditions rather than one exact result.

Do not treat event as a synonym for one trial. The trial is the act of rolling; the event is the condition evaluated in its outcome. Several events can be true for the same outcome. Rolling a six satisfies both “even” and “greater than four”. Overlap between events becomes important later.

Use event of interest, the event occurs and outcomes included in the event. Tricia lists the outcomes before calculating. For practice, write the event “number greater than three” as a set of die outcomes and explain why four, five and six belong while three does not.

3. Sample space

A sample space is the full set of possible recorded outcomes under the model. For one fair six-sided die, the sample space is {1, 2, 3, 4, 5, 6}. For one coin whose sides are recorded as H and T, the sample space is {H, T}. The set defines the reference from which events are formed.

A sample space must match the experiment. If two coins are tossed and order matters, HH, HT, TH and TT are distinct ordered outcomes. Writing only “zero, one or two heads” is a different sample space that records the count of heads rather than the exact sequence. Both can be useful, but their outcome probabilities are not automatically equal.

Use complete sample space, outcome in the sample space and define the sample space. Kai Kai checks that nothing possible is omitted and nothing impossible added. For practice, list the sample space for drawing one letter from the word LEVEL when positions are equally likely but the recorded outcome is the letter shown. Explain why repeated letters affect probability even though the recorded labels are not all distinct.

4. Trial

A trial is one performance of a chance experiment. Tossing a coin once is one trial; rolling a die once is one trial; drawing and replacing one counter can be one trial. Repeating trials provides a sequence of observed outcomes that can be compared with a probability model.

Trials need consistent rules if their results are to be combined. Drawing a counter and replacing it before the next trial is different from drawing without replacement because the composition of the bag changes in the second procedure. A learner should not combine those outcomes as though the probability stayed constant unless the model justifies it.

Use repeat the trial, one outcome per trial and independent trials only where independence is justified. Alicia records each result before the next draw. For practice, explain what part of a counter experiment must be reset between trials if the intention is to keep the probability the same.

5. Experiment

A probability experiment is a repeatable chance process whose possible outcomes are defined. Tossing two coins, spinning a fair spinner or drawing one counter from a bag can serve as simple probability experiments. The word experiment here concerns uncertainty, not necessarily the causal experiments discussed in Volume 7.

A probability experiment can have known or modelled probabilities even when the next result is unknown. Conversely, collecting real outcomes without a clear procedure can make interpretation difficult. State the experiment’s rules, including replacement, ordering and what is recorded, before assigning likelihoods.

Use chance experiment, repeat the experiment and experimental result. Tricia distinguishes the model from one realised outcome. For practice, rewrite “We got heads five times” so the sentence also identifies how many tosses were performed and whether each toss followed the same rule.

6. Possibility

A possibility is something that can occur under the stated rules. On a standard die, rolling six is possible and rolling seven is not. Possibility answers whether an outcome or event can occur; probability answers how likely it is within the model.

Possible does not mean probable. An event with probability one in a thousand is possible but unlikely. Likewise, saying an outcome is possible does not imply that all possible outcomes are equally likely. The physical or mathematical model determines the probabilities.

Use a possible outcome, within the set of possibilities and possible but unlikely. Kai Kai separates can happen from likely to happen. For practice, give one event that is possible but rare and one that is impossible under a simple stated model.

7. Impossibility

Impossibility means that an event cannot occur under the stated model or rules. Rolling a seven on a standard six-sided die is impossible. Its probability is zero. The statement belongs to the model: if the object is not actually a standard six-sided die, the conclusion may not apply.

A zero observed frequency does not automatically prove impossibility. An event with small positive probability may simply not appear in a limited number of trials. The distinction between never observed and impossible is fundamental. Evidence from finite trials cannot usually turn absence into probability zero without additional structural information.

Use impossible under the stated rules, probability zero in this model and not observed but not impossible. Alicia checks the sample space. For practice, explain why seeing no sixes in ten die rolls does not make rolling a six impossible on the eleventh roll.

8. Certainty

Certainty means that an event must occur under the model. Drawing either red or blue from a bag containing only red and blue counters is certain. A certain event has probability one. The outcome can still be uncertain at a more detailed level: the colour category is certain while the exact colour drawn remains uncertain.

Certainty should be attached to the correct event. “Something will be drawn” may be certain, while “red will be drawn” is not. In real-world decisions, complete certainty is uncommon because models can be incomplete. This article reserves probability-one language for clearly defined mathematical or stipulated classroom models.

Use certain event, probability one and certain within the model. Tricia states the event before calling it certain. For practice, write one certain event and one uncertain event for the same bag without changing the bag’s contents.

9. Favourable outcome

A favourable outcome is an outcome that belongs to the event being counted for a particular probability question. The word favourable here does not necessarily mean good or desirable. For the event “roll an even number”, outcomes 2, 4 and 6 are favourable because they satisfy the condition.

The same outcome can be favourable for one event and unfavourable for another. Rolling six is favourable for “greater than four” and unfavourable for “less than three”. The term is therefore question-relative. Avoid letting its everyday positive meaning distort the mathematics.

Use three favourable outcomes, favourable to the event and count the favourable cases. Kai Kai states the event first. For practice, identify the favourable outcomes for “number at most two” on a six-sided die and explain why two itself is included.

10. Equally likely

Equally likely means that outcomes have the same probability. On a fair six-sided die, the six faces are modelled as equally likely. This permits the simple calculation of favourable outcomes divided by total outcomes when the counted outcomes are truly equally likely.

Distinct outcomes are not automatically equally likely. When two coins are summarised by number of heads, the recorded outcomes 0, 1 and 2 heads are not equally likely because one head can occur as HT or TH. Treating the three summaries as equal would ignore their different numbers of underlying ordered outcomes.

Use equally likely outcomes, not equally likely and fair model. Alicia checks the construction before using simple case counting. For practice, explain why a spinner with one large red sector and one small blue sector has two possible colour outcomes without making those colours equally likely.

Entries 11–20: describe how likely an event is without turning a probability into a promise

11. Likelihood

Likelihood is how likely an event is to occur under the stated model or evidence. In elementary probability, likelihood can be described with words, fractions, decimals or percentages. An event with probability three quarters has greater likelihood than one with probability one quarter.

Likelihood does not identify the actual next outcome. A more likely event can fail to occur on the next trial. Nor should vague words such as likely replace an available exact probability when precision matters. Use the numerical measure and the verbal description together when doing so improves understanding.

Use greater likelihood, likelihood of the event and likely but not certain. Tricia compares probabilities before choosing the adjective. For practice, order events with probabilities 0, 0.2, 0.5, 0.8 and 1 using both numerical and verbal descriptions.

12. Likely

Likely describes an event whose probability is comparatively high in the relevant context. In a bag with nine blue and one red counter under an equal-draw model, blue is likely. The word is relative and should be used with awareness of the model and alternatives.

Likely does not mean certain. A ten-per-cent red outcome can still occur. Nor is there one universal numerical threshold at which every context must switch from possible to likely. In classroom probability, the word can be anchored to clear comparisons such as more likely than not when probability exceeds one half.

Use more likely than, likely but uncertain and most likely among the listed events. Alicia compares events instead of treating likely as an absolute guarantee. For practice, explain why a 70% event is likely but still leaves substantial chance for the 30% alternative.

13. Unlikely

Unlikely describes an event with comparatively low probability in the stated context. Drawing the single red counter from a bag with nine blue and one red is unlikely. Its occurrence would be surprising relative to blue, but entirely compatible with the model.

Unlikely does not mean impossible and should not be reported as “cannot happen”. A rare event’s occurrence also does not automatically show that the process is unfair. The correct response is to compare the outcome with the model and, over many trials, examine whether the broader pattern remains plausible.

Use unlikely event, unlikely but possible and less likely than. Kai Kai avoids treating one rare outcome as proof of cheating. For practice, write a sentence describing a one-in-ten event after it occurs once, preserving both its prior low probability and the fact that it actually happened.

14. Probable

Probable means likely to occur or be true given the model or evidence. In formal mathematical work, the numerical probability is clearer than relying on the adjective alone. In explanatory prose, probable can summarise a high-probability judgement when the basis is already stated.

Probable should not be used to disguise unsupported confidence. A student cannot call tomorrow’s result probable merely because they hope for it. The word requires a model, frequency, evidence or justified comparison. It also does not imply certainty; a probable outcome can fail.

Use probable under this model, more probable than and the most probable listed outcome. Tricia states the basis before the adjective. For practice, compare probable, possible and certain in one three-sentence example, ensuring that each term refers to a different probability level.

15. Chance

Chance can refer to the possibility or probability that an event occurs. In everyday language, “a 25% chance of red” means a probability of one quarter. The word can also mean an opportunity, so scientific or mathematical context should make the intended sense clear.

Chance does not mean absence of structure. A fair die has chance outcomes governed by a clear probability model. Repeated results can be analysed systematically even though no one knows the next face in advance. Randomness and lawful probability can coexist.

Use chance of the event, a one-in-four chance and chance outcome. Alicia translates between fraction and percentage. For practice, express probability 0.25 in fraction, percentage and ordinary chance language without implying that one success must occur exactly every four trials.

16. Random

Random, in a probability model, means that the particular outcome is governed by chance according to a stated process rather than chosen to produce a desired result. A well-shuffled draw may be modelled as random when every eligible item has the intended selection probability.

Random does not mean patternless in every finite sequence. Three heads in a row can occur in random coin tossing. It also does not mean fair: a biased coin can still produce random outcomes with unequal probabilities. Fairness concerns the probabilities; randomness concerns unpredictability under the process.

Use random outcome, random selection and random does not mean equally likely. Kai Kai separates unpredictability from fairness. For practice, describe a biased coin whose next toss is random but whose two outcomes are not equally likely.

17. Fair

Fair, in elementary probability, describes a device or process whose relevant symmetric outcomes are equally likely as intended. A fair coin gives heads and tails equal probability in the model. A fair six-sided die assigns equal probability to its six faces.

Fair does not mean a short run must display equal counts. Five heads and one tail in six tosses can occur with a fair coin. Judging fairness from data requires more than expecting perfect balance in every small sample. The term also has moral meanings outside mathematics, which should not be confused with equal-probability modelling.

Use fair coin, fair spinner and consistent with a fair model. Tricia distinguishes model assumption from observed frequency. For practice, explain why ten tosses split 7–3 do not by themselves prove the coin is unfair.

18. Biased

Biased, in a simple probability device, means the outcomes do not have the equal probabilities that a fair version would assign. A weighted die can be biased toward some faces. In sampling and judgement, bias has wider meanings, but here the immediate sense concerns unequal outcome tendencies or systematic distortion.

One unusual result does not establish bias. Evidence for bias comes from the construction of the device, a validated mechanism or enough data to make the fair model implausible under an appropriate analysis. Do not turn surprise into accusation. A rare sequence can occur by chance.

Use biased coin, bias toward and evidence of bias. Alicia asks whether the claim comes from the device design or just a short sequence. For practice, compare a coin known to have uneven sides with an ordinary coin that produces four heads in a row. Explain which case provides structural evidence of bias.

19. More likely than not

More likely than not means having probability greater than one half. An event with probability 0.6 is more likely than not; one with probability exactly 0.5 is not. The phrase gives a clear comparative threshold without implying certainty.

The threshold concerns probability, not consequence. An event can be more likely than not but relatively harmless, while a less likely event can have severe consequences. Decision-making later in this volume combines likelihood with impact rather than choosing solely on which event crosses fifty per cent.

Use more likely than not, probability exceeds one half and not quite more likely than not. Kai Kai tests the boundary at 0.5. For practice, classify probabilities 0.49, 0.50 and 0.51 under the phrase and explain why the exact threshold matters.

