# Accelerated Expertise

## Study guide, v3 revised

**Book:** *Accelerated Expertise: Training for High Proficiency in a Complex World*  
**Authors:** Robert R. Hoffman, Paul Ward, Paul J. Feltovich, Lia DiBello, Stephen M. Fiore, Dee H. Andrews  
**Edition:** Psychology Press, first edition. Copyright page 2014; Perlego metadata 2013.  
**Public study edition.**

**Scope.** These notes cover all sixteen chapters and the appendix. **Book** reports the authors' position; **My reading** and the applications connect it to your work. Figures are historical findings reported by the book, with study-specific limits discussed in the chapter notes and evidence table. Page references use the book's printed labels. These are paraphrases, not a collection of direct quotations.

This revision builds on v2 and v3. The practice and retention passages and selected appendix material were checked again in Perlego; the other chapters retain the earlier reading. The Commoncog connection and the Arthur retention paper were checked separately. See [source notes](#source-notes).

## Contents

- [Retrieval sheet](#retrieval-sheet) · [Central argument](#the-central-argument) · [Chapter notes](#chapter-notes)
- [Evidence table](#what-the-demonstrations-establish) · [Contrast pairs](#the-ideas-that-matter-as-contrast-pairs)
- [What this changes for your work](#what-this-changes-for-your-work)
- [Worked exercise](#worked-exercise) · [One decision worth revisiting](#one-decision-inside-the-trough)
- [Templates](#templates) · [Assessment](#assessment-of-the-book)
- [Recall questions](#recall-questions) · [Answer key](#answer-key) · [Glossary](#glossary) · [Source notes](#source-notes)

## Retrieval sheet

Twenty findings and distinctions to retrieve quickly. The chapter notes hold the detail.

1. **Five targets.** Faster initial training, earlier advanced proficiency, better retention, faster cognitive task analysis, and faster movement of field lessons into training. (Ch. 2, pp. 13-16)
2. **Curricula lag the field.** The authors report a six-to-nine-month delay in their operational context. (Ch. 2, p. 16)
3. **Three outcomes.** Robustness holds performance across conditions. Resilience restores it after disruption. Adaptivity changes the method. (Ch. 2, p. 14)
4. **Practice has a structure.** Amount, specificity, and deliberateness matter; time in grade proves little. (Ch. 3, p. 27)
5. **Simplification can become a trap.** Eight dimensions describe lost complexity; 23 knowledge shields describe ways to defend a flawed model. (Ch. 3, pp. 29-30)
6. **Triangulate proficiency.** Use career history, standards, measured performance, and peer identification together. (Ch. 3, pp. 26-27)
7. **Spacing's benefit shrank with complexity.** Across four task categories in a 63-study meta-analysis, effect sizes ranged from 0.97 to 0.07. (Ch. 4, p. 33)
8. **Feedback quality shapes expertise.** Type 1 domains offer clearer experts and feedback; Type 2 domains make both harder to establish. (Ch. 4, p. 40)
9. **Errors can teach.** Error-management training improved later performance when learners could interpret task feedback. (Ch. 4, p. 44)
10. **Transfer is earned.** Expertise at judging swine did not carry over to judging cattle. (Ch. 5, p. 49)
11. **One difference can matter enormously.** Similar tasks can fail to transfer because the viewpoint, equipment, or habitual response changes. (Ch. 5, pp. 60-61)
12. **Start the break proficient.** The authors identify pre-hiatus performance, including overlearning, as the best predictor of retained skill. (Ch. 6, p. 80)
13. **Retention has striking reported losses.** The book gives below 92% of prior performance after a year of non-use, and nearly 50% loss of newly learned declarative material within a day without review. See the measurement note in Chapter 6. (pp. 68-69)
14. **Knowing can return before doing.** A tank-gunner refresher restored knowledge scores but not scenario performance. (Ch. 6, p. 78)
15. **Guidance changes.** Less help on mastered steps, more coaching on newly hard problems. (Ch. 7, p. 85)
16. **Teams need selective sharing.** Everyone receiving everything is not the goal. (Ch. 8, p. 96)
17. **Sherlock compressed troubleshooting experience.** About 25 hours reportedly matched four years of experience on assessed troubleshooting tasks. (Ch. 9, p. 102)
18. **Just-in-time has a limit.** A reminder cannot supply the judgment a demanding situation already requires. (Ch. 12, p. 139)
19. **Train the combination.** If the job demands concurrent activities, practise their coordination. (Ch. 13, p. 153)
20. **Bring experience forward.** The authors' pre-burning illustration shifts earlier proficiency forward; it is not a fixed timetable for expertise. (Ch. 15, p. 170)

## The central argument

**Book.** Expertise develops slowly because ordinary work supplies the hard cases rarely and unevenly, and when a rare case does arrive a senior person usually handles it. Training can supply those cases deliberately, with expert feedback attached, and so compress the experience-feedback cycle. The opportunity is to bring consequential experience forward. The demonstrations show earlier proficiency in specific capabilities, not a universal timetable for becoming a complete expert.

The book is a research synthesis and a research agenda, funded by the US Air Force and written for a military and industrial readership. Its examples are pilots, forecasters, technicians, utilities engineers and intelligence analysts. Its theory is Cognitive Flexibility Theory (Spiro and Feltovich) merged with Cognitive Transformation Theory (Klein).

