Build the skills to define worthwhile projects, measure processes credibly, distinguish association from causation, test improvements, and sustain results. This guide contains twenty original multiple-choice practice questions and sixty worked concepts across the six ASQ CSSGB domains, using the supplied 2022 Body of Knowledge linked by ASQ in October 2026. It is a representative preview, with domain allocations of 2/4/4/4/3/3 questions and 7/12/12/11/9/9 concepts. Explanations emphasize assumptions, calculations, and plausible decision errors. Scope is framed for October 17, 2026; no future version is represented. Source tags establish syllabus relevance; worked technical explanations use established subject knowledge.
What this guide covers
Practice 20 original CSSGB questions and learn 60 concepts spanning Six Sigma, DMAIC, statistics, and lean.
Overview: Six Sigma and the Organization
1. Shewhart, Deming, Juran, and Ishikawa
Six Sigma combines several established quality ideas rather than relying on one statistical technique. Shewhart distinguished common and special causes through statistical control. Deming emphasized understanding systems and variation, learning through improvement cycles, and management responsibility. Juran's trilogy separates quality planning, quality control, and quality improvement. Ishikawa advanced cause-and-effect analysis and broad employee participation in quality. These contributions guide different decisions: stabilize unusual variation, design quality into work, organize improvement, and investigate plausible causes. A Green Belt selects tools according to the problem rather than treating every quality problem as an inspection shortage.
Worked example: A packaging team first uses a Shewhart chart to identify an unusual shift. It then organizes possible causes with an Ishikawa diagram and uses Juran's planning perspective to revise packaging requirements before testing an improved method.
Common mistake: Attributing all quality improvement to final inspection overlooks the distinct roles of process stability, management systems, planning, and employee knowledge.
Official syllabus reference: [1]
2. SMART Goals and Balanced Process Metrics
A project should connect a measurable process result to an organizational driver such as customer satisfaction, cost, efficiency, or market performance. SMART goals specify the intended result, measurement, feasibility, relevance, and time boundary. Primary metrics show whether the target improves; consequential metrics detect harmful side effects. A scorecard helps prevent local optimization by displaying several linked outcomes. Process feedback also matters: changing an upstream input can alter downstream workload, inventory, or quality. A target is therefore incomplete if it improves one department while transferring greater cost or delay elsewhere.
Worked example: A service team targets a reduction in median approval time from five days to three within one quarter. Its scorecard also tracks returned applications and overtime, so faster approval cannot conceal increased rework or unsustainable staffing demands.
Common mistake: A numerically precise departmental target is not automatically useful when its effects on customers and downstream processes remain unmeasured.
Official syllabus reference: [1]
3. Constraint Capacity, Takt Time, and Pull
Takt time equals available production time divided by customer demand and expresses the required demand pace. Cycle time describes actual processing pace. Theory of constraints focuses improvement on the resource limiting system throughput; extra capacity elsewhere may merely create inventory. Just-in-time and pull approaches link replenishment to downstream consumption rather than unrestricted upstream production. These ideas interact but are distinct: takt expresses demand, the constraint limits supply, and pull governs authorization to produce. Variability, downtime, and product mix must be considered before treating nominal machine speed as sustainable effective capacity.
Worked example: A cell has 420 available minutes and demand for 140 units, giving takt of three minutes per unit. Its slowest operation requires four minutes. Improving a two-minute operation will not resolve the cell's current demand-capacity gap.
Common mistake: Equating takt with observed cycle time hides the distinction between the customer's required pace and the process's actual capacity.
Official syllabus reference: [1]
4. Value Stream Maps and Direct Flow Observation
A value stream map links material and information flow across an end-to-end process, including processing time, waiting, inventory, and scheduling signals. It distinguishes customer-valued transformation from necessary non-value-added work and avoidable waste. Inspection and rework generally do not transform the product in a way the customer values, although some checks remain necessary. Gemba observation verifies how work actually occurs. A spaghetti diagram complements the map by tracing physical movement. Together these tools reveal whether long lead time arises from processing, queues, transport, storage, or excessive movement rather than assuming every delay needs faster equipment.
Worked example: A laboratory map shows 18 minutes of testing within a two-day lead time. Gemba observation finds samples waiting for batch release, while a spaghetti diagram reveals repeated trips to shared storage. Queue and layout changes become stronger candidates than faster testing.
Common mistake: Calling every required activity value-added prevents the team from distinguishing customer-valued transformation from currently necessary support work and avoidable waste.
Official syllabus reference: [1]
5. DMADV, IDOV, Verification, and Validation
DMAIC improves an existing process through Define, Measure, Analyze, Improve, and Control. Design for Six Sigma addresses new designs or substantial redesign when the existing architecture cannot meet requirements. DMADV means Define, Measure, Analyze, Design, Verify; IDOV means Identify, Design, Optimize, Verify. Both develop requirements and evaluate a proposed design before release. Verification checks whether specified design requirements were met. Validation evaluates whether the resulting product or process serves its intended use and customer need. Passing a technical specification does not alone establish that the chosen specification represents what users actually need.
Worked example: A team designing a refrigerated delivery container verifies that prototypes maintain the specified temperature range in defined tests. It validates the design through representative delivery use, checking whether handling, loading, and actual routes meet customers' intended needs.
Common mistake: Treating verification and validation as interchangeable can produce a design that meets its written specifications but fails its intended use.
Official syllabus reference: [1]
6. FMEA Ratings and Risk Priority Numbers
Failure mode and effects analysis links a potential failure mode to its effects, causes, current controls, and recommended actions. Traditional RPN multiplies severity, occurrence, and detection-difficulty ratings defined by the team's scales. Higher detection-difficulty ratings usually indicate weaker detection, but the scale must be confirmed. Equal RPNs can represent very different risks, and rating scales are ordinal rather than precise physical quantities. Severity-based rules and the nature of the consequence can therefore outweigh simple RPN ranking. Actions should target prevention or detection, identify ownership, and be reassessed after their effectiveness is established.
Worked example: A missing seal receives severity 8, occurrence 3, and detection difficulty 5, producing RPN 120. An interlock prevents cycling without a seal. After verification, the team reassesses occurrence; it does not lower severity merely because prevention improved.