20. Uncertain

Uncertain means that the outcome, value or state is not known with certainty. A future draw from a mixed bag is uncertain even when its probabilities are known exactly. Uncertainty can therefore coexist with a precise probability model.

Uncertain does not mean unknowable in every respect. We may know the sample space, probabilities and consequences while still not know which outcome will occur. In other cases, the probabilities themselves are uncertain because the model or data are incomplete. These are different layers of uncertainty and should not be merged.

Use uncertain outcome, uncertain probability estimate and known model but unknown result. Tricia names which part is uncertain. For practice, compare a fair-coin toss with a forecast based on limited past data. Explain why both involve uncertainty but in different ways.

Entries 21–30: turn likelihood into a measure without turning the measure into a promise

21. Probability

Probability is a numerical measure of how likely an event is within a model, usually expressed from zero to one or from zero per cent to one hundred per cent. Zero represents impossibility under the model and one represents certainty. A probability of one half does not promise alternating success and failure; it describes likelihood across the chance process.

Probability belongs to an event, not to an outcome after it has already happened. Once a die shows four, the statement “the probability of four was one sixth before the roll” remains meaningful, but the realised outcome is no longer uncertain. Probability describes uncertainty before the result is known or across future repetitions.

Use probability of the event, a probability of 0.25 and probability model. Alicia always names the event. For practice, translate one quarter into decimal and percentage form, then explain why all three notations describe the same likelihood.

22. Theoretical probability

Theoretical probability is probability derived from the structure and assumptions of a model rather than estimated from observed trial counts. For a fair six-sided die, the theoretical probability of rolling an even number is three favourable faces out of six equally likely faces, or one half.

Theoretical does not mean imaginary or always correct in practice. The model may be an approximation of a real device. If the die is biased, the fair-die theoretical calculation no longer describes its true long-run probabilities. A theoretical result is only as suitable as the assumptions supporting the model.

Use theoretical probability under the fair model, calculated from equally likely outcomes and model-based probability. Tricia states the fairness assumption. For practice, explain why the same favourable-over-total method cannot be used directly when the outcomes are not equally likely.

23. Experimental probability

Experimental probability is a probability estimate based on observed results from repeated trials. If red appears 18 times in 60 spins, the experimental probability of red is 18/60, or 0.3. It summarises the observed relative frequency in that set of trials.

Experimental probability can differ from the theoretical model, especially in a small sample. That difference is not automatically an error. As trials accumulate under a stable process, the observed relative frequency may move closer to the model probability, but short-run variation remains possible.

Use experimental probability from sixty trials, observed relative frequency and estimated from the data. Kai Kai reports both numerator and total trials. For practice, compare 18/60 with a theoretical probability of one third and explain why a small difference does not by itself show that the spinner is unfair.

24. Relative frequency

Relative frequency is the fraction, decimal or percentage of trials in which an event occurs. Twenty blue results in eighty trials give a relative frequency of 20/80 = 0.25. The denominator is the number of relevant trials, not the number of possible outcomes in the sample space.

Relative frequency is an observed result, while theoretical probability is model-based. Their closeness can support a model, but exact equality is not required in every finite experiment. Also, different sample sizes can produce the same relative frequency while providing different amounts of evidence.

Use relative frequency of blue, twenty out of eighty trials and observed share of trials. Alicia keeps trial count beside the proportion. For practice, compare 2/8 and 25/100: both equal one quarter, but explain why the larger set contains more observations.

25. Long-run frequency

Long-run frequency refers to the proportion of times an event occurs across a large number of repeated trials under a stable probability process. For a fair coin, the fraction of heads tends to settle near one half as the number of tosses becomes large, although local streaks and short-run imbalances continue to occur.

Long-run does not mean a fixed finite number such as exactly one hundred trials. Nor does it promise that the cumulative proportion moves steadily closer at every new toss. One additional outcome can temporarily move the relative frequency farther from the theoretical probability.

Use long-run proportion, stabilises near and large number of trials. Tricia separates long-run tendency from next-trial prediction. For practice, explain why a coin showing six heads in a row can still be consistent with a long-run probability of one half.

26. Odds

Odds compare favourable outcomes with unfavourable outcomes rather than favourable outcomes with all outcomes. If an event has three favourable and one unfavourable equally likely outcome, the odds in favour are 3:1 while the probability is 3/4. These are related but different expressions.

Do not read odds 3:1 as probability three to one hundred or as three quarters without converting the relationship correctly. The total number of ratio parts is four. Likewise, odds against reverse the order. Technical and everyday uses of odds can vary, so state the intended direction clearly.

Use odds in favour, odds against and convert odds to probability. Kai Kai writes both favourable and unfavourable counts. For practice, convert odds 2:3 in favour into probability and explain why the result is 2/5 rather than 2/3.

27. Fraction

A fraction expresses a part relative to a whole. In equally likely probability models, the numerator can count favourable outcomes and the denominator the total outcomes in the sample space. A probability of 3/8 says three of eight equally likely cases satisfy the event.

The fraction form is meaningful only when numerator and denominator correspond to the model. If the eight listed labels are not equally likely, counting three labels over eight does not automatically give probability 3/8. A simple fraction can therefore hide an incorrect assumption about equal weighting.

Use probability fraction, three eighths and equivalent fraction. Alicia simplifies after checking the model. For practice, explain why 6/16 and 3/8 express the same probability while still representing different numbers of counted cases.

28. Decimal probability

A decimal probability expresses probability as a number from zero to one. A probability of 0.75 is three quarters or seventy-five per cent. The decimal can make comparisons straightforward, especially when probabilities are not simple familiar fractions.

A decimal written to several places may be rounded or estimated. Do not treat 0.333 as exactly one third unless the context says it is a rounded representation. Similarly, 0.50 and 0.5 express the same mathematical probability even though one shows more displayed digits.

Use decimal probability, probability 0.6 and rounded to three decimal places. Tricia keeps exact fractions where helpful. For practice, convert 7/20 to decimal and percentage form and explain which notation is easiest to compare with 0.4.

29. Percentage probability

A percentage probability expresses likelihood out of one hundred. Probability 0.18 is 18%. This form can be accessible in ordinary language but should still be connected to the same event and model as the fraction or decimal version.

Percentage probability is not the same as percentage change. A chance rising from 20% to 30% rises by ten percentage points and by 50% relative to the original probability. The units and question determine which comparison is relevant.

Use an 18% chance, probability expressed as a percentage and ten percentage points higher. Kai Kai checks whether the sentence compares probabilities or describes one probability. For practice, explain the difference between “30% probability” and “a 30% increase in probability”.

30. Complement

The complement of an event is the event that the original event does not occur. If A is “draw blue”, its complement is “do not draw blue”. Their probabilities add to one because together they cover every possible outcome without overlap.

Complement is useful when the opposite event is easier to calculate. The probability of at least one success can sometimes be found by subtracting the probability of no successes from one. The method works only when the event and its complement truly partition the entire sample space.

Use complementary event, probability one minus and event and its complement. Alicia checks that no outcome belongs to both and none is omitted. For practice, give the complement of “roll a number greater than four” on a standard die and verify that the probabilities sum to one.

Entries 31–40: use experiments and simulations to explore uncertain processes

31. Simulation

A simulation is a model that imitates the important chance structure of a process so it can be explored repeatedly. A spinner, random-number generator or shuffled set of labelled cards can simulate another situation when the outcome probabilities are matched appropriately.

A simulation result belongs first to the simulation. Its usefulness for the real situation depends on whether the model preserves the relevant probabilities and conditions. A convenient simulation that misrepresents the probabilities can produce many precise-looking results while answering the wrong question.

Use run a simulation, simulation model and simulate the stated probability. Tricia checks the mapping between model outcomes and target events. For practice, design a simple ten-equal-sector spinner to simulate an event with probability 0.3 and explain which sectors count as success.

32. Random number

A random number is a number generated according to a specified random process, often used to select outcomes or simulate events. Digits 0–9 can simulate a 30% event by assigning three digits to success and seven to failure when each digit is equally likely.

Random numbers need a mapping rule. Assigning success to 0, 1 and 2 gives the same target probability as assigning 4, 7 and 9 when all digits are equally likely. What matters is the number and probability of assigned cases, not the visual order of the labels.

Use random-number simulation, assign random digits and mapping rule. Kai Kai writes the rule before generating outcomes. For practice, create a random-digit mapping for probability 0.6 and explain why six of ten digits are required.

33. Frequency table

A frequency table records how many times each outcome or event occurs across repeated trials. For a coin simulation, the table might show heads 48 and tails 52 after one hundred tosses. The counts preserve the observed outcomes before they are converted to relative frequencies.

A frequency table does not by itself give theoretical probability. It shows what happened in the trials. The trial count also matters: a 5–5 split after ten tosses and a 500–500 split after one thousand share the same relative frequencies but contain very different amounts of observed data.

Use frequency table of outcomes, observed frequency and total number of trials. Alicia checks that the category frequencies sum to the trial total when the outcomes are mutually exclusive and complete. For practice, find the missing category frequency from a complete table.

34. Convergence

Convergence describes values moving closer to a limiting value or pattern. In repeated probability trials, the running relative frequency may converge towards the model probability as the number of trials grows. The word describes a tendency, not a guarantee that every new observation moves closer.

A short run can wander away before later returning closer. Convergence also depends on a stable process. If the device changes halfway through the experiment, one long-run probability may no longer describe the whole sequence.

Use converges towards, running proportion and stable probability process. Tricia plots cumulative relative frequency rather than isolated block frequencies. For practice, explain how 0.7 after ten trials and 0.52 after one hundred can be consistent with convergence to 0.5.

35. Variation

Variation is the natural difference among observed outcomes or frequencies across repeated sets of trials. Two groups each tossing a fair coin twenty times can obtain different numbers of heads. That variation is expected in random processes and does not automatically indicate different coins.

Variation generally has a larger proportional effect in smaller samples. A difference of two heads changes a ten-toss proportion more than a thousand-toss proportion. Recognising variation prevents the learner from expecting every finite experiment to reproduce theoretical proportions exactly.

Use random variation, variation between samples and expected variation. Kai Kai compares sample size before judging discrepancies. For practice, explain why 7 heads in 10 tosses is a larger proportional departure from 0.5 than 52 heads in 100 tosses.

36. Streak

A streak is a consecutive run of the same or similar outcome. HHHH is a streak of four heads. Streaks can occur in random sequences and should not automatically be interpreted as a changing probability or a device “wanting” to continue the pattern.

After four heads from independent fair tosses, the next toss still has probability one half of heads. The previous streak does not create a debt that tails must repay. This is a useful place to distinguish remembered sequence from current trial probability.

Use streak of four, random streak and streak does not alter the next-trial probability. Alicia checks independence before making that statement. For practice, explain why the next toss after HHHH is not made more likely to be tails by the previous results alone.

37. Law of large numbers

The law of large numbers is the principle that, under suitable stable conditions, averages or relative frequencies tend to move closer to their expected values as the number of independent observations becomes large. At this level, the term is useful as an extension concept for explaining why long runs can stabilise even though individual outcomes remain unpredictable.

The law does not say that short-run imbalances must be corrected immediately. Ten heads in a row do not create a mathematical force pushing the next toss to tails. Nor does it say the counts of heads and tails must become exactly equal; the relative proportion can approach one half while the absolute count difference still changes.

Use long-run stabilisation, law of large numbers and relative frequency tends toward. Tricia keeps next-trial probability separate from long-run proportion. For practice, explain why a future run can move the cumulative proportion toward one half without requiring a perfectly alternating sequence.

38. Sample size

Sample size is the number of observations or trials included in an experimental probability estimate. A sample size of twenty means twenty recorded trials under the stated procedure. Larger samples often produce more stable relative frequencies, assuming the probability process remains comparable.

A large sample does not repair a biased device or changing procedure. Ten thousand trials from the wrong simulation model can estimate the wrong target very precisely. Quantity of observations and quality of model must be considered together.