**My reading.** The practical test is whether the next unfamiliar case changes what you notice, predict, or do.

## Chapter notes

### Chapter 1. Why organisations need expertise (pp. 1-8)

**Book.** Senior practitioners carry the history, the interdependencies and the judgment that only becomes visible when normal operations fail. Retirement removes it. Knowledge repositories capture some of it but have been the biggest disappointment of knowledge management, because a document does not develop anyone's ability to recognise when it applies. The authors estimate that established methods can train around 85 percent of military jobs to journeyman level; the remaining work raises harder proficiency questions. Automation can hide complexity without removing it, so operators need more expertise precisely when automated help is least reliable.

**Details worth keeping.** The Hawley analysis of fratricide incidents, which illustrates automation bias and weaknesses in the use of automation and the crews had done what they were trained to do (p. 7). Journeyman to supervisor is a transfer across responsibilities, not across situations, and the old skills are still expected while the new ones are learned (p. 4).

### Chapter 2. What acceleration means (pp. 9-16)

**Book.** The chapter traces popular accelerated-learning claims to a 1965 book promising photographic memory and a language in twelve days. The authors separate themselves from that and from "relative acceleration," which is fixing bad training. Their five senses are in the retrieval sheet. Rapidized training has a stricter definition than finishing faster. It must also hold retention, and the methods that speed acquisition (massed, blocked practice) typically hurt retention.

**Details worth keeping.** The three properties, robustness, resilience, adaptivity (p. 14). The six to nine month lag (p. 16). The three requirements for rapid transposition, transfer from practitioner to trainer without adding workload, transformation into shareable products, transmission to the trainee (p. 16).

### Chapter 3. What changes as someone becomes expert (pp. 17-31)

**Book.** The declarative versus procedural split is called slippery and replaced with three kinds of knowledge, teachable, experiential and conjectural. The ten-year and 10,000-hour rules are approximate with wide variation. Livestock judges and auditors typically take 20 to 30 years. Time in grade guarantees nothing. The authors identify five conditions for developing high proficiency, critical skills for the mission, constant stretching by tough cases, high intrinsic motivation, rich feedback, and mentoring.

**Details worth keeping.**

- The seven-level craft guild scale the book uses throughout. Naive, novice, initiate, apprentice, journeyman, expert, master (Table 3.3, p. 25). The book drops "master" from its own scale because the term meant too many things in the guilds.
- The four scaling methods and the tripod rule (Table 3.4, pp. 26-27). In the Navy forecasting study the scale ran from junior apprentice to senior expert, with senior experts at as much as 50,000 hours.
- The eight dimensions of difficulty (Table 3.7, p. 29). Static or dynamic. Discrete or continuous. Separable or interactive. Sequential or simultaneous. Homogeneous or heterogeneous. Single or multiple representations. Mechanism or organicism. Linear or nonlinear. Learners read the second of each pair as the first, and that is the reductive tendency.
- Twenty-three knowledge shields have been catalogued. They are the mental moves that keep a wrong model intact when contrary evidence arrives (p. 30).
- Cognitive Flexibility Theory's four instructional rules. Multiple representations. No oversimplification. Case-based construction rather than transmission. Interconnected rather than compartmentalised knowledge (p. 31).

**My reading.** The dimensions table gives you a way to interrogate an explanation. Did it treat interacting causes as independent, or a moving process as a snapshot? Then look for the response to contrary evidence. A simplification becomes a shield when it is protected from a fair test.

### Chapter 4. Practice and feedback (pp. 32-48)

**Book.** The chapter's one confident generalisation is that generalisation is difficult. Much of the research uses students and simple tasks. Spacing usually helps retention, but one reviewed meta-analysis found a much smaller benefit for complex tasks. Interleaving slows acquisition and improves retention. Variable practice improves transfer. Teaching concepts before procedures produced better knowledge structures, but the learners were complete novices. Easy examples early teach simple rules that fail on hard examples. Guidance that helps novices can hurt advanced learners, the expertise reversal effect. Immediate feedback can block the reflection that produces understanding. Experts seek corrective feedback and fall down more in practice because they attempt what they have not mastered.

**Details worth keeping.**

- Donovan and Radosevich's 63-study meta-analysis. Across four categories from simple physical tasks to combined physical and mental tasks, spacing effect sizes dropped from 0.97 to 0.07; complexity correlated negatively with effect size, r = -0.25 (p. 33). This compares categories of tasks, not stages of one learner's progress.
- Spacing should be considered alongside the delay before later use or testing (p. 34). A schedule suited to tomorrow's test may not suit a skill needed months later.
- Type 1 and Type 2 domains (p. 40). The application to strategy consulting appears in the personal section below.
- The error-management meta-analysis, 24 studies, advantage after training but not during, and only with interpretable task feedback (p. 44).
- Bjork's desirable difficulties, delayed feedback, variable examples, mixed-part sequences, all worse now and better later (p. 45).
- The three traits of an expert mentor. Forms a rich model of the learner's knowledge, anticipates when the learner will form a reductive model, and can pick the case that forces the learner past it (p. 47).
- The chapter's own conclusion. Problem-based learning shows promise for dynamic domains. Feedback is necessary but timing rules are always tentative. Method depends on whether learning is initial, intermediate or advanced (p. 48).

### Chapter 5. Transfer (pp. 49-65)

**Book.** Transfer failures are easier to find than successes. Skill at judging swine does not transfer to cattle. Americans driving in England is negative transfer at the primary task, but almost everything else, keys, mirrors, seat, transfers fine, so a failure at the headline task does not mean nothing transferred. The authors contrast common elements (Thorndike) with general experience (Griffith). The first concerns overlap between tasks, the second transfer of principles through analogy. Neither reliably predicts a given case. Physical fidelity is not the critical feature. Cognitive fidelity is, and low-fidelity simulations have trained complex skills.