Common mistake: A lower occurrence rating does not automatically reduce severity, because severity describes the consequence if the failure actually occurs.
Official syllabus reference: [1]
7. Design FMEA versus Process FMEA
Design FMEA examines how a product or service design could fail to perform its intended functions. Process FMEA examines how execution, manufacturing, assembly, or service delivery could introduce failures. The analyses connect but address different causal mechanisms. A component may be inadequately specified even when manufactured perfectly, while a sound design may fail because it is assembled incorrectly. Design controls include analysis, specification changes, and design tests; process controls include fixtures, parameter controls, and error prevention. Keeping the distinction clear helps assign action to the team capable of changing the relevant cause.
Worked example: A connector loosens because its specified locking geometry is insufficient: that is a design issue. The same connector can also loosen when assembly torque is missed: that is a process issue requiring a different prevention and verification strategy.
Common mistake: Assigning every field failure to process FMEA can leave an inadequate product design unchanged even when assembly fully meets instructions.
Official syllabus reference: [1]
Define Phase
8. Project Selection and Comparable Benchmarking
DMAIC suits a meaningful, measurable performance gap in an existing process when causes or solutions require investigation. A known, low-risk correction may need a simpler improvement approach, while a fundamentally new design may call for DfSS. Selection considers strategic alignment, customer impact, data access, resources, scope, and expected benefit. Benchmarking supplies comparison opportunities: competitive benchmarking studies competitors, collaborative benchmarking shares information with partners, and best-practice benchmarking can cross industries. Comparisons become misleading when definitions, product mix, service commitments, or exposure differ. A benchmark identifies a possible gap, not proof that another organization's method will transfer unchanged.
Worked example: A distributor compares order accuracy with a retailer but discovers that one counts errors per order and the other per line item. After aligning definitions and product mix, it selects a bounded picking-error project rather than copying the retailer's target.
Common mistake: Selecting the largest reported benchmark gap without checking comparable definitions can launch a project around a measurement artifact.
Official syllabus reference: [1]
9. SIPOC Boundaries and Stakeholder Ownership
SIPOC summarizes suppliers, inputs, the high-level process, outputs, and customers. Its first practical decision is the start and end boundary, because those boundaries determine which activities and handoffs belong in the project. Inputs are resources or information entering the process; outputs are its results. Customers may be internal or external. The process owner is accountable for ongoing process performance, while other stakeholders may supply resources, receive consequences, or approve changes. Cross-functional boundaries require particular attention because one function's output often becomes another's input, creating opportunities for unclear requirements and hidden queues.
Worked example: For purchase-order approval, the team starts at receipt of a complete request and ends at an approved order. Requesters supply information, procurement owns the process, and receiving is a downstream customer that needs accurate order details.
Common mistake: Drawing a SIPOC before agreeing on boundaries can mix upstream causes, project activities, and downstream customers into inconsistent categories.
Official syllabus reference: [1]
10. Voice-of-Customer Sampling and Question Design
Voice-of-customer collection seeks evidence about needs from relevant internal and external customer segments. Interviews explore meaning, focus groups reveal shared and conflicting views, surveys estimate structured responses, and observation reveals behavior customers may not articulate. Each method has selection and response limitations. Questions should avoid leading language, double-barreled wording, vague time frames, and ambiguous scales. Complaint records overrepresent people who report problems and may miss silent dissatisfaction. Effective collection links the sampling frame, customer segment, question wording, and intended decision so the team knows whose needs the evidence actually represents.
Worked example: Instead of asking whether delivery is 'fast and reliable,' a team asks separately about promised-date performance and elapsed delivery time during the previous month. It samples both frequent and occasional customers rather than relying only on complaint callers.
Common mistake: A large number of responses does not remove bias when the invitation method systematically excludes an important customer segment.
Official syllabus reference: [1]
11. CTQ Trees and Operational Requirements
A critical-to-quality tree decomposes a broad customer need into drivers and measurable requirements. For example, dependable delivery may involve completeness, timeliness, and damage-free condition. Each terminal requirement needs an operational definition describing the unit, event, measurement method, and acceptance rule. CTX extends the same reasoning to other critical outcomes, such as cost or safety. A requirement should express the needed result without prematurely prescribing a solution. Customer importance and business feasibility inform prioritization, but existing performance should not silently become the acceptance standard merely because it is easy to measure.
Worked example: Customers request 'accurate invoices.' The team separates price accuracy, quantity accuracy, and correct customer identification. For quantity accuracy, it defines a defect as any invoice line whose billed quantity differs from the accepted shipment record.
Common mistake: Naming a metric such as turnaround time is insufficient until its start, stop, unit, and acceptance requirement are operationally defined.
Official syllabus reference: [1]
12. QFD Relationships and Kano Requirement Types
Quality function deployment translates customer requirements into technical characteristics through an explicit relationship matrix. Customer importance weights and relationship strengths help identify which characteristics deserve design attention, while a correlation roof can reveal technical tradeoffs. The Kano model distinguishes must-be requirements, performance requirements, and attractive features. Missing a must-be requirement creates dissatisfaction, whereas improving an attractive feature may delight customers without making its absence unacceptable. These categories can change with customer expectations. QFD weights support structured judgment, but numerical totals depend on the scoring assumptions and should not conceal contradictory or weak customer evidence.
Worked example: For a booking service, correct reservations are a must-be need, shorter confirmation time is a performance need, and a helpful optional itinerary may delight users. QFD links those needs to validation checks, response latency, and recommendation features.
Common mistake: A high weighted QFD score does not prove a feature is essential when the underlying relationship ratings or customer importance estimates are weak.
Official syllabus reference: [1]
13. Project Charters, Scope, and Consequential Metrics
A charter authorizes a bounded improvement effort by stating the business case, problem, baseline, goal, scope, team, ownership, and expected timing. The problem statement describes an observed gap without embedding an unverified cause or favored solution. Scope specifies included products, locations, process boundaries, and exclusions. Primary metrics measure the targeted improvement; consequential metrics check for unwanted effects. Pareto analysis can help select a major defect category, but a large category is not automatically a proven cause. Charter changes should be explicit when new findings alter resources, boundaries, or objectives.