Use sample size of, larger sample and stability with increasing sample size. Kai Kai reports the denominator beside every experimental probability. For practice, explain why 30 successes out of 100 and 300 out of 1000 have the same relative frequency but different sample sizes.

39. Estimate

An estimate is an approximate value used to represent an unknown probability, frequency or future quantity. Experimental probability can estimate an underlying chance when the process is stable but its exact probability is not already known.

An estimate should remain labelled as an estimate. Increasing trial count can improve stability, but finite data still leave uncertainty. A clean decimal displayed by a calculator does not turn the estimate into a known exact probability.

Use estimate the probability, probability estimate and estimated from repeated trials. Alicia separates exact model probabilities from empirical estimates. For practice, explain why 37 successes in 100 trials supports an estimate of 0.37 without proving the true probability is exactly 0.37.

40. Model

A probability model specifies possible outcomes and assigns probabilities to them. A fair coin model has outcomes H and T with probability one half each. A bag model may assign probabilities according to the proportion of equally selectable counters of each colour.

A model is a representation, not the physical object itself. If the selection process favours some counters because of shape or placement, the simple proportion model may fail. Comparing experimental data with model predictions can help evaluate whether the assumptions are adequate.

Use probability model, model assumption and consistent with the model. Tricia lists outcomes and probabilities before simulation. For practice, write a model for a spinner with unequal sectors and explain why counting sector labels is insufficient unless sector sizes are included.

Entries 41–50: combine events without losing the relationships between them

41. Combined event

A combined event involves two or more conditions considered together. “Roll an even number and then draw blue” combines results from two parts of a process. The probability depends on how the parts are related and whether the second stage is affected by the first.

Combined does not automatically mean multiply. Multiplication rules require conditions such as independence or conditional probabilities that describe how the stages connect. The learner should first write the event in words, then identify the structure before choosing an operation.

Use combined event, two-stage event and the event occurs only if both conditions are met. Alicia sketches the stages before calculating. For practice, compare “red and even” with “red or even” and explain why they include different outcomes.

42. Independent events

Independent events are events where the occurrence of one does not change the probability of the other. In a model of two separate fair coin tosses, the result of the first toss does not change the probability of heads on the second.

Independent does not mean unrelated in language or occurring far apart in time. It has a precise probability meaning. Two events can be close together but independent, or physically separate yet dependent through a shared condition. Always test the probability relationship rather than relying on everyday intuition.

Use independent events, probability unchanged by and independence assumption. Tricia checks whether replacement restores the original bag composition. For practice, explain why two draws with replacement can be independent while two draws without replacement usually are not.

43. Dependent events

Dependent events are events where one outcome changes the probability of another. Drawing a red counter without replacement changes the bag before the second draw, so the probability of another red draw may change.

Dependence does not tell the learner whether the second event becomes more or less likely without inspecting the case. Removing one red may reduce another-red probability; removing a blue may increase it. The direction follows the changed composition.

Use dependent events, the first result changes and conditional on the first draw. Kai Kai updates the denominator after each draw. For practice, compare a bag with two red and two blue counters before and after one red is removed.

44. Mutually exclusive

Mutually exclusive events cannot occur together in the same trial under the stated definitions. On one die roll, “roll a two” and “roll a five” are mutually exclusive. “Roll even” and “roll greater than three” are not, because four and six satisfy both.

Mutually exclusive does not mean independent. If one mutually exclusive event occurs, the other becomes impossible in that trial, so their relationship is strongly dependent except in special zero-probability cases. These two technical terms answer different structural questions.

Use mutually exclusive events, cannot occur together and overlapping events for the opposite situation. Alicia lists the intersection. For practice, classify “even” and “odd” versus “even” and “greater than four” on a standard die.

45. Overlap

Overlap occurs when outcomes belong to more than one event. In one die roll, outcome six belongs to both “even” and “greater than four”. If the probabilities of those events are added without adjusting for overlap, the shared outcomes are counted twice.

Overlap can be represented with lists, tables or Venn diagrams. The visual tool is less important than preserving the logic. A learner should be able to say which outcomes are shared before applying a formula.

Use overlapping events, shared outcomes and avoid double-counting the overlap. Tricia marks the common outcomes once. For practice, list the overlap between “number at least three” and “even” on a standard die.

46. Conditional probability

Conditional probability is the probability of an event given that another condition is known to have occurred. If a drawn card is known to be red, the probability it is a heart is evaluated within the red-card subset rather than the whole deck.

Conditioning changes the reference set. A probability can rise, fall or stay the same after new information. The word given is consequential because it tells the learner which cases remain possible. Conditional probability is not the same as multiplying two unrelated probabilities.

Use conditional on, given that and updated sample space. Kai Kai redraws the possible cases after learning the condition. For practice, explain why knowing that a die result is even changes the probability that it is greater than three.

47. Tree diagram

A tree diagram is a branching representation of sequential outcomes. Each level can show a stage of the process, and each branch can carry an outcome or probability. It is especially useful when the probabilities change after earlier outcomes.

A tree diagram is only as good as its branches. Missing a possible outcome or assigning the wrong conditional probability produces an incorrect model even if the diagram looks tidy. Branch probabilities from one node should represent the complete possibilities from that state.

Use branch of the tree, two-stage tree diagram and follow the path. Alicia labels before multiplying. For practice, describe in words the four branches produced by two coin tosses and connect each terminal path to an ordered outcome.

48. Path probability

A path probability is the probability of following a particular sequence of branches in a multi-stage model. For independent fair coin tosses, the path HH has probability one half times one half, or one quarter.

Multiplication along a path reflects the probability of each stage given the path so far. If later branch probabilities change after earlier outcomes, use those conditional probabilities rather than repeating the initial probability.

Use probability of the path, multiply along the branches and conditional branch probability. Tricia explains the branches before the product. For practice, calculate red-then-red from a two-red, two-blue bag without replacement.

49. Union

The union of two events is the event that at least one of them occurs. In ordinary school language this often corresponds to “A or B”, including outcomes that belong to both. On a die, even or greater than four includes 2, 4, 5 and 6.

Everyday or can sometimes be interpreted exclusively, but probability union normally includes the overlap unless the problem states otherwise. The learner should inspect the event definition rather than import conversational assumptions.

Use union of events, A or B and at least one event occurs. Kai Kai lists the combined outcomes once. For practice, find the union of “less than three” and “even” on a standard die.

50. Intersection

The intersection of two events is the event that both occur. On one die roll, the intersection of “even” and “greater than four” is {6}. The intersection identifies the overlap that must be handled correctly when combining event probabilities.

An empty intersection means the events are mutually exclusive. A non-empty intersection does not tell the learner its probability until the probabilities of the shared outcomes are known. In equally likely finite models, counting can be enough.

Use intersection of events, A and B and shared outcomes. Alicia writes the intersection set before using it in a probability calculation. For practice, find the intersection of “at least two” and “at most four” on a six-sided die.

Entries 51–60: use probability to forecast without pretending the future has already happened

51. Expected frequency

An expected frequency is the number of times an event would be expected on average across many similar sets of trials under a probability model. With probability 0.3 over 100 trials, the expected frequency is 30.

Expected does not mean guaranteed. A real run can produce 27, 34 or another count. Expected frequency is a model-based centre across repeated comparable runs, not a promise for one particular run.

Use expected frequency of thirty, on average across repeated sets and actual frequency may differ. Tricia distinguishes expectation from observation. For practice, calculate the expected number of sixes in sixty fair-die rolls and explain why ten is not mandatory.

52. Expected value

Expected value is the probability-weighted average of numerical outcomes in a model. It represents the long-run average result across many repetitions, not necessarily a value that can occur in one trial. A game paying 0 points with probability one half and 4 points with probability one half has expected value 2 points.

Expected value does not tell the learner whether the outcomes are risky or evenly spread. Two games can have the same expected value and very different possible consequences. Decision-making may therefore need both expected value and information about variation or downside.

Use expected value of, probability-weighted average and long-run average. Alicia lists every outcome and probability before summing. For practice, compare a certain two points with a 50–50 chance of zero or four points.

53. Forecast

A forecast is a statement about a future outcome or quantity based on a model, evidence or assumptions. A forecast can assign probabilities to several possible outcomes rather than select one as certain.

A forecast should not be judged solely by whether the most likely outcome happened once. A 70% event can fail 30% of the time. Calibration over many comparable forecasts is a stronger way to judge whether stated probabilities match observed frequencies.

Use probabilistic forecast, forecast assigns and forecast uncertainty. Kai Kai distinguishes most likely from guaranteed. For practice, explain why a 70% rain forecast is not automatically wrong if it does not rain that day.

54. Scenario

A scenario is a described possible future or condition used to explore consequences. A decision packet may include low-demand, typical-demand and high-demand scenarios. Scenarios organise possibilities; they are not predictions unless probabilities are assigned.

A plausible scenario can be useful even when it is not the most likely outcome. Decision-makers often test whether a plan remains acceptable under several possibilities. The point is to understand sensitivity and consequences, not to pretend one imagined future is certain.

Use scenario analysis, under the high-demand scenario and plausible scenario. Tricia separates scenario from forecast. For practice, create three fictional weather-independent classroom attendance scenarios without assigning probabilities, then explain what extra information would be needed to rank their likelihood.

55. Prediction interval

A prediction interval is an advanced extension term for a range intended to contain a future observation with a stated level of probability under a model. At Primary 6 level, the central idea is that responsible forecasts can use ranges instead of pretending one exact future value is known.

A prediction range is not a guarantee that every future result will fall inside it. Its meaning depends on the model and coverage probability. This volume uses the term conceptually rather than requiring formal interval calculations.

Use prediction range, forecast interval and future value remains uncertain. Alicia explains the idea in ordinary language first. For practice, compare an exact claim “tomorrow will be 24” with a probabilistic range statement and discuss which is more honest when uncertainty is real.

56. Base rate

A base rate is the background frequency or probability of an event before considering more specific information. If 5% of ordinary fictional system checks show a fault, that 5% is a base rate for the fault in that setting.

Ignoring base rates can make rare events seem more probable than they are after an imperfect signal. At this level, learners need not perform advanced Bayes calculations to understand the principle: evidence should be interpreted against how common the event was to begin with.

Use base rate of the event, background frequency and rare before the new evidence. Kai Kai asks how common the condition is before reacting to a signal. For practice, compare a clue about a very common event with the same clue about a very rare event.

57. Calibration

Calibration, in forecasting, describes how well stated probabilities match observed frequencies across many comparable predictions. If events forecast at around 70% occur around 70% of the time over a large suitable set, the forecasts are well calibrated at that level.

Calibration is not the same as always choosing the right most-likely outcome. A forecaster can be well calibrated while some high-probability events fail and some low-probability events occur. Probability statements are evaluated over groups of comparable forecasts.

Use well-calibrated probabilities, forecast calibration and compare stated probability with frequency. Tricia checks many predictions rather than one. For practice, explain why a single failed 90% forecast does not by itself establish poor calibration.

58. Plausible

Plausible means reasonably possible given the available evidence or model. A plausible scenario is not necessarily the most likely one. The term is useful when exact probabilities are unavailable but the outcome remains consistent with what is known.

Plausibility should not become an excuse to treat every imaginable story equally. A scenario that contradicts known constraints is not plausible merely because it can be described. Evidence narrows the space of reasonable possibilities.

Use plausible explanation, plausible range and less plausible given. Alicia names the evidence supporting the scenario. For practice, distinguish plausible, probable and certain using three fictional future outcomes.

59. Robust forecast

A robust forecast is a forecast whose central conclusion remains reasonably stable under modest changes in assumptions, data or modelling choices. At this level, robust means not overly dependent on one fragile assumption.

Robust does not mean certain or perfect. A robust forecast can still be wrong because the future contains genuine uncertainty. Robustness is about stability of the reasoning, not guaranteed success.