**Details worth keeping.**

- Haskell's nine senses of transfer (Table 5.1, p. 52). Content to content, procedure to procedure, declarative to procedural, procedural to declarative, self-aware strategy, cross-subdomain, vertical, cross-domain by analogy, lateral skill.
- The book's own definition. Transfer is when something learned in one context is used successfully in another that has important similarities and substantive differences (p. 53).
- The transfer surprises (Table 5.2, pp. 60-61). Aircraft to drone piloting fails on one difference, angle of approach versus angle of view. Rifle range safety to pistol range fails on muzzle length. Rodeo to jousting fails because riders look back on exiting the gate. B-52 to B-1 succeeded and the pilot credited confidence.
- The self-trained group in the luggage-screening study performed worst during training and best afterwards (p. 56).
- Healy's six principles for promoting transfer. Deliberate practice, variable practice, contextual interference, mixed schedule, distributed practice, explicit error analysis with process and outcome feedback (p. 64).
- The default prediction is no transfer, yet expertise is defined by coping at the edge of the envelope, so some transfer must exist at the boundary (p. 65).

### Chapter 6. Retention and decay (pp. 66-83)

**Book.** The usual decay curve drops quickly, then flattens. Faster relearning shows that some learning can survive even when immediate performance has declined. The factors are retention interval, overlearning, task type, how learning and retention are measured, retrieval conditions, training method and individual differences. The authors present overlearning as the most important modifiable determinant of retention. Simple motor skills barely decay over a year. Instrument flying decays more than basic flying. The measurement problem is severe. One errorless trial is often the criterion for mastery, and testing itself provides practice.

**Details worth keeping.**

- Two months after training, over half of Army trainees can still do tasks under nine steps and under 10 percent can do tasks with fifteen (p. 68).
- The book reports performance below 92 percent of its pre-break level after a year of non-use, citing Arthur et al. (1998) (p. 68). It also describes nearly 50 percent forgetting within a day for newly learned declarative material without review (p. 69). These are different tasks, time intervals, and measures.
- In one military study, higher- and lower-ability soldiers forgot at similar rates from different starting points (p. 69). Starting level and rate of loss are separate quantities.
- Interviewed F-16 refresher trainers estimated returns below 80 percent on standard manoeuvres after more than a few months away, and 30 to 40 percent after long breaks (p. 74). These are practitioner estimates. Four tracks depend on time away, from 242 academic modules and 38 sorties for eight-plus years out down to 165 modules and 11 sorties for under three.
- The two attitudes on return. "I have to catch up" belongs to the more expert. "This will be easy" belongs to mid-level pilots and is the dangerous one (p. 74).
- Pilot interviews, four factors. Skill before the break, resemblance of the staff job to flying, practice during the break, and the design of the return course. "An ace before, an ace afterwards" (p. 77).
- The up-or-out critique. The Air Force moves good people off the job because they are the future, and the Canadians, British and Dutch do not, which allows mentoring (p. 77).
- The tank gunner refresher day that fixed knowledge and not performance (p. 78). Overlearning's advantage on geology facts gone by week nine (p. 82). Student tutors retained more than the students they tutored (p. 82).
- The two headline generalisations. Pre-break proficiency predicts retention; recreating relevant learning conditions supports retrieval (p. 80).

**Measurement note.** Arthur et al.'s paper itself uses the below-92-percent wording, but its measured result after more than 365 days is a standardised difference of d = -1.4. That is not a raw percentage of skill lost or retained. Keep the reported wording attached to its source rather than converting it into an individual forecast. [Original study, pp. 57 and 85](https://gwern.net/doc/psychology/spaced-repetition/1998-arthur.pdf).

**My reading.** A break creates two problems, lost fluency and a changed world. The F-16 material is the clearest statement in the book that refresher training has to address both.

### Chapter 7. Problem and scenario-based training (pp. 84-91)

**Book.** Technicians troubleshoot by telling stories, and case libraries of expert stories improve prediction and explanation. Problem-based learning suits advanced learners; worked examples suit beginners. The guidance-fading model is incomplete because expertise requires increasing difficulty with mentoring added, not withdrawn. Situation awareness has three levels, noticing, understanding and projecting, and a measure of it predicted simulator performance better than flying hours did.

**Details worth keeping.**

- The assistance dilemma. Current theory does not say how much help to give at first, or when and how fast to withdraw it (p. 86).
- Virtual reality plus knowledge. A virtual valve with expert commentary hyperlinked to its parts, estimated to cut maintenance time by five to ten percent, and the models were used to elicit the knowledge in the first place (p. 88).
- Situation awareness originally meant getting inside the opponent's decision loop. That meaning has been lost in most later uses (p. 89).
- The aviation maintenance situation-awareness course rated "very useful" by participants with no performance measure and no control, which the book flags as notable (p. 91).

### Chapter 8. Team training (pp. 92-100)

**Book.** A team of experts is not an expert team. Complex cognitive work is teamwork by nature. Communication is exchanging information. Coordination is sequencing interdependent actions. Collaboration is helping another member reach their goal. A shared mental model does not require identical information in every member's head. Transactive memory is knowing who knows what. Some mixed-membership studies found a tradeoff between immediate performance and later teamwork skill. Training with perturbations, static on the comms line, produced the best performers, but the causal path was unclear.