Worked example: A charter reports that 14% of complete applications required correction last quarter and targets 7% within four months. It includes one application type, excludes policy redesign, and tracks approval delay to detect any burden created by extra checks.
Common mistake: Writing 'install automation to fix careless staff' as the problem statement assumes both the solution and cause before analysis.
Official syllabus reference: [1]
14. WBS, Critical Paths, PERT, and Project Cadence
A work breakdown structure decomposes deliverables into manageable work packages. A Gantt chart displays scheduled activities over time, while a network diagram makes dependencies explicit. The critical path is the longest-duration dependent path and determines earliest completion under stated resource assumptions. PERT commonly estimates duration as (optimistic + 4 × most likely + pessimistic)/6. Shortening one activity may expose another critical path. Top-down planning establishes overall milestones and constraints; agile methods use iterative delivery and feedback where requirements or solutions evolve. Neither approach removes the need to manage dependencies and acceptance criteria.
Worked example: A pilot requires fixture preparation and operator training before execution. Preparation takes six days and training four, both after a two-day design task. With a three-day pilot, the controlling path takes eleven days, assuming parallel resources are available.
Common mistake: Adding every activity duration assumes all work is sequential and overstates completion time when independent tasks can proceed in parallel.
Official syllabus reference: [1]
15. Risk Registers, Tollgates, and Project Closure
A project risk register records uncertain events, their causes, potential consequences, ownership, and responses. Feasibility, resource availability, schedule, cost, stakeholder effects, and business continuity can all shape risk. Tollgate reviews assess whether phase evidence supports proceeding, revising, or stopping; completion of a slide deck is not sufficient evidence. Documentation should preserve definitions, assumptions, decisions, and results for reproducibility. Closure compares achieved outcomes with the charter, transfers ownership, records lessons, and identifies further opportunities. Communication should match the audience, using upward escalation, downward direction, and horizontal coordination where appropriate.
Worked example: A project depends on a supplier delivering trial material. Its risk register assigns an owner, a backup sourcing option, and a trigger date. At closure, the team records actual benefits, unresolved limitations, and the process owner's accepted control responsibilities.
Common mistake: Closing a project when the pilot succeeds leaves a gap if ownership, documentation, and sustainment responsibilities have not been transferred.
Official syllabus reference: [1]
16. Affinity, Interrelationship, Tree, and Matrix Diagrams
Management and planning tools organize different forms of complexity. An affinity diagram groups unstructured observations by natural themes without imposing a premature causal hierarchy. An interrelationship digraph explores directional influence among related issues, although arrows remain hypotheses until supported. A tree diagram decomposes an objective into progressively more specific means or requirements. A matrix diagram displays relationships between two or more sets, such as requirements and process steps. Selecting the tool depends on whether the immediate need is grouping, influence exploration, decomposition, or relationship mapping. None of these diagrams alone proves causation or priority.
Worked example: A team groups interview comments into themes with an affinity diagram, explores influences among themes with a digraph, decomposes the chosen objective with a tree, and maps resulting requirements to responsible functions in a matrix.
Common mistake: Treating the most connected digraph node as a proven root cause confuses a structured team hypothesis with verified causal evidence.
Official syllabus reference: [1]
17. Prioritization Matrices, PDPC, and SWOT
A prioritization matrix compares alternatives against explicit criteria and, when appropriate, weighted scores. Weights should reflect decision importance, and sensitivity analysis reveals whether small changes reverse the ranking. A process decision program chart anticipates what could go wrong during a proposed plan and identifies countermeasures. SWOT distinguishes internal strengths and weaknesses from external opportunities and threats. An activity network then sequences the chosen work through dependencies. These tools serve different decisions: choosing, preparing contingencies, assessing strategic conditions, and scheduling. Combining them is useful when numerical rankings alone would overlook feasibility or execution risk.
Worked example: A team scores three fixture designs on quality, cost, and changeover time. It tests whether modest weight changes alter the winner, uses PDPC to prepare for installation failures, and schedules approved tasks through their predecessor relationships.
Common mistake: A weighted score does not eliminate judgment; unstable rankings or poorly defined criteria can create numerical precision without a dependable decision.
Official syllabus reference: [1]
18. DPU, DPMO, RTY, and Cost of Poor Quality
Defects per unit equals total defects divided by units inspected; one unit may contain multiple defects. DPMO equals defects divided by total defined defect opportunities, multiplied by one million. Opportunity definitions must remain comparable across products and periods. Rolled throughput yield multiplies conditional first-pass stage yields to measure end-to-end passage without failure. Cost of poor quality includes internal and external failure costs such as scrap, rework, returns, and warranty handling. Translating defect rates into sigma levels requires an explicitly stated distribution and shift convention; the number is not meaningful without those assumptions.
Worked example: Inspecting 200 forms reveals 30 defects across four defined opportunities per form. DPU is 30/200 = 0.15, and DPMO is 30/800 × 1,000,000 = 37,500. The defective-form percentage still requires counting how many forms contain defects.
Common mistake: Using defect count as defective-unit count overstates the number of affected units whenever one unit can contain multiple defects.
Official syllabus reference: [1]
19. Team Development, RACI, and Structured Decisions
Teams commonly move through forming, storming, norming, performing, and adjourning, with recognition supporting closure. Conflict can reflect unresolved roles rather than lack of commitment. RACI identifies who is responsible, accountable, consulted, and informed for a deliverable. Champions and sponsors secure organizational support, Black Belts provide methodological guidance, and process owners sustain results. Brainstorming generates options; nominal group technique adds silent generation and structured sharing; multivoting narrows a list. Facilitators should address domination, reluctance, groupthink, and premature agreement. Clear communication channels help distinguish evidence, interpretation, decisions, and unresolved concerns.
Worked example: A dominant specialist shapes every meeting. The facilitator first collects ideas silently, then gives each person a turn before multivoting. A RACI clarifies that the process owner approves the revised workflow while the Green Belt coordinates analysis.
Common mistake: Apparent agreement does not demonstrate consensus when quieter members have lacked a structured opportunity to present evidence or concerns.