Use robust to small changes, robust forecast and sensitive rather than robust. Tricia changes one assumption at a time. For practice, compare a forecast that stays near 60% under several reasonable inputs with one that swings from 20% to 80%.

60. Sensitivity

Sensitivity is how much a result, forecast or decision changes when an assumption or input changes. If a small shift in estimated probability reverses the preferred option, the decision is sensitive to that estimate.

Sensitivity is not automatically bad. It tells the learner which uncertain inputs matter most and therefore deserve better information. A decision that is insensitive to reasonable variation may be easier to defend.

Use sensitivity analysis, sensitive to the probability estimate and robust across the tested range. Kai Kai varies one assumption while holding others fixed. For practice, identify which of two uncertain inputs changes a fictional decision threshold.

Entries 61–70: combine likelihood with consequence

61. Risk

Risk is the possibility of an unwanted outcome considered together with its likelihood and consequence. An event can be unlikely but still represent meaningful risk when its potential harm is large. Another event can be common but low-risk when its consequence is minor.

Risk is not identical to probability. Saying an event has a 10% chance does not tell the reader how serious it would be. Nor does a severe consequence tell the probability. A responsible comparison keeps likelihood and impact visible as separate dimensions before combining them.

Use risk of the outcome, low-probability high-consequence risk and risk depends on both likelihood and impact. Alicia names both parts. For practice, compare a 50% chance of losing one point with a 1% chance of losing one hundred points without deciding from probability alone which matters more.

62. Consequence

A consequence is what happens if an outcome occurs. Consequences can be positive, negative or mixed. In a fictional classroom planning game, choosing a longer route may reduce the chance of delay but require more time even when no delay occurs.

Consequence should be distinguished from probability. A serious consequence can be rare, and a frequent event can have little consequence. Decisions under uncertainty often become clearer when the learner lists possible outcomes first and assesses consequence separately from likelihood.

Use consequence of the event, potential consequence and consequence if the event occurs. Tricia avoids calling the worst consequence the most likely outcome. For practice, write two consequences for one decision and assign different probabilities to them.

63. Impact

Impact is the magnitude or importance of a consequence. A missed deadline may have small impact in one practice task and large impact in another context. The term therefore needs a reference: impact on what objective, person or outcome?

Impact is not automatically numerical. It can be described using categories when the criteria are clear. However, words such as high or severe should not be assigned merely to make a scenario dramatic. The judgement needs a stated basis.

Use high-impact event, impact on the objective and low likelihood but high impact. Kai Kai states the criterion for impact. For practice, rank three fictional consequences for a classroom exhibition using a declared scale and explain the ordering.

64. Severity

Severity is how serious an undesirable consequence would be. In risk discussions, severity can be considered separately from likelihood. A rare severe outcome may justify precautions even when a more common mild outcome is more probable.

Severity categories need definitions. “Severe” should not mean whatever the writer dislikes most. A simple classroom model can define low, medium and high severity through specified consequences. Real safety-critical decisions require expert and institutional guidance beyond this educational vocabulary exercise.

Use severity of the consequence, high-severity outcome and severity scale. Tricia checks the description before the label. For practice, create a three-level severity scale for fictional project delays without using any real safety or medical context.

65. Exposure

Exposure is the extent to which a person, system or plan is subject to a source of uncertain loss or change. In a fictional planning model, relying on one supplier creates more exposure to that supplier’s delay than using two independent alternatives.

Exposure is not identical to realised loss. A plan can be exposed to a risk that never occurs. Conversely, a small exposure can still produce a loss if the adverse outcome happens. The word helps distinguish being vulnerable to an event from actually experiencing it.

Use exposure to delay, reduce exposure and remaining exposure. Alicia asks what event the plan depends on. For practice, compare two fictional schedules with different dependence on one uncertain delivery.

66. Vulnerability

Vulnerability is a weakness or condition that makes a system more susceptible to harm when an adverse event occurs. A project with no backup copy is more vulnerable to one file becoming unavailable than a project with an independent verified backup.

Vulnerability does not mean the adverse event is likely. It describes how badly the system may cope if it happens. A robust design can face the same external probability with lower vulnerability because it has alternatives, buffers or recovery procedures.

Use vulnerability to, reduces vulnerability and single point of vulnerability. Kai Kai separates event probability from system weakness. For practice, explain why adding a backup changes vulnerability even if it does not change the probability that the primary copy fails.

67. Precaution

A precaution is an action taken in advance to reduce the likelihood or consequence of an unwanted outcome. Saving a verified backup of a project file is a precaution against data loss. It may reduce consequence without changing the probability of the original device failing.

A precaution has costs or trade-offs, such as time or resources. The goal is not always to eliminate every conceivable risk. A proportionate precaution addresses meaningful risk at reasonable cost within the decision context.

Use reasonable precaution, precaution against and reduces the consequence. Tricia asks which part of risk the action changes. For practice, identify whether a fictional precaution reduces likelihood, consequence or both.

68. Mitigate

To mitigate is to reduce the likelihood, impact or severity of an unwanted outcome. A project can mitigate schedule risk by adding a time buffer or preparing an alternative route. The event may remain possible while its effect becomes smaller or easier to manage.

Mitigate does not mean eliminate. A backup plan can reduce exposure without guaranteeing success. Scientific and planning language should state the remaining risk rather than using mitigation as a promise that the problem cannot occur.

Use mitigate the risk, mitigation measure and partly mitigated. Alicia names the residual risk. For practice, distinguish mitigation from avoidance in a fictional decision where one option removes an uncertain dependency and another merely reduces its consequence.

69. Residual risk

Residual risk is the risk that remains after precautions or mitigation have been applied. A backup copy reduces the consequence of losing one file, but both copies might still become unavailable under a shared failure. The remaining possibility is residual risk.

Residual risk prevents a misleading all-or-nothing view. A plan can become much safer without becoming risk-free. Whether the remaining risk is acceptable depends on the context, criterion and authority responsible for the real decision.

Use residual risk remains, after mitigation and remaining exposure. Kai Kai checks shared dependencies. For practice, add one mitigation to a fictional scenario and identify one risk it does not address.

70. Risk tolerance

Risk tolerance is the amount or type of uncertainty and potential loss a decision-maker is prepared to accept in pursuit of an objective. Different contexts can justify different tolerances. A classroom game may accept outcomes that would be inappropriate in a real safety-critical decision.

Risk tolerance is not a mathematical fact derived from probability alone. It reflects goals, constraints and consequences. In real organisations it may be set by policies or authorities. This volume uses the term only in fictional, low-stakes decision exercises.

Use within the stated risk tolerance, low tolerance for delay and acceptable risk threshold. Tricia keeps values and preferences separate. For practice, compare two fictional decision-makers with different deadlines and explain why the same probability can lead to different choices.

Entries 71–80: choose among uncertain options without pretending one criterion answers everything

71. Decision

A decision is a choice made among alternatives using goals, evidence, constraints and uncertainty. A decision is evaluated by the reasoning available at the time, not only by whether the realised outcome later turns out well.

A good decision can produce a bad outcome because uncertainty remains. A poor decision can occasionally get lucky. Separating decision quality from outcome quality is one of the central lessons of probability-aware reasoning.

Use decision under uncertainty, decision rule and quality of the decision process. Alicia records the probabilities known before the outcome. For practice, give an example where the higher-probability choice loses once and explain why that single outcome does not prove the decision was irrational.

72. Alternative

An alternative is one option among the available choices. A decision becomes clearer when alternatives are defined consistently enough to compare. “Do nothing”, “prepare a backup” and “change the schedule” can be alternatives if each is feasible under the stated rules.

An alternative should not be included merely as an impossible straw option. Its costs, consequences and constraints should be described fairly. New information can also create or remove alternatives, changing the decision problem.

Use available alternative, alternative option and compare alternatives. Kai Kai lists the choices before ranking them. For practice, identify a missing but feasible alternative in a fictional two-option scenario.

73. Criterion

A criterion is a standard used to judge alternatives. A project might compare options by expected completion time, cost, chance of delay or robustness. Different criteria can favour different options.

A criterion should be relevant to the objective and defined before ranking where possible. Choosing a criterion only after seeing which option wins can make the evaluation arbitrary. Multiple criteria may need explicit weighting or priority.

Use decision criterion, meets the criterion and criteria conflict. Tricia states the objective beside the criterion. For practice, compare two options where one has lower expected time and the other lower worst-case delay.

74. Trade-off

A trade-off occurs when improving one desirable feature comes at a cost in another. A route with a lower chance of disruption may be longer even when nothing goes wrong. A decision must decide how much extra time is worth the reduced uncertainty.

Trade-off is not the same as a mistake. Many real choices involve objectives that cannot all be maximised simultaneously. The useful language names what is gained and what is given up rather than labelling one option simply best.

Use trade-off between, accept the trade-off and reduces risk at the cost of. Alicia writes both sides. For practice, create a two-criterion choice where neither option dominates the other.

75. Opportunity cost

Opportunity cost is the value of the best alternative given up when a choice is made. In a fictional classroom planning case, spending the last ten minutes checking one display means those ten minutes cannot be used to rehearse another activity.

Opportunity cost is not necessarily money. It can be time, flexibility or another missed benefit. The concept helps learners recognise that choosing one uncertain strategy can close off another.

Use opportunity cost of, foregone alternative and what must be given up. Kai Kai identifies the best missed option, not every rejected option. For practice, compare the opportunity cost of two mutually exclusive uses of one hour.

76. Expected utility

Expected utility is an advanced extension concept: a decision model can combine outcome probabilities with values representing how desirable or undesirable those outcomes are. This recognises that two outcomes with equal numerical rewards may not have equal practical meaning to every decision-maker.

At Primary 6 level, the important idea is qualitative rather than formal: probability alone may not determine the preferred option because consequences matter differently. This volume does not require assigning real utility numbers to personal or high-stakes decisions.

Use value-weighted decision, expected utility as recognition vocabulary, and probability is only one input. Tricia first discusses consequences in words. For practice, explain why two options with the same expected points can feel different when one has a wide spread of outcomes.

77. Threshold

A threshold is a boundary at which a rule changes. A fictional plan may trigger a backup when the estimated chance of delay exceeds 30%. Exactly 30% and more than 30% are different under that rule.

A threshold is a decision rule, not a natural law. It may be chosen for practical reasons or policy. Different contexts can set different thresholds even with the same probability estimate.

Use decision threshold, exceeds the threshold and at or above when that is the rule. Alicia tests boundary cases. For practice, compare “more than 30%” with “30% or more” using probabilities 0.29, 0.30 and 0.31.

78. Constraint

A constraint is a limit that restricts which alternatives are feasible. A plan may have a fixed budget, deadline or room capacity. A probabilistically attractive option cannot be selected if it violates a hard constraint.

Constraints differ from preferences. A preference can be traded off; a true constraint may remove an option entirely. Clear decision writing labels which is which.

Use within the constraint, hard constraint and constrained by. Tricia eliminates infeasible options before ranking the rest. For practice, compare a preferred arrival time with a fixed closing deadline.

79. Dominant option

A dominant option is an option that performs at least as well as another across every relevant stated criterion and better on at least one, under the comparison being made. When domination exists, the inferior alternative can often be removed without needing a difficult trade-off.

Dominance depends on the chosen criteria and accurate information. Adding another important criterion can remove the apparent dominance. Do not call an option dominant merely because it wins on the writer’s favourite measure.

Use dominates the alternative, dominant under these criteria and no option dominates. Kai Kai compares rows criterion by criterion. For practice, construct two options where one is cheaper and faster with equal risk.

80. Decision rule

A decision rule is an explicit method for choosing among alternatives. A rule might select the lowest expected time provided the chance of missing the deadline stays below a stated threshold. Making the rule explicit improves consistency and allows others to examine the assumptions.