**Details worth keeping.**

- The 2010 meta-analysis did not detect differences in training effectiveness between civilian and military teams, lab and field, or ad hoc and intact teams, so the student literature may generalise (p. 92).
- "Seven people, one command line." A systems administration incident where seven people troubleshot through one person with access, each wanting something different, and the root cause was one person's misreading of a configuration parameter (p. 94).
- Psychological safety and collective efficacy as trainable attitudes. Feedback to individuals works better delivered to the whole team in the team context (p. 97).
- Navy pilots who received simulated combat training out-performed Air Force pilots who did not, in the lull in the Vietnam air war (p. 98).

### Chapter 9. Demonstrations of accelerated expertise (pp. 101-119)

**Book.** The demonstrations range from intelligent tutors, with reported gains around one standard deviation, to surgery, perceptual training, business simulation, and military decision exercises. Their results and boundaries differ.

**Details worth keeping.**

- Sherlock. Senior apprentices with three years' experience beat experts with ten-plus after tutoring. About 25 hours equalled four years on the job, because the real world mostly supplies the routine and rare cases go to the established experts (p. 102). The book's caution is that Sherlock trains fixed tasks, and complex work is not fixed tasks (p. 103).
- Navy information technology tutor. Two weeks inside a sixteen-week course. Tutored students solved 97 percent of problems with 90 percent rated excellent, versus experienced technicians at 85 percent and 60 percent. New technicians moved from the 50th to the 98th percentile, and on advanced content matched twelve years of experience (p. 103).
- Teacher coaching. The teachers who got the most coaching were the new ones, so floor effect and Hawthorne effect are both live explanations (p. 104).
- Surgical cognitive task analysis. Better learning, faster procedures, fewer errors, but the "novices" were medical students and residents, so the level shift was probably junior journeyman to journeyman (p. 105).
- Perceptual training. Temporal occlusion, a few hours of low-fidelity practice, gains in novice and moderately skilled performers, and methods that educate attention rather than teach rules hold up better under stress (p. 106).
- The biotech operational simulation. Backorders as high as $800,000 per day. Four teams of 20 to 25, each playing twice on consecutive Fridays, one business quarter compressed so each week ran fifteen minutes. The best first run was worse than the worst second run. After five years, backorders around $1,100 (pp. 109-111).
- Decision-making exercises. A map, a one-page narrative building to a dilemma, no single right answer, playable in a lunchroom. The point is not whether the first prediction is right but how fast the person improves (p. 112).
- Think Like a Commander. Experience trumped rank on the vignettes. Simple vignettes may beat costly simulation. Instructorless practice worked in the Army studies but not in the Dutch replication, where scaffolding was required (pp. 115-116).
- Top Gun. Navy kill ratio went to 12 to 1 while the Air Force stayed at 2 to 1, over 100 engagements (p. 117). National Training Center reported post-training success rates of 30 to 1 for light infantry, 15 to 1 for combined arms, 5 to 1 for brigades (p. 118).
- Table 9.1, what effective scenario and engagement training shares. Tailored to the learner. Built from lessons learned and from how experts apprehend problems. Tough for the level. A daunting adversary. Fidelity as high as needed and no higher. Expert observers. Multiple reviews with outcome and process feedback, including emotional state (p. 119).

### Chapter 10. Domain case studies (pp. 120-131)

**Book.** The authors argue that established second-language programmes leave limited room to accelerate native-level fluency. The historical utilities examples report a median workforce age of 50 and 65 percent of senior engineers nearing retirement, and expertise at the "franchise expert" level takes 25 to 35 years because the equipment spans decades. Fighter pilot proficiency is measured in flying hours, 500 tactical hours as the standard, but hours are no longer the right kind. Research in real domains is slow. One retention study planned for eighteen months took three years.

**Details worth keeping.**

- Five of 1,500 people at a plant are irreplaceable (p. 122).
- The retired relay protection engineer replaced by four engineers with ten years each, and brought back when a car hit a pole and both lines tripped across fifty-year-old relay schemes (p. 123).
- The book uses $35,000 as the cost of an F-22 flight hour (p. 124).
- Table 10.1, the eight career paradoxes (pp. 129-130). The policy-proficiency paradox, promotion requires breadth which prevents depth. The culture-measurement clash, people must look perfect to advance so nobody measures development. The myth of the career path. The irony of transitioning, two-month courses give exposure not comprehension. The ironies of demotivation. The ironies of reassignment, moved off the job just as they become expert. Under-utilised resources, debrief records nobody mines. The consequences of limiting resources, fewer flying hours means more retraining later.
- Twelve years of operational assignments is the country's return on a pilot's training, and anyone can be moved off the path after two (p. 128).
- Intelligence analysts spend up to half their time on briefings and reports, and on returning to an analytical role nobody knows them and their mentors are gone (p. 127).

### Chapter 11. Forging a theory (pp. 132-137)

**Book.** Cognitive Load Theory splits load into intrinsic, extraneous and germane, and says design should minimise the second. The authors' objection is that real work is full of distractions and workarounds, and training that strips them out does not prepare anyone. Cognitive Flexibility Theory says advanced learning is the ability to assemble knowledge from multiple cases and perspectives to fit the case at hand, and incremental complexification breeds reductive models. Cognitive Transformation Theory says learning is the elaboration and replacement of mental models, and a better model is harder to abandon because it explains away more. The two are merged into one syllogism.

**Details worth keeping.**

- Some workers plateau because they are so effective at defending flawed models. High proficiency includes the ability to lose confidence in your own model (p. 136).
- Cognitive Flexibility Theory is called lopsided. It does not say where the sweet spot is, and a learner who over-complexifies may be paralysed (p. 136).
- The merged syllogism (Table 11.3, p. 137). All models are limited. Training must help learners past reductive explanation. Shields lead to wrong diagnoses. Learning therefore includes unlearning, and advanced learning is promoted by interconnected cases and concepts along multiple dimensions.