Official syllabus reference: [1]
Measure Phase
20. Process Maps versus Written Work Instructions
A process map represents actual sequence, decisions, rework loops, and handoffs. Written procedures describe required arrangements, while work instructions provide task-level execution detail. Comparing them reveals gaps between intended and actual practice. Swimlane maps make functional ownership and cross-department transfers visible. A useful decision symbol states a question with clearly labeled branches, and a rework loop returns to the correct step. The map should reflect observed work rather than merely reproduce an approved document. Differences can indicate obsolete instructions, informal adaptations, unclear criteria, or missing controls and should be investigated before drawing conclusions.
Worked example: An approved procedure sends incomplete requests back immediately, but observation shows clerks holding them in a shared queue. A swimlane map identifies the queue owner and exposes an unrecorded delay that the original procedure does not describe.
Common mistake: A clean flowchart copied from a procedure can conceal the very workarounds and handoffs that explain actual process performance.
Official syllabus reference: [1]
21. Data Types and Measurement-Scale Boundaries
Continuous data represent measured quantities such as time or diameter, while discrete data represent counts or classifications. Nominal scales identify categories without order; ordinal scales add order without established equal spacing. Interval scales support meaningful differences but lack a meaningful absolute zero, as with Celsius temperature. Ratio scales also support meaningful ratios, as with elapsed duration or mass. The scale determines which comparisons have defensible meaning. Numeric coding does not upgrade a category's measurement properties. Separately, a defective-unit indicator differs from a defect count because one unit may contain several defects.
Worked example: A survey codes poor, fair, good, and excellent as one through four. The order is meaningful, but a response of four is not established as twice the satisfaction of two. Recorded waiting time in minutes supports a ratio interpretation.
Common mistake: Assigning numbers to ordered categories does not establish equal intervals or justify interpreting ratios between their codes as physical quantities.
Official syllabus reference: [1]
22. Random Sampling, Stratification, and Data Checks
Random sampling gives eligible units a defined selection chance and reduces selection bias when implemented correctly. Stratified sampling divides the population into relevant groups and samples within each, improving representation or precision. Disproportionate stratum sampling requires appropriate weighting for an overall estimate. A collection plan defines the unit, variables, timing, method, sampling frame, ownership, and quality checks. Check sheets and coding rules make collection consistent. Range checks identify impossible values, while missing and duplicate records require separate handling. Plausible extremes should be verified rather than automatically deleted because they may reveal real process conditions.
Worked example: A center samples 30 day-shift and 30 night-shift cases, although day shift handles 80% of volume. It reports shift-specific rates and weights the overall estimate by actual volumes instead of averaging the two rates equally.
Common mistake: Equal samples from unequal strata improve coverage but do not justify an unweighted overall estimate when population proportions differ.
Official syllabus reference: [1]
23. Center, Dispersion, and Distribution Displays
The mean uses every value and is sensitive to extremes; the median identifies the middle ranked value and is more resistant to skew. Range, variance, standard deviation, and interquartile range describe different aspects of dispersion. Sample variance uses squared deviations with denominator n − 1. Histograms show distribution shape but depend on binning. Boxplots summarize median and quartiles, stem-and-leaf plots preserve individual values, and cumulative frequency plots show proportions at or below thresholds. A normal probability plot assesses compatibility with normality. No distribution summary replaces time order when investigating process stability.
Worked example: Waiting times of 2, 3, 3, 4, and 18 minutes have mean six and median three. A boxplot highlights the long upper tail, while a time plot can show whether the 18-minute wait coincided with a specific disruption.
Common mistake: A histogram that looks approximately normal does not establish stability, because rearranging the same observations in time leaves the histogram unchanged.
Official syllabus reference: [1]
24. Probability Rules, Independence, and Counting
For any events A and B, P(A and B) equals P(A)P(B given A). Independence permits replacing the conditional probability with P(B). Mutually exclusive events cannot occur together, so their intersection probability is zero; positive-probability exclusive events are therefore not independent. The addition rule subtracts overlap when calculating the probability of either event. Permutations count ordered arrangements, while combinations count selections where order does not matter. Sampling without replacement can create dependence. Correct modeling begins by defining the event and sampling mechanism rather than applying a multiplication rule merely because two percentages are available.
Worked example: From ten distinct inspectors, choosing a three-person team gives C(10,3) = 120 possibilities. Assigning those three people to three different audit roles instead gives 10 × 9 × 8 = 720 arrangements because role order matters.
Common mistake: Mutually exclusive and independent describe different relationships; confusing them produces incorrect joint probabilities even when the individual event probabilities are known.
Official syllabus reference: [1]
25. Central Limit Theorem and Standard Error
The central limit theorem explains why averages of independent, identically distributed observations with finite variance tend toward an approximately normal sampling distribution as sample size increases. The original observations need not become normal. For independent observations, the mean's standard error is sigma divided by the square root of n, estimated using sample standard deviation when appropriate. Confidence intervals quantify uncertainty in a population parameter through a repeated-sampling procedure. Strong skew, heavy tails, dependence, or small samples can weaken simple approximations. Increasing sample size reduces random error but does not correct systematic measurement or selection bias.
Worked example: Individual service times are right-skewed, with standard deviation estimated at 12 minutes. For independent samples of 36 cases, the estimated standard error of the sample mean is two minutes, although individual service times remain right-skewed.
Common mistake: The central limit theorem concerns the distribution of a statistic across samples, not a promise that raw process observations become normal.
Official syllabus reference: [1]
26. Normal, Binomial, and Poisson Models
The normal distribution models continuous, symmetric variation through a mean and standard deviation. A binomial model counts successes in a fixed number of independent trials with a constant success probability. A Poisson model counts events within a defined exposure when events occur independently at a constant average rate; its mean and variance are equal under the model. Exposure must be comparable or explicitly incorporated. Clustering, changing rates, or dependence can invalidate simple count models. Model selection follows the data-generating mechanism, not merely whether a histogram appears vaguely bell-shaped or whether observations happen to be integers.
Worked example: The number of defective items among 20 independent inspected items fits a binomial setup. The number of surface blemishes over a fixed inspected area may fit Poisson assumptions. Measured shaft diameter instead requires a continuous model.
Common mistake: A count is not automatically Poisson; a fixed number of binary trials usually calls for a binomial model instead.