A rule can be poor even when followed consistently, and a good rule can occasionally produce an unfavourable outcome because of chance. Evaluate the rule by its objective and long-run consequences rather than one lucky or unlucky case.

Use decision rule, apply the rule consistently and rule depends on. Alicia writes the threshold before seeing the realised outcome. For practice, design a simple rule for choosing between two fictional routes using expected time and maximum tolerated delay probability.

Entries 81–90: judge uncertain evidence without confusing confidence with certainty

81. Confidence

Confidence is the degree of trust placed in a judgement, estimate or conclusion. Confidence should respond to evidence quality, sample size, model fit and uncertainty rather than to how strongly someone wants a result to be true.

High confidence is not certainty. A well-supported 90% forecast still allows failure. Likewise, low confidence does not mean every alternative is equally plausible. Calibrated confidence preserves both what is supported and what remains open.

Use high confidence in the estimate, confidence limited by and confident but not certain. Alicia names the evidence behind the judgement. For practice, compare a probability from a known model with one estimated from ten observations.

82. Ambiguity

Ambiguity is uncertainty caused by unclear meaning, incomplete specification or multiple possible interpretations. “There is a good chance” is ambiguous when the reader does not know the probability scale, event or evidence behind good.

Ambiguity differs from randomness. A fair coin’s next toss is uncertain even though the model is clear. An ambiguous forecast may be unclear because the model itself has not been specified. Good writing can reduce ambiguity without reducing the underlying chance uncertainty.

Use ambiguous statement, reduce ambiguity and uncertainty caused by missing definition. Tricia asks what “likely” refers to. For practice, rewrite a vague forecast with a named event, time window and probability.

83. Unknown

Unknown describes a value or fact that is not currently known. An outcome can be unknown before a trial even when its probability is known. A probability itself can also be unknown when the process has not been adequately modelled or measured.

Unknown should not be replaced with zero, fifty-fifty or “anything can happen”. The absence of knowledge is not itself a numerical probability. A bounded unknown can still be constrained by known information.

Use unknown outcome, unknown probability and known bounds but unknown exact value. Kai Kai separates missing information from measured zero. For practice, describe a future draw with known bag composition and compare it with a bag whose composition is hidden.

84. Evidence

Evidence is information used to support, update or evaluate a probability judgement. Repeated trial results, device structure and reliable prior records can all contribute. Evidence quality matters as much as quantity.

One dramatic outcome can feel persuasive without being strong evidence against a model. A rare event is expected to occur occasionally. The relevant question is how probable the entire observed pattern would be under the model, not whether one outcome felt surprising.

Use evidence for the probability estimate, weak evidence and update with new evidence. Alicia looks at the full sequence. For practice, compare one unusual toss with a long-run persistent imbalance.

85. Update

To update is to revise a probability judgement when relevant new evidence becomes available. A forecast based on yesterday’s information may change after a new observation. The revision should reflect the evidence rather than the desire for a different conclusion.

Updating does not mean overreacting to every outcome. One trial may provide little information about an underlying probability. Strong evidence, repeated patterns or direct structural information should produce larger justified changes than a single noisy observation.

Use update the estimate, probability updated by evidence and small update. Tricia asks how informative the new evidence is. For practice, compare a single coin toss with discovering that the coin is physically weighted.

86. Base-rate neglect

Base-rate neglect is the error of ignoring how common an event was before considering new evidence. A clue can seem dramatic while still leaving a rare event relatively uncommon if the clue also appears frequently in ordinary cases.

This is an advanced reasoning term. At Primary 6 level, the useful habit is simple: ask “How common was this to begin with?” before reacting to a signal. No formal Bayesian calculation is required to learn that background frequency matters.

Use consider the base rate, base-rate neglect and rare before the signal. Kai Kai checks the denominator population. For practice, explain why a signal seen in both rare and common conditions may not make the rare condition probable.

87. Gambler’s fallacy

The gambler’s fallacy is the mistaken belief that, in independent random trials, a run of one outcome makes the opposite outcome “due” on the next trial. After five heads from a fair coin, tails is not made more likely by the streak; the next toss remains one half under the independent fair model.

The fallacy should be applied only when independence and stable probabilities are appropriate. In sampling without replacement, earlier outcomes genuinely change later probabilities. The learner must check the process before accusing a prediction of this error.

Use gambler’s fallacy, outcome is not due and independent next trial. Alicia checks the replacement rule. For practice, contrast five coin tosses with drawing counters without replacement.

88. Overconfidence

Overconfidence is having greater certainty in a judgement than the evidence justifies. A learner who sees three successes in three trials and declares the true probability to be 100% is overconfident because the sample is too limited to establish certainty.

Overconfidence is about calibration between confidence and evidence, not about personality. The remedy is clearer reasoning: larger samples, explicit assumptions, ranges and comparison with alternative explanations.

Use overconfident conclusion, confidence exceeds the evidence and calibrate confidence. Tricia replaces always with a bounded statement. For practice, revise a universal claim based on five trials.

89. Surprise

Surprise in probability reasoning is the reaction to an outcome or pattern that had low probability under a model. Surprise can be informative, but it is not itself proof that the model is wrong.

A rare event must happen sometimes. The appropriate response depends on how rare the full pattern is and whether the model assumptions remain credible. Several surprising outcomes in one direction may justify stronger investigation than one isolated event.

Use surprising under the model, unexpected but possible and degree of surprise. Kai Kai separates emotional reaction from mathematical probability. For practice, explain why a one-in-twenty outcome is surprising when it occurs but still compatible with the model.

90. Robustness

Robustness is the extent to which a conclusion or decision remains similar under reasonable changes in assumptions or estimates. A route choice that remains preferred whether the delay probability is 20%, 25% or 30% is more robust than one that reverses after a one-percentage-point change.

Robustness does not mean the inputs are certain. It means the decision is less fragile to their uncertainty. When a result is not robust, gathering better information about the sensitive input can be especially valuable.

Use robust decision, robust across the range and not robust to. Alicia tests nearby assumptions. For practice, compare two decisions with identical expected outcomes but different sensitivity to probability estimates.

Entries 91–100: communicate uncertainty so another reader can reason with it

91. Define

To define is to state what an event, outcome, criterion or category means in the current problem. Probability calculations become unreliable when the event changes halfway through the answer. “Success” must mean the same thing in numerator, denominator and conclusion.

A definition should be precise enough to classify boundary cases. “On time” might mean arrival before 3:00 p.m. or at 3:00 p.m. or earlier. The difference changes counts and probabilities.

Use define the event, operational definition where appropriate, and boundary included. Tricia tests an edge case. For practice, define “delay” so a reader can classify exactly 10 minutes late.

92. Quantify

To quantify is to express a feature numerically. Instead of writing “quite likely”, a writer may quantify the probability as 0.72 when the model supports that precision. Quantification can improve comparison while still requiring explanation of uncertainty and source.

Not every judgement can be quantified meaningfully. Inventing numbers for poorly defined concepts can create false precision. Use numerical probabilities when there is a defensible model or estimate.

Use quantify the likelihood, numerical estimate and cannot be quantified from the supplied information. Kai Kai avoids arbitrary percentages. For practice, identify which of three statements contains enough information for a probability calculation.

93. Compare

To compare uncertain options is to examine their probabilities, consequences or decision criteria on a common basis. A 20% risk of a ten-point loss can be compared with a 5% risk of a forty-point loss only after the learner states which measure matters.

Comparison should not reduce every difference to one number when important dimensions differ. Expected loss can be useful, while worst case, variability or threshold probability can matter too.

Use compare probabilities, compare expected consequences and not directly comparable without. Alicia names the common criterion. For practice, compare two fictional options by both expected points and maximum loss.

94. Estimate

To estimate is to produce an approximate value from data, a model or reasonable assumptions. Experimental probability estimates an underlying chance from repeated trials; a forecast estimates a future probability from available information.

An estimate should retain its source and sample size. “Estimated at 30% from 20 trials” communicates more than “30%”. The additional words help the reader judge stability and scope.

Use estimate from, probability estimate and approximate rather than exact. Tricia keeps the evidence beside the number. For practice, rewrite a bare percentage with its trial count and event definition.

95. Interpret

To interpret a probability is to explain what it means for the event and process. A 70% chance means that under the model the event is more likely than not, not that 70% of one event will happen or that it must occur seven times in every ten-trial block.

Interpretation connects the number with the unit of repetition. In long-run terms, similarly modelled cases would show the event around seventy per cent of the time, subject to variation.

Use interpret the probability as, long-run meaning and not guaranteed on the next trial. Kai Kai paraphrases before calculating. For practice, explain 0.8 to a reader without using the words four fifths.

96. Qualify

To qualify a probability statement is to add the conditions or limits needed to keep it accurate. “The model gives a 60% chance” may need “assuming the spinner sectors are equally selectable as specified.”

Qualification should target the uncertainty that matters. Adding “maybe” to every sentence does not improve reasoning. State whether the issue is small sample size, model assumption, changing process or consequence uncertainty.

Use qualified by, under the stated assumptions and estimate based on. Alicia adds the smallest meaningful limit. For practice, repair an overconfident forecast using one evidence-based qualification.

97. Justify

To justify is to explain why a probability, decision or comparison is warranted. A decision might be justified by lower expected loss, lower chance of missing a deadline or greater robustness under uncertain estimates.

Justification should name the criterion and evidence. “Option A feels safer” is weaker than “Option A keeps the delay probability below the stated 20% threshold across all tested scenarios.”

Use justify the choice, justified by the model and decision criterion. Tricia separates reason from outcome. For practice, justify a choice before revealing which outcome actually occurred.

98. Revise

To revise is to change a probability estimate, model or decision after relevant new information appears. Revision is rational when the evidence changes. Refusing to revise can be as unreasonable as changing direction after every noisy outcome.

The size of revision should reflect the strength of new evidence. A structural discovery about the device can justify a large update; one ordinary trial may justify only a small change.

Use revise the estimate, revision after new evidence and decision remains unchanged when robust. Kai Kai records the old and new basis. For practice, explain why discovering a biased spinner should trigger a larger revision than one unexpected spin.

99. Communicate

To communicate uncertainty well is to state the event, probability or range, source, consequences and necessary limits clearly enough that another reader does not mistake likelihood for certainty. Good communication preserves what is known and unknown.

Communication should avoid both false precision and vague hedging. “About 30%, estimated from 100 comparable trials” can be more informative than “maybe” and more honest than “30.000%” when the data do not justify three decimal places.

Use communicate uncertainty, state the probability and condition and avoid false certainty. Alicia asks a partner to paraphrase the forecast. For practice, revise a probability sentence whose reader wrongly thinks the event is guaranteed.

100. Decide

To decide is to select an action or option using the available objectives, probabilities, consequences, constraints and preferences. A probability model informs the decision; it does not make the decision by itself.

A defensible decision can later produce an unfavourable outcome. That does not retroactively change the probabilities known beforehand. Review the quality of the reasoning separately from the luck of one realised result.

Use decide under uncertainty, decision based on and defensible before the outcome was known. Tricia records the decision rule first. For the final vocabulary task, choose between two fictional low-stakes options and explain the probability, consequence and criterion driving the choice.

Six original probability and decision laboratories with worked answers

All games, bags, routes, scores and forecasts in these laboratories are fictional teaching models. They are not gambling advice, financial advice, safety advice or predictions about real services. Each packet fully states its own rules. The task is to reason from those rules and to write conclusions whose certainty matches the evidence.

Laboratory A: The spinner that looked unfair after twelve spins

A1 — model. A fictional spinner has four equal sectors: red, blue, green and yellow. Under the design model, each colour has theoretical probability one quarter. The pointer is spun twelve times using the same rule.

A2 — observed results. The sequence is R, R, B, R, G, R, Y, R, B, R, G, R. Red occurs seven times, blue twice, green twice and yellow once.