### Chapter 12. Designs for accelerated learning (pp. 138-146)

**Book.** Proficiency requires practice before it is needed. Returning practitioners must recover fluency and adapt to changed work. Motivation, causal feedback, and cognitively relevant practice matter. The authors favour tough cases scaled to proficiency, with expert help where needed. The design must preserve the relationships that make the real problem difficult.

**Pair.** Skill versus fact. The authors contrast stronger retention but narrower transfer of skills with weaker retention but broader transfer of facts (p. 140). The earlier chapters contain exceptions to this shorthand.

**Details worth keeping.**

- Just-in-time disaster (p. 139).
- Ericsson's three-step expert performance approach and the ExPerT fourth step, with the book's list of its limits, unsuited to changing domains, ill-structured tasks, and domains with few experts (pp. 141-142).
- Expertise is not holding a rich mental model. It is the ability to regenerate models as needed (p. 145).
- Concept-case coupling. Principles are the vehicle for transfer, but they only become usable through repeated application to cases, including cases where the principle must be modified or does not apply (p. 145).
- Mastery training over getting individual problems right (p. 146).

### Chapter 13. The focus of training (pp. 147-155)

**Book.** Perceptual learning concerns relations among cues across data and time. Experts recognise when a pattern is unreliable and switch to deliberation. Changing technology requires relearning perception. A knowledge model describes stored understanding; a mental model is the explanation active in the situation. Training must preserve the combined demands of real work.

**Details worth keeping.**

- The pattern ladder. Single cues, sets of separable cues, combinations, relations among cues, featureless family resemblances, integral cue configurations, and dynamic event-based configurations (p. 149). The forecaster's tornado signature is a nest of relations, not a feature.
- The moving target rule. Change the technology and you change the work even when the domain does not change (p. 150).
- Psychological support skills, mental rehearsal, self-talk, goal focus, can be trained and may accelerate novices (p. 151).
- The functional task principle. Train two things at once if the job demands two at once (p. 153).
- The twelve traits of the expert mentor (p. 154). Models decisions in context. Supports the learner under uncertainty and load. Helps juniors form first models. Reads the learner's model fast and predicts where it will go wrong. Crafts cases that push to the next level. Uses anomalies to trigger loss of faith in a model. Works real problems aloud with the apprentice. Teaches the apprentice to generate their own feedback. Scaffolds recognition of a flawed model by building a richer one. Trains on ill-structured problems. Uses extreme and non-routine cases. Emphasises interconnection.
- Experts do not necessarily make good mentors. The ones who make the job look easy often describe it worst (p. 154).

### Chapter 14. Training materials, methods and strategies (pp. 156-164)

**Book.** Training is contingent on domain. Perturbations belong in team training even where they are rare in the field. Klein's "deliberate performance" collects feedback during and after real work as the counterpart to deliberate practice. Three proficiency gaps matter, initiate to junior apprentice, senior apprentice to junior journeyman, senior journeyman to junior expert. Case libraries need to be large, well indexed, and scaled by level with "level stretchers." A retrospective case is not yet a scenario. Scenarios must carry the mess, interruptions, missing data, competing goals. Operational simulations and decision exercises differ from conventional simulation because they let new solutions emerge instead of testing adherence to a designed one.

**Details worth keeping.**

- Feedback must be timely, satisfying to sensemaking through causal explanation, and informative of both outcome and process (p. 158).
- The four-part scenario package. A first-person narrative, second-person analytical commentary, a timeline broken into sub-events, and a coded analysis of that timeline marking observations, decisions, actions, surprises, options and information requirements (p. 161).
- Build the neat version first, then create messy variants (p. 161).
- Scenarios should sometimes mislead the learner toward the common solution when the real cause is a little-known issue (p. 162).
- The authors describe groups reproducing their characteristic workplace failures in operational simulations (p. 163).
- There is no validated method for walking into an organisation and finding who is, or could become, a good mentor (p. 164).

### Chapter 15. Roadmapping demonstrations (pp. 165-173)

**Book.** The proposed research programme brings together ten kinds of specialist. Rapidized cognitive task analysis is itself an unsolved research problem. Measurement must work beyond the ceiling of easy tests, because routine accuracy tests can stop distinguishing experts who continue improving. Compression packs rare cases closer together or removes empty time inside scenarios. The cohort-select design tracks groups starting at different proficiency levels in parallel. The economic illustration is hypothetical.

**Details worth keeping.**

- The "5 percenters," the senior experts whose achievement rests on predisposition, curiosity and drive as much as training (p. 167).
- The Cochran-Weiss-Shanteau index, discrimination over inconsistency, as a way to distinguish performance after an easy accuracy test reaches its ceiling (p. 168).
- Pre-burning. The illustrative curve brings earlier proficiency forward while retaining a long development period (Fig. 15.1, p. 170). It illustrates an opportunity; it does not establish that later development cannot be accelerated.
- The economic illustration. 100 workers, two years of schoolhouse, three years to journeyman, five to junior expert, salaries $75,000 to $95,000, a two-year saving totalling $15 to $28 million (p. 172). Illustrative, and see the assessment below.
- The proposed vehicle is a university-industry research initiative over five to eight years (p. 173).