Official syllabus reference: [1]
27. Student's t, Chi-Square, and F Distributions
Student's t accounts for uncertainty from estimating population standard deviation when making mean inferences under suitable assumptions. Its tails are heavier than the normal distribution's, especially at low degrees of freedom. A chi-square distribution arises from sums of squared standard normal quantities and supports variance inference under normality and categorical goodness-of-fit or association tests through corresponding approximations. An F distribution describes ratios of independent variance estimates under normal-model assumptions and underlies variance comparisons and ANOVA. Degrees of freedom shape all three distributions. Test names alone do not remove requirements concerning independence, normality, or adequate expected counts.
Worked example: For a normal sample of ten measurements, a mean interval uses a t reference with nine degrees of freedom. Comparing two independent normal-sample variance estimates uses an F reference with separate numerator and denominator degrees of freedom.
Common mistake: Using the normal reference whenever a sample mean appears ignores the extra uncertainty involved when population variation must be estimated.
Official syllabus reference: [1]
28. Gauge R&R and Precision-to-Tolerance
Gauge repeatability is variation when the same operator repeatedly measures the same item under consistent conditions. Reproducibility captures differences associated with operators or other defined measurement conditions. A crossed gauge R&R study has operators measure the same representative parts repeatedly, with suitable randomization to reduce memory and order effects. The study separates measurement variation from part-to-part variation. Precision-to-tolerance compares a defined measurement spread with specification width; the multiplier and acceptance rule must be stated. Percentage of study variation answers a different question because its denominator depends on the sampled process or part variation.
Worked example: Using a stated six-sigma convention, measurement standard deviation of 0.01 mm and tolerance width of 0.20 mm give P/T = 6 × 0.01/0.20 = 0.30. The team evaluates that value against its explicitly established application requirements.
Common mistake: A low percentage of total study variation can result from widely separated sample parts and does not alone establish adequate tolerance discrimination.
Official syllabus reference: [1]
29. Bias, Linearity, Agreement, and Correlation
Bias is the difference between average measured value and a suitable reference value. Linearity examines how bias changes across the operating range, so it requires reference information at multiple levels. Stability concerns measurement behavior over time. For attribute systems, repeated classifications assess within-appraiser consistency, between-appraiser agreement, and agreement with a reference. Overall percent agreement can conceal poor performance in a rare but important category, so category-specific results matter. Measurement correlation evaluates association between methods but does not establish interchangeability: a method with a consistent offset or proportional error can correlate almost perfectly with a reference.
Worked example: Two instruments have correlation 0.99, but the new instrument reads approximately 0.4 mm higher across the range. Their association is strong, yet the offset prevents assuming agreement. Multiple reference levels help distinguish constant bias from changing bias.
Common mistake: High correlation between two measurement methods can coexist with substantial bias, so correlation alone cannot demonstrate that their readings are interchangeable.
Official syllabus reference: [1]
30. Cp, Cpk, Pp, Ppk, and Target-Sensitive Cpm
Cp equals specification width divided by six within-process standard deviations. Cpk uses the smaller distance from the mean to either specification, divided by three within-process standard deviations, and therefore reflects centering. Pp and Ppk use overall standard deviation instead. A gap between within and overall indices suggests variation across the broader observation period that deserves investigation. Cpm incorporates deviation from a target through a denominator based on both variance and squared mean-to-target distance. Index interpretation requires clear estimators, credible measurements, and appropriate distribution assumptions; a width-only index cannot establish acceptable centering.
Worked example: With limits 0 and 12, mean 8, and within standard deviation 1, Cp is 12/6 = 2. Cpk is min(4/3, 8/3) = 1.33, revealing the effect of the mean's proximity to the upper limit.
Common mistake: A high Cp can coexist with poor conformance when the process mean is sufficiently close to, or beyond, a specification limit.
Official syllabus reference: [1]
31. Capability Preconditions and Sigma-Shift Conventions
Specification limits express requirements; natural process limits describe process behavior under a statistical model. They are not interchangeable. A capability study first establishes measurement adequacy, representative sampling, process stability, and a suitable distributional approach. Normal-based formulas can misrepresent tail risk for skewed or mixed populations. Short-term within variation and longer-term overall variation may differ because of drift, shifts, or changing conditions. A conventional 1.5-sigma adjustment is an assumption used in some Six Sigma reporting, not a universal physical law. When data permit, characterize observed changes rather than automatically imposing that convention.
Worked example: A stable short-term study places the mean five standard deviations from the nearest specification. Reporting a long-term value of 3.5 sigma requires explicitly adopting a 1.5-sigma shift convention; it does not demonstrate that such a shift actually occurred.
Common mistake: Applying a fixed sigma shift without stating the convention can turn a reporting assumption into an unsupported claim about actual process behavior.
Official syllabus reference: [1]
Analyze Phase
32. Multi-Vari Studies: Positional, Cyclical, Temporal
A multi-vari study separates patterns at several levels of variation. Positional variation occurs within a unit or across simultaneous positions, cyclical variation occurs between successive units or cycles, and temporal variation occurs across longer time intervals. The sampling design must preserve those levels: several positions within each unit, several nearby units, and repeated time periods. Charts then compare within-unit spread, unit-to-unit changes, and movement over time. The strongest observed component guides further investigation, but the pattern does not itself prove a physical cause. Inadequate sampling can confound position with time or product changes.
Worked example: A team measures thickness at three positions on each of four consecutive sheets every two hours. Large position differences within each sheet suggest a different investigation from a common upward shift across all positions during later sampling periods.
Common mistake: Measuring only one position per unit cannot separate within-unit positional variation from differences between units or changes over time.
Official syllabus reference: [1]
33. Pearson Correlation and Confounding
Pearson's correlation coefficient measures the direction and strength of linear association between two quantitative variables. Values near positive or negative one indicate strong linear patterns, while values near zero indicate little linear association, not necessarily no relationship. Scatterplots reveal curvature, clusters, outliers, and restricted ranges that a single coefficient can hide. Correlation does not establish temporal order, eliminate confounding, or demonstrate causation. Pooled data can show a relationship different from the relationship within relevant groups. Statistical significance of correlation also depends on sample size and does not by itself establish practical usefulness.
Worked example: Across several factories, larger staffing levels correlate with more defects because larger factories produce more units. Within each factory, defects per unit may be unrelated to staffing. Stratifying and defining comparable exposure changes the interpretation.