A3 — dispute. Kai Kai says the spinner is clearly biased because red appeared seven of twelve times. Alicia says the short-run result is unusual-looking but one sample does not identify the true probabilities. Tricia proposes repeating the spinner many more times while also inspecting whether the sectors are actually equal.

Questions A

AQ1. Calculate the experimental probability of each colour.

AQ2. Compare red’s experimental probability with its theoretical probability. Does the difference prove bias?

AQ3. What additional evidence would distinguish random variation from a biased spinner more effectively?

AQ4. Explain why yellow’s single appearance does not make yellow impossible or even establish probability one twelfth.

AQ5. Write a calibrated conclusion.

Worked answers A

AQ1. Red is 7/12, approximately 0.583. Blue and green are each 2/12 = 1/6, approximately 0.167. Yellow is 1/12, approximately 0.083. These are experimental probabilities from this twelve-spin sample.

AQ2. Red’s theoretical probability is 1/4 = 0.25, so the observed relative frequency is considerably higher in this small sample. That discrepancy is evidence worth investigating, but it does not by itself prove bias. Random variation can produce uneven short runs. A structural inspection or much larger set of consistent trials would provide stronger evidence.

AQ3. Repeat many more spins under the same procedure and inspect the physical equality of the sectors and pointer. If the long-run red frequency remains far above one quarter and a mechanical asymmetry is found, the fair model becomes less credible. The combination of structural and empirical evidence is stronger than one short sequence.

AQ4. Yellow appeared once because that was the realised count in twelve trials. The true model probability is determined by the spinner structure, not simply by one observed proportion. A fair quarter-sector spinner can produce one yellow in twelve by chance. Experimental probability estimates the underlying chance; it does not define it exactly from a small sample.

AQ5. “In twelve spins, red occurred seven times, giving an experimental probability of about 0.583 compared with the fair-spinner model value of 0.25. The short-run difference is large enough to motivate further checking, but these twelve results alone do not establish that the spinner is biased.”

Transfer A. Suppose a further 188 spins produce 46 red results, making 53 red outcomes in 200 spins overall. The combined red frequency is 0.265, much closer to 0.25. Explain how the larger sample changes the judgement about the first twelve spins without changing what actually happened in them.

Laboratory B: With replacement or without replacement?

B1 — bag. A bag contains two red counters and two blue counters. Every counter is equally selectable when the bag is mixed.

B2 — Experiment One. Draw one counter, record the colour, replace it, remix, then draw again.

B3 — Experiment Two. Draw one counter, do not replace it, then draw again from the remaining three.

B4 — question. Compare the probability of red followed by red in the two experiments.

Questions B

BQ1. Are the two draws independent in Experiment One? Calculate P(red then red).

BQ2. Are they independent in Experiment Two? Calculate the conditional probability of red on the second draw given red first.

BQ3. Calculate P(red then red) without replacement.

BQ4. Explain why using one half × one half for both experiments is a design-reading error.

BQ5. Write a comparison using the words independent, dependent and conditional.

Worked answers B

BQ1. With replacement, the bag returns to two red and two blue before the second draw. The second-draw red probability remains 1/2 regardless of the first result, so the events are independent. P(red then red) = 1/2 × 1/2 = 1/4.

BQ2. Without replacement, after red first the bag contains one red and two blue. The conditional probability of red second given red first is therefore 1/3. The second event depends on the first result.

BQ3. P(red then red) = 1/2 × 1/3 = 1/6.

BQ4. The factor one half for the second draw is valid only when the original composition is restored. Without replacement, the first outcome changes the sample space and probabilities. Reusing the initial probability ignores the actual procedure.

BQ5. “The two draws are independent with replacement because the first draw does not change the second-draw probability. Without replacement they are dependent: after a first red, the conditional probability of red on the second draw falls from one half to one third.”

Transfer B. Change the bag to three red and one blue. Recalculate both red-red probabilities and explain which structural distinction stays the same despite the new numbers.

Laboratory C: Expected points and the result of one game

C1 — Game Safe. A player receives exactly 2 points.

C2 — Game Swing. A fair coin is tossed. Heads gives 4 points; tails gives 0.

C3 — Game Rare. A fair ten-sector spinner has one gold sector. Gold gives 20 points; any other sector gives 0.

C4 — choice. Alicia asks which game has the greatest expected value. Tricia asks which has the most predictable single result. Kai Kai says Game Swing is worse than Safe after he plays it once and gets 0.

Questions C

CQ1. Calculate the expected value of all three games.

CQ2. Which game has no outcome uncertainty? Which has the widest spread of outcomes?

CQ3. Does one zero from Game Swing show that choosing it was a bad decision if the criterion was highest expected value?

CQ4. Why might a decision-maker still prefer Safe even though all three games have the same expected value?

CQ5. Write a paragraph distinguishing expected value, outcome and risk preference.

Worked answers C

CQ1. Safe has expected value 2. Swing has 0.5×4 + 0.5×0 = 2. Rare has 0.1×20 + 0.9×0 = 2. All three share expected value 2.

CQ2. Safe has no outcome uncertainty because the result is always 2. Rare has the widest range, from 0 to 20. Swing ranges from 0 to 4.

CQ3. No. If the predeclared criterion was expected value only, Swing tied Safe and Rare. Receiving zero is an unfavourable realised outcome, but it does not retroactively change the probabilities or expected value known before play.

CQ4. A decision-maker may prefer certainty or have low tolerance for zero outcomes. Expected value does not capture every preference about variability. Choosing Safe can therefore be rational under a criterion that values predictable outcomes.

CQ5. “All three games have expected value two points, but their possible outcomes differ. Safe guarantees two, Swing gives either zero or four, and Rare gives zero most of the time with a small chance of twenty. A single loss does not show the expected-value calculation was wrong; it shows why a decision can depend on risk preference as well as long-run average.”

Transfer C. Change Rare’s gold reward to 30 points. Recalculate its expected value and explain how the expected-value ranking changes while its high variability remains.

Laboratory D: The forecast that was wrong once

D1 — forecast. A fictional forecast system issues one hundred independent event forecasts, each assigned probability 70%. The event occurs in sixty-eight of the hundred cases.

D2 — single case. On Case 101 the system again assigns 70%, but the event does not occur. Kai Kai says the system is unreliable because “70% was wrong”.

D3 — comparison. Alicia notes that 68% realised frequency is close to 70% across the first hundred cases. Tricia argues that a probabilistic forecast should not be judged like a promise of one outcome.

Questions D

DQ1. Explain why a 70% forecast can be well calibrated even though 30% of such events fail.

DQ2. Compare the stated 70% probability with the observed 68% frequency. Does the two-percentage-point difference establish poor calibration?

DQ3. Why is the Case 101 failure compatible with the forecast?

DQ4. What evidence would provide a stronger calibration assessment than one failed case?

DQ5. Write a corrected response to Kai Kai.

Worked answers D

DQ1. A 70% forecast means the event is expected to occur about seventy per cent of the time across many comparable forecasts, not on every occasion. Failures are part of the model’s stated uncertainty.

DQ2. Sixty-eight per cent is close to seventy per cent in this one hundred-case sample. The small difference could arise from normal variation. A calibration judgement should consider more cases and possibly several probability bands before declaring the forecasting system systematically over- or under-confident.

DQ3. A 70% event has 30% probability of not occurring. The Case 101 outcome is therefore one of the explicitly allowed possibilities. Its failure does not contradict the probability statement.

DQ4. Group many forecasts assigned near 70% and compare the stated probabilities with observed frequencies, ideally across a large suitable dataset. Repeating this across other probability levels would show whether the system is broadly calibrated or systematically overconfident.

DQ5. “The forecast did not promise that Case 101 would occur. A 70% forecast allows failure in about thirty per cent of comparable cases. The stronger test is whether events given about 70% probability occur at roughly that rate over many forecasts; the first hundred cases, with 68 occurrences, are broadly consistent with that.”

Transfer D. Suppose only forty of the next one hundred 70% forecasts occur. The combined two-hundred-case rate becomes fifty-four per cent. Explain why the evidence for calibration becomes much weaker and why the correct response is to investigate the model rather than excuse every discrepancy as random variation.

Laboratory E: The safer route that takes longer

E1 — objective. A fictional exhibition team must move materials from Room A to Room B. Route Short normally takes 10 minutes but has a 20% chance of a temporary corridor delay adding 15 minutes. Route Long always takes 14 minutes under the simplified classroom model.

E2 — consequence. Arriving later than 20 minutes misses a rehearsal slot. Therefore Route Short reaches in either 10 minutes or 25 minutes; Route Long reaches in 14 minutes. No other delays are included.

E3 — dispute. Kai Kai chooses Short because its expected time is lower. Alicia chooses Long because it guarantees arrival before the 20-minute threshold. Tricia says the right answer depends on the decision criterion.

Questions E

EQ1. Calculate the expected travel time for Route Short.

EQ2. Compare expected times for Short and Long.

EQ3. Compare the probability of missing the rehearsal threshold.

EQ4. Explain why Kai Kai and Alicia can both have coherent decision rules.

EQ5. Which route is robust if the team has zero tolerance for missing the rehearsal?

Worked answers E

EQ1. Short’s expected time is 0.8×10 + 0.2×25 = 8 + 5 = 13 minutes.

EQ2. Short has lower expected time at 13 minutes versus Long’s certain 14. If expected time is the sole criterion, Short is preferred.

EQ3. Short misses the 20-minute threshold exactly when the corridor delay occurs, probability 20%. Long always takes 14 minutes in this model, so its miss probability is zero.

EQ4. Kai Kai uses expected duration as his criterion. Alicia uses deadline reliability. Because the objectives differ, the same probability information can support different choices. Neither decision is irrational merely because it gives different weight to consequence and uncertainty.

EQ5. With zero tolerance for missing the rehearsal, Long is the robust choice under this simplified model because it stays below the threshold in every listed scenario.

Transfer E. Change the delay probability on Short from 20% to 5%. Recalculate expected time and discuss whether a decision based on expected time, deadline probability and risk tolerance could change.

Laboratory F: The game that became “due”

F1 — process. A fair coin is tossed independently. The first five outcomes are H, H, H, H, H.

F2 — claims. Kai Kai says tails is now more likely because the sequence must “balance out”. Alicia says the next toss remains one half heads and one half tails. Tricia distinguishes next-trial probability from the long-run tendency toward roughly equal proportions.

F3 — extension. After the five heads, another ninety-five fair tosses are made, producing forty-four heads and fifty-one tails. The full one-hundred-toss sequence therefore has forty-nine heads and fifty-one tails.

Questions F

FQ1. What is the probability of tails on toss six under the fair independent model?

FQ2. Identify the gambler’s fallacy in Kai Kai’s reasoning.

FQ3. Explain how the final 49–51 total can move close to half without tails being “due” on toss six.

FQ4. Does the long-run near-balance prove that the sixth toss had to be tails?

FQ5. How would the reasoning change if counters were drawn without replacement instead of tossing an independent coin?

Worked answers F

FQ1. Tails remains probability one half on toss six because the fair tosses are independent.

FQ2. Kai Kai treats earlier heads as creating a debt that the next toss must repay. Independence means the coin has no memory of the previous sequence. The streak affects our description of the past, not the probabilities of the next toss.

FQ3. The later ninety-five tosses happened to include more tails than heads, bringing the overall proportion close to one half. This is compatible with long-run stabilisation, but the adjustment occurred through many independent outcomes rather than a rule forcing an immediate tail.

FQ4. No. Many other one-hundred-toss sequences could also have occurred. The final near-balance does not retroactively make one intermediate outcome necessary.

FQ5. Without replacement, earlier draws can genuinely change later probabilities because the composition changes. The “due” criticism must therefore begin by checking independence rather than applying the coin-toss lesson mechanically to every chance process.