### Chapter 16. Summary (pp. 174-180)

**Book.** The summary restates the five targets and acknowledges counterexamples to its generalisations. Its research agenda combines contrasting domains, faster expertise analysis, level-specific cases, mentors, and time compression. Career rotation and the gap between isolated tasks and real work remain obstacles.

**Details worth keeping.** The generalisation that experienced teams may communicate overtly less than weak ones (p. 177). The four difficulties in tracing training through to operational results (p. 178), a problem of attribution when many influences change together.

### Appendix. Proficiency scaling in weather forecasting (pp. 181-196)

**Book.** Two studies. The first scaled 22 Navy forecasters by career interviews, hours, breadth of experience and verification scores. The second, 65 Air Force forecasters, used a knowledge audit and a sorting task to derive four reasoning styles.

**Details worth keeping.**

- The hours table runs from about 1,850 hours for junior apprentices to about 55,400 for senior experts (Table A.2, p. 184). The Gilbreth correction estimates time on task at roughly half of time on duty (p. 185). The useful distinction is what the hours contain; halving the largest estimate does not make it equal to a universal 19,000-hour threshold.
- Verification data were available for only eight of the 22 interviewees, with routine scores between 81 and 85 percent. In the hard summer months experts averaged 83 and journeymen 80. Journeymen are often as good as experts, but when they miss they miss badly (p. 187). Forecast amendment biases scores upward (p. 187).
- The four descriptive styles. Scientist, builds a causal model first, uses many tools, may act like a Mechanic under stress. Proceduralist, completes a knowledge-based procedure, more time on guidance than on the model. Mechanic, fixed procedure, not motivated, disrupted by tool changes. Disengaged, dislikes the job, minimal products (Table A.4, pp. 190-191).
- Expert recognition can take longer than a journeyman's because it integrates cue configurations across data types rather than firing on one cue (p. 195).

## What the demonstrations establish

| Demonstration | Reported | Supports | Does not establish |
|---|---|---|---|
| Sherlock tutor | About 25 hours comparable to four years on assessed troubleshooting; apprentices beat experienced comparators | Concentrated cases and coaching can produce large gains on a bounded skill | Complete occupational expertise or a universal hours-to-years conversion |
| Navy IT tutor | 97 vs 85 percent solved; 50th to 98th percentile; matched twelve years | Rapidized training and accelerated proficiency with shipboard transfer, in one programme | A universal tutoring effect size |
| Surgical CTA training | Better learning, faster procedures, fewer errors | Multiple-expert elicitation reveals content ordinary explanations omit | Learners starting without medical expertise, or one consistent proficiency-level shift |
| Perceptual video training | Gains in hours, transfer to live play | Anticipation can be trained off-field | That recognition alone delivers the whole skill |
| Biotech operational simulation | Backorders reportedly fell from as high as $800,000 per day to around $1,100; five-year follow-up | Simulation, changed coordination, and operational improvement occurred together | Training alone causing the gains, or the same result elsewhere |
| Think Like a Commander | Better tactical analysis; experience beat rank | Vignette practice improves domain-specific judgment cheaply | Domain-general critical thinking, or whether scaffolding is needed (studies split) |
| Top Gun and NTC | 12 to 1 vs 2 to 1; 30 to 1, 15 to 1, 5 to 1 | Strong reason to study structured engagement training | Isolation of any single ingredient |
| Teacher coaching | Improvement with wide variation | Reason to study coaching conditions | That the gain was expertise rather than floor or attention effects |

## The ideas that matter, as contrast pairs

**My reading.** These distinctions are the questions to carry into work.

| Distinction | Question it raises |
|---|---|
| Rapidized / accelerated | Same standard sooner, or a higher standard sooner? |
| Duration / structure of experience | Which consequential decisions did those years contain? |
| Simplification / knowledge shield | Is the explanation useful, and will it survive a fair test? |
| Guidance fading / guidance changing | What is mastered, and where is help still needed? |
| Common elements / general experience | Which shared element or principle makes transfer plausible? |
| Lost fluency / changed world | What was forgotten, and what is now different? |
| Outcome / process feedback | What happened, and what in the reasoning should change? |
| Type 1 / Type 2 domain | How would we know who is good, and whether a decision worked? |
| Knowledge model / mental model | Can the stored knowledge produce a useful explanation now? |
| Archive / exercise | Where must the learner decide before seeing what happened? |
| Isolated task / actual work | Which simultaneous demands disappeared from the exercise? |
| Deliberate practice / deliberate performance | What can we learn in training, and during real work? |

## What this changes for your work

**My reading.** These applications connect the book to case comparison, a collection of evidence notes, and a prediction journal. They are proposed uses, not systems validated by the book.

### Commoncog gives the theory a practical form

The connection is explicit. Commoncog identifies Cognitive Flexibility Theory and *Accelerated Expertise* as foundations of its case library, using linked concepts and cases to develop business judgment. [Commoncog's explanation](https://commoncog.com/c/about/).

That gives your comparison questions a purpose. What is surprisingly similar? What is surprisingly different? The first helps you recognise a concept in an unfamiliar form. The second stops the concept from becoming a rule pasted onto every company. A good comparison changes what you would notice or predict in the next case.

### Cases and evidence notes serve different parts of learning

An evidence note gives you a claim to retrieve. A case gives that claim conditions, a sequence, and consequences. Reading both can improve understanding; practising requires a decision of your own.

For a selected case, separate the evidence available then from what became clear afterwards. Stop at one consequential choice. Record the decision, prediction, and confidence. Reveal the outcome, discuss the reasoning, and choose a related case that might defeat the lesson. The existing material becomes the raw material for practice without needing to rebuild the whole library.