Common mistake: A statistically significant correlation does not identify which variable causes the other or exclude a third variable affecting both.
Official syllabus reference: [1]
34. Regression Slopes, Residuals, and Prediction
Simple linear regression models the conditional mean response as an intercept plus slope times a predictor. The slope expresses expected response change per predictor unit, while residuals are observed minus fitted values. Residual plots help assess nonlinearity, unequal variance, unusual observations, and dependence. A confidence interval for a mean response differs from a prediction interval for an individual future observation; the latter includes additional outcome variation. Predictions within the observed range are interpolations, while extrapolations need stronger justification. A significant slope supports association under the model, and R-squared alone does not validate assumptions or causality.
Worked example: A model predicts processing time as 12 + 0.4 × batch size, with time in minutes. For a batch of 20, the fitted mean is 20 minutes. Individual batches still vary around that mean because residual variation remains.
Common mistake: Treating a fitted response as an exact future outcome ignores residual variation and confuses mean estimation with individual prediction.
Official syllabus reference: [1]
35. Alpha, Beta, Power, and Practical Significance
A Type I error rejects a true null hypothesis; its planned probability is controlled by significance level alpha under the test assumptions. A Type II error fails to reject when a specified alternative is true, with probability beta. Power equals one minus beta and depends on effect size, variability, sample size, and the decision rule. A p-value measures compatibility of the observed or more extreme result with the null model, not the probability the null is true. Practical significance compares effect magnitude with a meaningful operational requirement and should be evaluated alongside statistical uncertainty.
Worked example: A study can detect a 0.1-minute reduction with a very small p-value because it has thousands of observations. If implementation requires at least a one-minute saving to justify cost, statistical significance alone does not support adoption.
Common mistake: Failure to reject a null hypothesis does not prove equivalence; a low-powered study may simply miss an important difference.
Official syllabus reference: [1]
36. Paired t-Tests Analyze Differences
A paired t-test applies when observations form meaningful pairs, such as the same unit before and after a change or carefully matched units. Define one difference per pair and test the mean difference using t = mean difference divided by its standard error. The effective sample size is the number of pairs, with n − 1 degrees of freedom. Independence is required between pairs, and the usual small-sample normality assumption concerns the differences. Pairing can improve precision by removing stable between-unit variation. It is inappropriate to invent pairing merely because two datasets have equal lengths.
Worked example: Ten pumps are tested before and after adjustment. The ten differences have mean −2 liters per minute and standard deviation 1.5. The test statistic is −2/(1.5/√10), using nine degrees of freedom and preserving each pump's match.
Common mistake: Checking normality separately in the before and after datasets misses the distribution that directly supports the paired t-test: within-pair differences.
Official syllabus reference: [1]
37. Independent Mean Comparisons and Confidence Intervals
An independent two-sample mean comparison evaluates groups whose observations are not naturally paired. Welch's t approach estimates the standard error from each group's variance and sample size without assuming equal population variances. A pooled t approach requires an equal-variance assumption in addition to independence and suitable distribution conditions. The confidence interval for the mean difference communicates both direction and plausible magnitude. Group definitions, sampling, and potential confounding determine whether the comparison supports a causal interpretation. If observations are clustered by operator or machine, treating every reading as independent can understate uncertainty.
Worked example: Two independently sampled suppliers have mean strengths of 52 and 49 MPa. A confidence interval for the difference of 0.5 to 5.5 MPa supports a positive mean difference, but whether that difference matters depends on the engineering requirement.
Common mistake: Two independent samples with equal sizes do not become paired samples; matching requires a substantive relationship between particular observations.
Official syllabus reference: [1]
38. F-Tests for Variance Ratios
An F-test can compare the variances of two independent normal populations using the ratio of sample variances. The numerator and denominator each contribute their own degrees of freedom, n − 1. Tail selection depends on the alternative hypothesis and the chosen ratio convention. Variance testing addresses consistency rather than a difference in means. The classical F-test is sensitive to nonnormality and outliers, so distribution assessment is particularly important. A statistically detectable reduction in variance still requires interpretation in process units, such as standard deviation or expected conformance, before its practical value is clear.
Worked example: Independent normal samples have variances of 9 and 4 square millimeters, yielding an F ratio of 2.25 with the first variance in the numerator. The reference distribution uses each sample's separate degrees of freedom before significance is assessed.
Common mistake: A larger sample variance alone does not establish a population variance difference; sampling uncertainty and distribution assumptions still determine the inference.
Official syllabus reference: [1]
39. ANOVA Separates Between- and Within-Group Variation
One-way ANOVA tests equality of several population means by comparing between-group variation with within-group variation. Its F statistic is the between-group mean square divided by the within-group mean square. The conventional model assumes independent observations, approximately normal errors, and common error variance. A significant overall test indicates that at least one mean differs, without identifying every differing pair. Planned contrasts or suitable follow-up comparisons address specific differences while managing multiplicity. Effect sizes and confidence intervals help distinguish a detectable difference from an operationally meaningful one. Random assignment strengthens causal interpretation where feasible.
Worked example: Three curing methods have different sample means, but substantial variation also exists within each method. ANOVA evaluates whether the between-method differences are large relative to that background variation before the team investigates specific method comparisons.
Common mistake: Rejecting the overall ANOVA null does not establish that every pair of methods differs or identify the best method automatically.
Official syllabus reference: [1]
40. Proportions and Chi-Square Association
A proportion compares the count of units with a characteristic to the total eligible units. Independent-group proportion tests evaluate differences in those rates rather than differences in raw counts. A chi-square independence test compares observed contingency-table counts with expected counts calculated from row total times column total divided by grand total. Its usual degrees of freedom are (rows − 1)(columns − 1). Expected-count adequacy and independence matter for the approximation. A significant association does not establish causation or reveal every contributing cell; rates, residuals, and the sampling design guide interpretation.
Worked example: One shift has 12 defective units among 100 inspected, while another has 18 among 300. Comparing counts alone favors the first shift incorrectly; their defective proportions are 12% and 6%, and inference must account for both denominators.
Common mistake: Raw failure counts cannot establish which group performs worse when the groups have different numbers of inspected units or exposure.