Transfer F. Replace the fair coin with a known biased coin whose heads probability is 0.7. After five heads, the next-head probability remains 0.7 under independent tosses. Explain why bias changes the probability value but does not create memory between tosses.

What the six laboratories teach together

Laboratory A separates short-run variation from evidence of bias. B shows how replacement changes dependence. C separates expected value from realised outcome and risk preference. D shows why probabilistic forecasts must be judged over many comparable cases. E combines likelihood with consequence and decision criterion. F exposes the gambler’s fallacy by separating independent next-trial probability from long-run frequency.

The common lesson is calibration. Probability language should be neither more certain nor more vague than the model and evidence justify. A learner who can explain why a 70% forecast may fail, why an unlikely event can occur and why a sensible decision can still turn out badly is using probability as a reasoning system rather than a collection of fractions.

Original story: The Good Decision That Lost

The following scene is fictional. It uses the Route Short and Route Long logic from Laboratory E but changes the setting and details so the vocabulary can work inside a narrative. It is not transport advice, risk advice or a report about real pupils.

Kai Kai chose Door B because the poster beside it said “FASTEST ROUTE”. The exhibition hall had two internal corridors leading to the rehearsal room. Door A opened onto the long corridor: fourteen minutes according to the staff timing sheet, every time in the simplified exercise. Door B opened onto the short corridor: ten minutes unless the movable display wall was being repositioned, in which case the wait added fifteen minutes.

The wall delay had occurred in four of the previous twenty comparable rehearsal movements. The class had agreed to use that one-fifth frequency as the working probability for the fictional planning exercise. Their rehearsal slot began in twenty minutes.

“Expected time is thirteen minutes,” Kai Kai said, writing the calculation on his clipboard. “Door A is fourteen. B wins.”

Alicia looked at the next line. “And the chance of missing the slot?”

Kai Kai had already calculated that too. If there was no wall delay, they arrived in ten minutes. If there was a delay, twenty-five. Under the model, one route had a twenty-per-cent chance of missing the rehearsal. The other had none.

“We said our criterion was average time,” he replied.

Tricia checked the planning sheet. The line under Objective did indeed say: choose the route with the lower expected journey time. The next line, added later in pencil, said: missing rehearsal is undesirable but not prohibited.

They went through Door B.

At the second turn, the movable wall was across the corridor.

For three seconds, nobody said anything. Then a member of staff raised a hand and pointed them towards a waiting area. “Fifteen minutes,” she said.

Kai Kai looked at the clock. “Wrong choice.”

“Bad outcome,” Alicia said.

“Same thing.”

Tricia shook her head. “Not if the decision rule was expected time.”

They missed the first five minutes of rehearsal. The team leader did not argue about probabilities in the corridor. She simply asked them to record what happened so the planning note could be discussed afterward.

Back in class, Kai Kai crossed out Door B on the worksheet. “Never again.”

Alicia took another sheet. “Suppose we ran the same choice a hundred times under the same model.”

“I know. B would usually be ten minutes.”

“Then why are you writing never?”

He stared at the word. The delay had been annoying enough to make the twenty per cent feel much larger than it had on paper. Before leaving the room, one in five had seemed like a number. After sitting beside the blocked wall, it felt like the whole experiment.

Tricia drew two headings: Decision quality and Outcome quality.

Under Decision quality, she wrote: Did we use the agreed objective? Were the probabilities and consequences represented correctly? Did we know the uncertainty before choosing?

Under Outcome quality, she wrote: What actually happened this time?

Kai Kai put the clipboard between the headings. “So a good decision can lose.”

“And a poor decision can get lucky,” Alicia said.

The team leader added another question. “Was expected time really the right criterion for us?”

That changed the discussion. If being late cost only a few minutes of informal practice, perhaps expected time was reasonable. If the rehearsal had been a one-time technical check that could not be repeated, the twenty-per-cent chance of missing it might matter much more. The probabilities had not changed. The consequence and objective had.

They tested three decision rules. Rule One chose the lowest expected time. It selected Door B. Rule Two required zero probability of arriving after twenty minutes. It selected Door A. Rule Three allowed Door B only when the delay probability was below ten per cent. With the current one-fifth estimate, it selected A.

None of the rules knew which route would be blocked on the next trip. They organised choices around different priorities.

Kai Kai erased never from his worksheet. In its place he wrote: “Door B produced a bad outcome this time. Whether it was a bad decision depends on the criterion chosen before the outcome was known.”

The sentence was longer. It was also harder to shout across a corridor. That was probably why “wrong choice” had arrived first.

At the bottom of the page, Tricia added one final line: “Probability does not tell us what will happen once. It helps us choose how to act before we know.”

Reading workshop: luck, hindsight and decision quality

Question 1. Why is “Wrong choice” an overstatement immediately after the delay? Worked answer: It judges the decision only from the realised outcome. Before the route was chosen, the class knew Door B had lower expected time and accepted a twenty-per-cent delay chance under its stated criterion. The unlucky delay does not change the information available beforehand.

Question 2. What changes when the team leader asks whether expected time was the right criterion? Worked answer: The probability model remains the same, but the value assigned to missing rehearsal becomes more important. Decision-making combines probabilities with objectives and consequences. A different criterion can rationally select another route without changing the chance calculation.

Question 3. Why does Kai Kai’s “never again” reflect overreaction to one outcome? Worked answer: One realised delay gives no new long-run probability beyond being one more observation. The emotional salience of the event makes it feel more informative than it necessarily is. A larger evidence update would require repeated comparable outcomes or structural information about the corridor process.

Question 4. Explain the difference between decision quality and outcome quality. Worked answer: Decision quality concerns whether the choice used appropriate information, probabilities, criteria and reasoning before the result was known. Outcome quality concerns what actually happened on that occasion. Uncertainty means the two can diverge.

Writing transfer. Rewrite the corridor event from Kai Kai’s point of view immediately after the delay, then add a reflective paragraph after the class discussion. Keep his first reaction emotionally convincing without presenting it as the final mathematical judgement.

Probability writing clinic: protect the event, denominator and time perspective

A probability sentence should tell the reader what event is being measured. “The chance is 30%” is incomplete unless the event is already unmistakable. “There is a 30% chance that the fictional delivery arrives after 4 p.m. under the model” gives the number a subject, condition and time meaning.

Keep the denominator visible when it changes the claim. Three successes in ten trials is not the same evidence as thirty in one hundred even though both relative frequencies equal 0.3. “One in ten” can describe a probability, an observed frequency or an expected rate depending on context. State which.

Preserve the time perspective. “Had a 70% probability” describes uncertainty before an outcome was known. “Occurred” describes what later happened. After the event, do not rewrite its prior probability as either zero or one merely because hindsight now knows the result.

Twelve precision repairs with explanations

1. “It happened, so its probability was 100%.” Repair: “The event occurred, but its probability before the outcome was known remained the model probability.” Realisation and prior likelihood are different.

2. “It has not happened yet, so it is impossible.” Repair: “It has not been observed in the current trials; impossibility requires probability zero under the model.” Absence in a finite sample is not impossibility.

3. “There are three possible results, so each has probability one third.” Repair: “Three labels do not imply equal probabilities; check whether the outcomes are equally likely.” Distinct is not equal-weighted.

4. “Seven heads in ten tosses proves the coin is biased.” Repair: “Seven heads is an uneven short-run result; stronger evidence is needed to distinguish random variation from bias.” Surprise is not proof.

5. “After five heads, tails is due.” Repair: “For independent fair tosses, the next tail probability remains one half.” The past sequence does not create a debt.

6. “Expected value two means I will get two.” Repair: “Expected value is a probability-weighted long-run average and may not be an available single-trial outcome.” Average is not promise.

7. “The 70% forecast was wrong because the event failed.” Repair: “A 70% forecast allows failure; calibration is assessed over many comparable forecasts.” Probabilistic forecasts are not categorical predictions.

8. “Route B was a bad decision because it was delayed.” Repair: “Route B had a bad realised outcome; decision quality depends on the predeclared criterion and information available before the delay.” Outcome and process differ.

9. “The rare event is low risk because it is unlikely.” Repair: “Risk depends on likelihood and consequence; a rare severe outcome can still matter.” Probability is only one dimension.

10. “The event is likely, therefore we should choose it.” Repair: “A decision also depends on the consequences, constraints and objective.” Most likely is not automatically best.

11. “More trials guarantee the theoretical probability.” Repair: “More trials usually stabilise relative frequency under a stable process, but finite results can still differ from the theoretical probability.” Long-run tendency is not exact finite balance.

12. “Uncertain means we know nothing.” Repair: “An outcome can be uncertain while the possible outcomes, probabilities and consequences are well characterised.” Uncertainty can be structured.

A six-session teaching sequence for probability and decision vocabulary

This sequence is a suggested route through the material, not a validated fixed programme or promise of mastery after six lessons. A session can be divided, repeated or reordered. Keep the learner’s active set small enough that the words can be applied in fresh problems instead of merely recognised beside familiar examples.

Session 1: outcome, event and sample space

Use entries 1–20. Start with simple dice, coin and spinner models described in words. Ask the learner to list outcomes, then build events from those outcomes. Compare possible, impossible, certain, likely and unlikely. Finish by presenting three distinct outcomes with unequal probabilities so the learner cannot rely on the shortcut that every listed outcome must be equally likely.

Record whether the learner can explain the difference between “possible” and “probable” without prompting. A useful exit question is: “Can an unlikely event happen?” The answer should be yes, with a modelled example. If unlikely keeps turning into impossible, return to the numerical scale before introducing more technical vocabulary.

Session 2: theoretical and experimental probability

Use entries 21–40 and Laboratory A. Calculate model probabilities first, then compare them with short-run observed frequencies. Repeat the reasoning with a larger supplied sample. The learner should explain why disagreement in a small sample can be ordinary variation and why increasing sample size does not repair a biased device or wrong model.

Introduce simulation only after the event and mapping rule are clear. Ask the learner to design a ten-digit simulation for probabilities 0.2, 0.3 and 0.7. They should justify how many digits represent success and recognise that the labels assigned to those digits do not matter when the random digits are equally likely.

Session 3: dependence, replacement and combined events

Use entries 41–50 and Laboratory B. Begin with replacement because the physical rule makes dependence visible. Ask what changes in the bag after the first draw. Then connect the changing composition with conditional probability. Only after the learner can explain the relationship in words should multiplication along a tree path become the main calculation.

Contrast mutually exclusive with independent explicitly. These words are often confused because both describe relationships between events. Give one pair that is mutually exclusive, one pair that is independent and one pair that is dependent but not mutually exclusive. The learner should justify the classification rather than select a label from appearance.

Session 4: expected value, variation and forecast language

Use entries 51–60 and Laboratories C and D. Compare a certain result with uncertain games sharing the same expected value. Ask why a long-run average need not appear in one trial. Then examine a 70% forecast that fails once. The learner should distinguish an event probability from a categorical promise and explain why calibration needs many comparable forecasts.

Use “expected”, “observed”, “forecast” and “scenario” in separate sentences. This helps prevent future estimates from being written as facts that have already happened. A strong learner should also recognise that a scenario can be plausible without having a supplied probability.

Session 5: risk and decision-making

Use entries 61–80 and Laboratory E. Separate probability from consequence before combining them. Compare a lower expected time with a lower chance of crossing a deadline threshold. Ask the learner to state the decision criterion before choosing. Then change the criterion and see whether the preferred option changes without any probability changing.

Introduce trade-off, precaution, mitigation and residual risk through low-stakes fictional planning. Avoid real medical, financial or safety-critical recommendations. The educational target is understanding that a decision uses probability but is not mechanically determined by probability alone.