### Strategy work needs a feedback channel

Strategy consulting often has the Type 2 problem. Outcomes arrive late, several causes move together, and a persuasive explanation can outlive the evidence for it. This makes your prediction journal useful.

Write the prediction before the result, including what would change your mind. Later, distinguish a good decision from a favourable outcome. Revisit patterns across decisions. The ledger cannot make a noisy result conclusive, but it can stop the original claim from changing unnoticed after the fact.

### A library probe can prepare cognitive task analysis

Searching evidence notes and then the wider library can identify recurring failures and candidate incidents. Take those leads into an expert interview. What did the practitioner notice? What alternative did they reject? What mattered that never reached the written account?

This preserves the useful connection without equating search with cognitive task analysis. Retrieval finds the account; the interview investigates the judgment behind it.

### Use the eight dimensions when reading an argument

For a filing or business explanation, ask which interacting causes were treated separately, which moving process became a snapshot, and which viewpoint was omitted. Then ask what evidence would distinguish the offered explanation from an alternative. Complexity is useful when it changes the test, not merely when it produces a longer account.

### A difficult period contains a case worth building

A period of stalled progress contains many decisions. Select one attempt to break it, or the moment your explanation changed. That is where the book's method becomes concrete. The prompts below are aimed at that incident, not a retrospective story about the whole period.

## Worked exercise

**Fictional practice case.** The purpose is to choose evidence that separates explanations and revise a decision as information arrives. The example illustrates the method; it is not a completed assessment of retail expertise.

### Stage 1. Uneven sales

A brand runs a promotion across several stores. Total units rise. Some stores barely move. The first recommendation is a deeper discount in those stores.

Write your answers before continuing.

1. What could explain the difference?
2. Which evidence would you request first, and which explanations would it separate?
3. What do you predict it will show?
4. What would you do now, and how confident are you?

**Stop here before revealing Stage 2.**

### Stage 2. Availability evidence

The low-growth stores had frequent shelf gaps. Price and display were executed as planned. The strongest early-selling products ran out soonest.

How does this change your explanation, recommendation, and confidence? What remains unresolved? Name evidence that would challenge the availability explanation.

**Stop here before revealing Stage 3.**

### Stage 3. Contribution remains poor

Availability is restored in a second test. Units improve. Contribution from the promotion is still poor.

What is the question now? Which evidence would you request next, and what conclusion must wait until you see it?

**Stop here before reading the facilitator notes.**

### Facilitator notes

Have participants commit individually before discussion. Evaluate how they choose evidence and update their explanation, not whether they guessed stockouts immediately.

Stage 1 offers a sales pattern, not a cause. Stage 2 increases the plausibility that supply constrained observed sales. It does not settle every question about price or demand. Stage 3 changes the problem from unit sales to contribution. Plausible lines include margin per unit, product mix, promotion costs, substitution, and purchases pulled forward from later periods. The supplied facts do not choose between them.

Ask which assumption changed, which evidence changed it, and whether the stockout diagnosis became an answer to everything. For a transfer test, use a different case with strong availability and uneven execution. Later, remove these prompts. Look for independent investigation rather than repetition of the previous answer.

## One decision worth revisiting

**Interview prompts for your own case.** Choose the first intervention you thought would break the plateau, or a specific decision where your understanding changed. Give it a date and a clear beginning and end. These prompts adapt the incident-focused approach in Chapter 14 (pp. 160-161).

1. What happened that made you decide to act then?
2. What did you believe the constraint was, in the words you would have used at the time?
3. What information was available before that decision? What became known only later?
4. What options did you consider, and why did you reject the others?
5. What did you predict your chosen action would produce, and by when?
6. What actually happened? Which record could verify the sequence?
7. Which observation changed your explanation, if any? What did the revised model predict?
8. Which competing goals, people, or practical constraints shaped the choice?
9. What might a more experienced operator have noticed earlier? What would a junior have missed?
10. What difference in another business would make this lesson fail?

Build the learner version up to that choice. Keep the later evidence and your interpretation in the facilitator version. Preserve uncertainty where the outcome does not identify the cause. A second case should test the lesson's boundary.

## Templates

### Case record

```markdown
# Case title
Source and date:
Domain and subdomain:
Intended learner level (guild scale):
Capability practised:
Why this case is useful (which dimension of difficulty it exercises):

## Learner version
Situation and goal:
Information available now:
Constraints and competing demands:
Decision required:
Prediction required:

## Staged information
Next evidence:
Next decision:

## Facilitator version
Actual outcome and source:
Expert account of the cues that mattered:
Reasoning and alternatives:
What remains uncertain:
Likely learner misconception (which reductive simplification):
Debrief questions:

## Follow-up
Related but different case:
What changes in that case:
Delayed assessment:
Conditions that require updating this case:
```

### Decision and debrief record

```markdown
Before the outcome
My interpretation:
My prediction:
My chosen action:
My confidence and reasons:
Missing information that matters:
Evidence that would change my mind:

After the outcome
What happened:
What I had right:
What I missed or misread:
Was the outcome informative or partly ambiguous:
Which assumption changes:
The next case that would test the revision:
```

### Mentor's diagnosis

```markdown
What the learner can already do:
What the learner currently seems to believe:
Evidence for that diagnosis:
The specific gap:
Case chosen to address it:
Guidance to give now:
Guidance to withhold for now:
What independent improvement would look like:
What happened on the next attempt:
```

## Assessment of the book

**My reading.** Its strongest contribution is connecting cases, mentoring, retention, and career design. An organisation can preserve knowledge while preventing anyone from developing the judgment to use it.