Official syllabus reference: [1]
41. Gap Analysis with Comparable Current and Future States
Gap analysis compares current performance with a defined desired state using consistent metrics. The gap can be expressed as an absolute difference, relative change, or service shortfall, but the direction and denominator must be explicit. Segmenting the gap by product, location, or process stage identifies where improvement is needed. A target state should include consequential requirements rather than a single attractive number. Gap analysis locates and quantifies a problem; it does not establish its root cause. Apparent improvements can result from changed product mix or metric definitions, so comparisons require a common measurement basis.
Worked example: Current first-pass yield is 88% and the target is 94%, a six-percentage-point gap. That is a 6/88 ≈ 6.8% relative increase in yield, while the corresponding failure rate falls from 12% to 6%, a 50% reduction.
Common mistake: A six-percentage-point improvement is not the same as a six-percent relative improvement; the comparison denominator determines the latter.
Official syllabus reference: [1]
42. Fishbone, Five Whys, and Fault-Tree Logic
A cause-and-effect diagram organizes plausible causes into categories so investigation extends beyond the first explanation. Five Whys follows a causal chain toward actionable underlying conditions, but there is no requirement to stop after exactly five questions. Fault-tree analysis starts with an undesired event and uses logical gates to show combinations of contributing events. An OR gate represents alternative sufficient paths, while an AND gate requires the connected conditions together. Relational matrices can organize suspected links between causes and outcomes. All these tools generate or structure hypotheses; data, observation, and targeted tests establish which links are credible.
Worked example: A wrong label may result from either an incorrect file selection or a printer using an obsolete cached file, represented as alternative fault-tree paths. The team checks logs and reproduces each mechanism instead of treating the diagram as proof.
Common mistake: Stopping at 'operator error' usually leaves the enabling conditions unexplained and provides little guidance for a dependable process-level countermeasure.
Official syllabus reference: [1]
Improve Phase
43. DOE Factors, Levels, Treatments, and Responses
In designed experiments, factors are controlled input variables, levels are their selected settings, and responses are measured outcomes. A treatment is a particular combination of factor levels. The experimental unit is the smallest unit independently assigned a treatment, which determines the true replication structure. Effects compare response behavior across settings, while experimental error represents unexplained variation under the design. A factorial experiment varies several factors together, allowing interaction assessment that one-factor-at-a-time trials can miss. Levels should be feasible, safe, and sufficiently separated to answer the practical question without extending conclusions beyond the studied region.
Worked example: A coating experiment varies temperature at two settings and conveyor speed at two settings, producing four treatment combinations. Each independently processed panel receives one combination, and coating thickness is the response. Temperature and speed are inputs, not responses.
Common mistake: Calling every recorded variable a factor confuses deliberately varied inputs with measured outcomes and uncontrolled conditions that the design must address.
Official syllabus reference: [1]
44. Randomization, Blocking, Replication, and Repeated Readings
Randomization assigns treatments or run order to reduce systematic association with uncontrolled influences. Blocking groups experimental units by a known nuisance source, such as day or material batch, so treatment comparisons can separate that source. Replication repeats treatment application on independently treated experimental units and supports error estimation. Multiple readings on one treated unit improve measurement information but do not create independent treatment replication. Terminology for repetition can vary, so the physical assignment structure should always be stated. Poorly designed experiments can confound treatment with time, operator, or batch even when many measurements are collected.
Worked example: A team tests every low-temperature run on Monday and every high-temperature run on Tuesday. Temperature is confounded with day. Running both temperatures within each day, with randomized order and independent panels, provides a stronger blocked design.
Common mistake: Ten measurements from one treated panel are not ten independent experimental replicates, because treatment was applied only once.
Official syllabus reference: [1]
45. Main Effects and Interaction Plots
A main effect compares average response across a factor's levels, averaging over the other factors in the design. An interaction occurs when one factor's effect changes with another factor's level. On an interaction plot, nonparallel lines indicate differing conditional effects in the displayed means; uncertainty assessment still requires the experimental error information. Strong interaction can make an average main effect misleading or even zero despite substantial conditional effects. Factor settings should therefore be selected jointly when interaction matters. Crossing lines are a clear pattern, but interaction can also occur when lines differ in slope without crossing.
Worked example: Raising pressure improves yield from 80% to 90% at low temperature but lowers it from 95% to 85% at high temperature. The average pressure effect is zero, yet choosing pressure without considering temperature would miss important performance differences.
Common mistake: A near-zero main effect does not establish that a factor is unimportant when its effect changes across another factor's levels.
Official syllabus reference: [1]
46. Proof of Concept, Simulation, and Representative Pilots
A proof of concept establishes whether a proposed mechanism can work under limited conditions. Try-storming uses rapid, practical trials to learn about possible changes. Simulation explores system behavior through a model whose assumptions and inputs require validation. A pilot tests implementation on a controlled scale under conditions relevant to the intended rollout. Acceptance criteria should include the primary outcome, consequential outcomes, usability, and operational reliability. Representative conditions matter more than simply repeating a convenient setting many times. Rollout planning also needs ownership, training, resources, contingencies, and a way to confirm results after deployment.
Worked example: A simulated staffing rule reduces predicted waiting time, so a center pilots it during both busy and quiet periods. It measures waiting, abandoned calls, overtime, and adherence before deciding whether the modeled benefit transfers to actual operations.
Common mistake: A successful proof of concept establishes feasibility under its test conditions, not reliable performance across every condition expected during full deployment.
Official syllabus reference: [1]
47. Pull Replenishment and Kanban Sizing
A pull system authorizes production or replenishment in response to downstream consumption. Kanban provides a visible authorization signal and limits work in process when the rules are followed. A common sizing model divides expected demand during replenishment lead time, adjusted for an explicitly chosen allowance, by container quantity and rounds up to whole containers. Units and time bases must match. More cards generally permit more inventory, which can mask problems. Reliable replenishment, defined container quantities, and disciplined signal return are necessary; kanban is not simply a label placed on unrestricted production.
Worked example: Demand is 40 parts per hour, replenishment takes two hours, containers hold ten parts, and the scenario specifies a 25% allowance. Required cards are 40 × 2 × 1.25/10 = 10, assuming those inputs remain representative.