Session 6: judgement, hindsight and communication

Use entries 81–100, Laboratory F and The Good Decision That Lost. Ask the learner to identify gambler’s fallacy, overconfidence and hindsight judgement. Then provide one decision before the outcome and ask for an evaluation after a bad realised result. The learner should preserve the information state that existed before the outcome was known.

Finish with a writing task: state one uncertain event, quantify or estimate its probability, identify a consequence, apply a criterion, choose an option and explain one limitation. A partner should be able to paraphrase the reasoning without turning “likely” into “certain” or “good decision” into “guaranteed good outcome”.

Assess the distinction the learner can actually use

Use specific observations rather than a vague label such as “good at probability”. Useful notes include “lists the sample space but assumes all recorded summaries are equally likely”, “calculates experimental probability accurately but treats it as exact”, or “distinguishes decision quality from realised outcome independently”. These observations lead to different next tasks.

Where helpful, mark demonstrated independently, demonstrated with support and not yet demonstrated in this task. Record the support supplied. A learner who selects the correct complement after being reminded that probabilities sum to one has shown partial control; a learner who identifies the complement spontaneously in a new setting has shown stronger transfer.

Test boundary cases. Use probability zero and one, exactly one half, an exact decision threshold, a repeated streak and an event that is unlikely but occurs. Boundary examples expose whether the learner understands the concept rather than merely associates a word with a middle-of-the-road example.

Questions about learning probability, risk and decision vocabulary

Must a Primary 6 learner know every term? No. Outcome, event, probability, likelihood, experimental probability, independent, dependent and fair are broadly useful. Expected utility, calibration, base-rate neglect and robustness are advanced extensions. A learner can understand the underlying idea in ordinary language before independently retrieving the technical term.

Is probability mainly mathematics? It is mathematical, but language determines what is being calculated and how the result is interpreted. Many errors occur before arithmetic: the wrong event, unequal outcomes treated as equal, replacement ignored, or an expected value mistaken for a guaranteed result.

Why include decision-making? Probability becomes useful when people use uncertainty to reason about choices. The decision section teaches that likelihood and consequence are different, and that a good decision can produce a bad outcome. The cases remain fictional and low-stakes rather than advising on real financial, medical or safety decisions.

Does a long streak mean the process has changed? Not necessarily. Streaks can occur in random independent trials. The right response is to examine the probability of the broader pattern, sample size and structural evidence rather than assume the next outcome must reverse or continue.

Should experimental probability equal theoretical probability? Not exactly in every finite experiment. Repeated trials often make relative frequency more stable around the model probability when the process is stable, but random variation remains. Large persistent discrepancies can motivate checking the model or device.

What should a parent correct first? Correct a meaning error before a calculation style issue. If the child treats three possible labels as equally likely without justification, fixing the fraction alone will not repair the underlying probability model. Ask the learner to explain the sample space and why each outcome has its assigned probability.

Reference notes and neighbouring eduKate routes

For the broader mathematical context, What Is Statistics? | Data, Variation, Probability, Inference and Decisions Under Uncertainty develops a full-world explanation of how probability fits inside statistical reasoning. This volume is narrower and learner-facing: it concentrates on the vocabulary and sentence control needed for upper-primary probability and decision explanations.

For numerical data language, use Volume 6: Charts, Tables and Data Interpretation. For fair tests, variables and experimental conclusions, use Volume 7: Experimental Design, Fair Tests and Scientific Conclusions. Volume 8 extends those skills into chance processes, forecasts, risk and choices under uncertainty.

For broad level-by-level vocabulary foundations, return to the Primary 6 Advanced Vocabulary Collection or the Vocabulary Article Directory. The collection is designed for selective use: choose the application that matches the learner’s current reading or writing problem rather than treating every volume as compulsory sequential homework.

The final principle: uncertainty is not ignorance

Probability is useful because the future can be uncertain while still having structure. We can know the possible outcomes, model their likelihoods, observe long-run frequencies, compare consequences and make defensible decisions without knowing what will happen next. The language of probability protects that middle ground between false certainty and helpless guessing.

Name the event. Check the sample space. Ask whether outcomes are equally likely, independent or conditional. Separate theoretical probability from observed frequency. Combine likelihood with consequence only when making a risk judgement. Choose a decision criterion before the outcome is known. Then communicate the result so another reader can tell exactly what is likely, what remains uncertain and why the choice is defensible.

Word families: change the grammar without changing the probability claim

Probable, probability and probably. “The event is probable” gives a verbal judgement of likelihood. “Its probability is 0.7” supplies a numerical measure. “The event will probably occur” communicates expectation but still does not promise occurrence. These forms share a root but do different grammatical jobs. A writer should not turn probably into certainly merely because the probability exceeds one half.

Possible, possibility and possibly. “Red is possible” says the event belongs to the allowed outcomes. “The possibility of red remains” names that status as a noun. “Red could possibly occur” may sound weaker but does not give a numerical likelihood. When an exact probability is available, use it instead of asking the adverb to carry information it does not contain.

Risk, risky and riskier. “There is a risk of delay” names the uncertain unwanted event. “Route B is riskier under the deadline criterion” compares options. The comparative requires a basis: higher probability of delay, greater consequence, or both. A writer should not call an option riskier merely because one unlucky instance occurred.

Expect, expected and expectation. “We expect thirty successes” under a 0.3 model across one hundred trials describes the model’s expected frequency. “The expected frequency is thirty” states the mathematical result. “Our expectation was thirty” can also describe a human prediction and should therefore be grounded in the model. Expected is not the same as guaranteed.

Decide, decision and decisive. “The group decides to take Route A” reports the choice. “The decision is based on a zero-delay threshold” states the rule. “The evidence is decisive” is a much stronger evaluation meaning it settles the matter sufficiently for the question. Similar spelling does not make these uses interchangeable. A decision can be made while the evidence remains uncertain.

Robust and robustness. “The choice is robust across probability estimates from 0.2 to 0.3” describes stability. “Its robustness comes from remaining below the threshold throughout that range” explains why. Do not replace robust with correct. A robust decision can still encounter an unfavourable outcome; it simply does not reverse easily when reasonable assumptions change.

Decision-matrix laboratory: when the winner changes with the criterion

The fictional choice. Three exhibition delivery plans have been modelled. Plan Alpha has expected completion time 12 minutes, a 25% probability of taking longer than 20 minutes, and a worst listed time of 30 minutes. Plan Beta has expected time 14 minutes, a 5% probability of exceeding 20 minutes, and a worst listed time of 24 minutes. Plan Gamma has expected time 16 minutes, zero probability of exceeding 20 minutes in the simplified model, and a worst listed time of 18 minutes.

Criterion 1 — lowest expected time. Alpha wins because 12 is lower than 14 and 16. This criterion accepts a substantial chance of crossing the deadline. It does not say the team prefers delay; it says expected speed has been given priority. If Alpha later takes 30 minutes, that realised outcome is unfortunate but does not change the earlier expected-time calculation.

Criterion 2 — probability of missing 20 minutes below 10%. Alpha becomes infeasible because its 25% miss probability exceeds the threshold. Beta and Gamma remain eligible. If the secondary criterion is expected time, Beta wins at 14 minutes. The preferred option changes because the decision rule changes, not because any probability has been recalculated.

Criterion 3 — zero tolerance for missing 20 minutes. Only Gamma meets the threshold in the supplied model. Its higher expected time is the trade-off for avoiding the listed late outcome. Calling Gamma “best” without naming this zero-tolerance criterion would hide the value judgement embedded in the choice.

Criterion 4 — minimise worst listed time. Gamma again wins at 18 minutes, followed by Beta at 24 and Alpha at 30. This criterion behaves conservatively toward the worst stated scenario. It ignores the probability distribution among other outcomes, so it should not be treated as universally superior to expected value.

Question 1. Which plan dominates another across all three supplied measures? Worked answer: Gamma has a higher expected time than Beta, so it does not dominate Beta despite having lower delay probability and lower worst time. Beta has higher expected time than Alpha but lower risk measures, so neither dominates there either. No plan dominates another across all three dimensions. A genuine trade-off remains.

Question 2. Suppose Alpha’s estimated miss probability is uncertain between 20% and 30%. Does that uncertainty matter under Criterion 1? Worked answer: Not directly if Criterion 1 uses only expected completion time and that expected time remains 12 minutes. The probability uncertainty matters under the deadline-threshold rules. This illustrates sensitivity: an uncertain input matters only when the decision depends on it.

Question 3. Suppose Beta’s miss probability estimate is 8% ± 4 percentage points rather than exactly 5%. Is the choice under the “below 10%” rule robust? Worked answer: No. The plausible range includes values below and above the threshold, so Beta’s eligibility is sensitive to the estimate. Better information about that probability has high decision value because it can reverse the choice.

Question 4. What additional information could matter even if all probabilities are correct? Worked answer: The consequence of lateness, cost of each plan, workload, flexibility, and whether the listed worst cases capture the relevant possibilities. Probability alone does not determine what the team should value.

Writing task. Write three separate conclusions, one for each of the first three criteria. Each conclusion should name the chosen plan, state the criterion, and identify the trade-off. Then write one sentence explaining why the three answers do not contradict one another.

Independent transfer brief: The mystery bag and the cautious planner

A sealed fictional bag contains one hundred counters. The manufacturer states that thirty are gold and seventy are grey, but the class cannot open the bag to inspect all counters. A drawing mechanism selects one counter at random, records its colour and returns it before the next draw. Across twenty trials, gold appears ten times. Across a second independent set of eighty trials using the same procedure, gold appears twenty-two times.

Task A — model and observation. State the claimed theoretical gold probability. Calculate the gold experimental probability after the first twenty trials and after all one hundred trials. Explain how the enlarged sample changes the evidence without rewriting the first twenty outcomes.

Worked answer A. The manufacturer claim implies theoretical probability 0.30 under random equal selection. The first twenty trials give 10/20 = 0.50. Across all one hundred, gold appears 10 + 22 = 32 times, giving 0.32. The early sample differs sharply from the model, while the combined sample is much closer. The first twenty remain a real unusually gold-heavy run; they do not need to be discarded for the larger sample to provide a more stable overall estimate.

Task B — judgement. Can the learner conclude that the bag contains exactly thirty-two gold counters because the experimental probability is 0.32? Worked answer: No. Replacement means the same physical counter can be drawn repeatedly, and experimental frequency estimates the selection probability rather than counting distinct contents. If the manufacturer statement is independently trusted, the model still says thirty gold counters. If the statement is uncertain, the trials provide evidence about selection probability, not a direct census of bag contents.

Task C — decision. A fictional planner receives one point for grey and loses two points for gold. They can either make one draw or take a certain score of zero. Under probability 0.30, calculate the draw’s expected value. Worked answer: Expected value = 0.70×1 + 0.30×(−2) = 0.70 − 0.60 = 0.10 point. The draw has slightly positive expected value, but it includes a 30% chance of losing two points. A planner who cares only about expected points may draw; one with a different risk criterion may choose the certain zero.

Task D — update. If the planner uses the combined experimental estimate 0.32 instead, expected value becomes 0.68×1 + 0.32×(−2) = 0.04 point. The preferred option under expected value remains the draw, but the margin becomes smaller. This shows that a decision can be robust to a moderate probability update even though the expected value changes.

Task E — communication. Write a final paragraph distinguishing the manufacturer’s model, the first-sample surprise, the one-hundred-trial experimental probability and the decision. A strong response will avoid saying that 0.32 is the “real number of gold counters” or that the positive expected value guarantees a gain.

Model response: “The manufacturer’s stated composition implies a gold probability of 0.30 under random selection with replacement. Gold appeared in half of the first twenty draws, an unusually high short-run frequency, but its combined frequency across one hundred trials fell to 0.32, close to the stated model. Using either 0.30 or 0.32, the fictional point game has a small positive expected value, although a single draw can still lose two points. The numerical choice therefore depends on the decision criterion as well as the probability estimate.”