**The evidence needs sorting.** Experiments, meta-analyses, interviews, and historical accounts answer different questions. The evidence table keeps their boundaries visible. The biotechnology result is particularly interesting because learning and organisational redesign happened together.

**The theory is useful without settling every design choice.** Cognitive flexibility explains why cases need multiple perspectives. Cognitive transformation explains why learning sometimes requires abandoning an explanation. Neither supplies a universal difficulty setting. The book recommends worked examples for beginners and calls its preferred theory lopsided.

**Read the review more closely than the summary.** The overlearning advantage disappears by week nine in one study; the concepts-first result involved complete novices. Those conditions matter more than a broad concluding slogan.

**Keep the units straight.** Sherlock's years-equivalent concerns troubleshooting. The Gilbreth correction distinguishes duty hours from task time. The retention paper reports a standardised effect, not a universal percentage-loss rule.

**The economics are illustrative.** Faster proficiency may create value, but development-period salaries are not automatically cash savings. Compare incremental benefit with expert time, case preparation, delivery, and measurement. The book's military costs, demographics, and research gaps belong to its early-2010s context.

## Recall questions

Answer before consulting the [answer key](#answer-key).

1. How can a shorter course fail to accelerate expertise?
2. What separates repetition from deliberate development?
3. Why is one correct answer weak evidence?
4. What distinguishes a knowledge shield from a useful simplification?
5. Must a surprising result overturn the model?
6. How do outcome and process feedback differ?
7. When is difficulty useful?
8. How could a vignette beat an elaborate simulator?
9. How do retention and transfer differ?
10. What are the two forms of time compression?
11. Why is mentoring distinct from expertise?
12. Why should a team avoid sending everything to everyone?
13. What two problems should a refresher address?
14. How does the Type 2 problem apply to strategy work?
15. What would demonstrate learning from this guide?

**Stop and write your answers before continuing.**

## Answer key

1. Course speed may improve without later independent performance, retention, or transfer. Specify the capability and comparison.
2. Development selects challenges, exposes weaknesses, provides feedback, and tests improvement. Repetition may stay inside routine.
3. It may be easy, familiar, lucky, or heavily assisted. Examine reasoning and performance across unfamiliar cases.
4. A useful simplification supports the decision and remains testable. A shield protects an inadequate model from contrary evidence.
5. No. Compare explanations and seek discriminating evidence; neither reflexive defence nor reflexive abandonment is learning.
6. Outcome concerns the result. Process concerns the reasoning and actions that produced it.
7. When it exercises the target capability and improves subsequent learning. Errors alone are not evidence of value.
8. It may preserve the relevant decisions, uncertainty, and feedback better than physical realism does.
9. Retention is capability after time passes. Transfer is capability under meaningful changes in task or context.
10. Pack rare cases closer together; remove empty intervals within scenarios. Watch for fatigue and lost timing or coordination demands.
11. A mentor must diagnose the learner's model and choose guidance. Performing well does not confer that skill automatically.
12. Members need relevant shared understanding and knowledge of who knows what. Indiscriminate sharing can overload them.
13. Lost fluency and a changed world. Restoring an old method does not establish readiness for changed work.
14. Feedback is delayed and ambiguous. Recording predictions helps prevent hindsight from rewriting the original judgment.
15. Better independent decisions, diagnosis, or case design later and in another relevant situation, beyond recalling vocabulary.

## Glossary

| Term | Meaning |
|---|---|
| Journeyman | Reliably handles ordinary work without close supervision |
| Expert | Demonstrates exceptional domain capability, including tough cases |
| Reductive tendency | Simplification that loses relationships needed to understand the situation |
| Knowledge shield | Defence of an inadequate model against conflicting evidence |
| Cognitive flexibility | Assembling knowledge across cases and perspectives to fit the situation |
| Cognitive transformation | Revising or replacing the model used to understand the work |
| Knowledge model | External representation of what someone knows |
| Mental model | Active explanation used to interpret and anticipate a situation |
| Cognitive task analysis | Methods for eliciting the knowledge and reasoning involved in work |
| Critical Decision Method | Incident-focused interview into cues, judgments, options, and decisions |
| Overlearning | Practice beyond an initial performance criterion |
| Savings | Faster relearning attributable to previous learning |
| Cognitive fidelity | Similarity of the thinking required in training and at work |
| Transactive memory | Knowing who on the team knows what |
| Concept-case coupling | Connecting principles with varied situations where they apply or fail |
| Deliberate performance | Seeking feedback and improvement during real work |

## Source notes

[Perlego edition](https://ereader.perlego.com/1/book/1617094/10?element_originalid=fm4&page_number=xv). All sixteen chapters and the appendix have notes above; printed page spans appear in their headings. Image-only tables and figures were not comprehensively checked in this revision.

Focused checks revisited Chapter 4's spacing comparison, Chapter 6's retention and overlearning discussion, and the appendix's proficiency measures. The [Arthur et al. study](https://gwern.net/doc/psychology/spaced-repetition/1998-arthur.pdf) supplies the measurement clarification in Chapter 6. [Commoncog's own account](https://commoncog.com/c/about/) supports the connection to its case method. Other underlying studies have not been independently reassessed.

**In simple terms:**

- Difficult cases and useful feedback can deliver experience sooner than routine work does.
- Your library becomes more useful for learning when you make decisions before revealing outcomes and test the lesson on another case.
- The prediction journal can keep your original predictions visible when hindsight begins rewriting them.
- Start your own exercise with one consequential decision from your own experience.