Common mistake: Adding kanban cards whenever a shortage occurs can increase inventory while leaving unstable replenishment time or unreliable signal handling unresolved.
Official syllabus reference: [1]
48. 5S and Standard Work
5S organizes work through sort, set in order, shine, standardize, and sustain. Its purpose includes easier abnormality detection, reduced searching, and consistent conditions, rather than appearance alone. Standard work specifies an agreed current method, including task sequence, expected timing, and necessary in-process inventory where applicable. It provides a stable baseline for training and improvement, while allowing controlled revision when better methods are demonstrated. Visual locations, cleaning-as-inspection, and routine checks support the standard. A rigid document that ignores actual work conditions will not create consistency simply because it is approved.
Worked example: A repair bench assigns each frequently used tool a visible location, removes obsolete items, and uses cleaning to expose leaks. A standard sequence then specifies inspection, disassembly, replacement, and verification so teams can compare improvements against a consistent baseline.
Common mistake: Treating 5S as a one-time cleanup misses the standardization and sustainment mechanisms that prevent searching, clutter, and hidden abnormalities from returning.
Official syllabus reference: [1]
49. Poka-Yoke Prevention and Detection
Poka-yoke makes an error impossible or makes it immediately evident before it becomes an escaped defect. Prevention approaches constrain selection, orientation, sequence, or completion; detection approaches signal or stop when an abnormal condition occurs. Contact methods check physical characteristics, fixed-value methods verify counts, and motion-step methods verify required sequences. The selected mechanism should address the demonstrated failure cause and be tested for false acceptance, false rejection, bypass, and recovery. A warning depends on a person's response, whereas an interlock can prevent continuation. Neither mechanism should create new hazards or obscure legitimate exceptions.
Worked example: A fixture accepts a housing only in the correct orientation, preventing reversed assembly. A separate sensor checks that all four fasteners are present before release. The team challenges both controls with deliberately incorrect conditions to verify their behavior.
Common mistake: An additional reminder is not equivalent to mistake prevention when the process still permits the same error without immediate detection.
Official syllabus reference: [1]
50. Continuous Flow and SMED Cycle-Time Reduction
Continuous flow reduces waiting and batch transfer by moving work through connected steps in small quantities where capacity and reliability permit. Lead time includes processing and waiting, so faster processing alone may leave the largest delay unchanged. SMED reduces changeover time by separating internal tasks that require stoppage from external tasks that can occur during operation, then converting and simplifying work. Smaller economic batches become more practical when changeovers shrink. Improvements must preserve necessary verification and safe execution. Balancing work, reducing queues, and preparing tools often deliver gains without increasing machine processing speed.
Worked example: A changeover includes 12 minutes of tool searching, eight minutes of fitting, and five minutes of verification. Staging tools externally removes the searching from downtime, reducing the stopped interval from 25 to 13 minutes without removing verification.
Common mistake: Reducing batch frequency by making larger batches can lower changeover counts while increasing waiting and inventory rather than shortening actual setup.
Official syllabus reference: [1]
51. Kaizen and Focused Kaizen Blitz Events
Kaizen is ongoing, incremental improvement through participation, observation, experimentation, and standardization. A kaizen blitz concentrates improvement work into a short, focused event with a bounded objective and prepared resources. The event format suits a problem where rapid cross-functional action and immediate trials are feasible; it does not replace deeper analysis when causes are uncertain or evidence takes time to collect. Both approaches require baseline measures and confirmation that changes improve the intended outcomes. Successful changes become the new standard, while unresolved actions retain owners and deadlines after the event ends.
Worked example: A daily kaizen routine gradually improves tool availability through small operator suggestions. A separate blitz brings maintenance, operators, and engineering together to redesign one changeover sequence, test it, update standard work, and assign remaining actions.
Common mistake: Finishing a scheduled blitz does not prove improvement; implemented changes still need outcome evidence, standardized methods, and ownership of unfinished actions.
Official syllabus reference: [1]
Control Phase
52. Common Causes, Special Causes, and Tampering
Common-cause variation arises from the current process system, while special-cause variation indicates an identifiable departure from that system's established behavior. Control charts compare time-ordered statistics with limits derived from a suitable baseline. A signal prompts investigation; an absence of signals supports, but does not prove, stability under the chosen rules. Specification limits concern requirements and are separate from control limits. Adjusting a stable process after every random fluctuation can increase variation, a problem called tampering. Improving common-cause variation generally requires changing the system, while special causes call for investigating the particular change.
Worked example: An operator repeatedly changes a filling setting whenever one bottle is slightly above or below target, although the chart is stable. The adjustments introduce additional oscillation. The team instead studies system-level sources of the stable filling variation.
Common mistake: A point inside customer specifications can still signal instability, while a statistically stable process can consistently produce output outside specifications.
Official syllabus reference: [1]
53. Rational Subgrouping Preserves Detectable Changes
Rational subgrouping collects observations so within-subgroup variation represents short-term behavior and meaningful changes appear between subgroups. Consecutive units under similar conditions often form useful subgroups, depending on the process question. Mixing different machines, shifts, or settings within each subgroup can inflate within-subgroup variation and hide differences in subgroup averages. Conversely, dependence among adjacent readings can distort conventional chart behavior. Sampling frequency must match the changes the team needs to detect. A subgroup is therefore a deliberate analytical choice, not merely a convenient number of measurements gathered during a day.
Worked example: Instead of averaging one reading from each of five machines into a subgroup, a team takes five nearby readings from each machine and charts machine behavior separately. This prevents machine-to-machine offsets from being absorbed into every subgroup's spread.
Common mistake: Combining unlike operating conditions within every subgroup can make control limits wider and conceal the shifts the chart should detect.
Official syllabus reference: [1]
54. X-Bar–R, X-Bar–s, and Median Charts
An X-bar chart monitors subgroup averages, paired with an R or s chart to monitor within-subgroup variation. R uses the range and is commonly convenient for small subgroups; s uses sample standard deviation and is generally more efficient as subgroup size grows. Chart constants depend on subgroup size and the chosen method. The variation chart should be examined because unstable spread undermines interpretation of mean-chart limits. A median chart monitors subgroup medians as an alternative location statistic. None of these charts should use specification width as a substitute for estimated process variation.
Worked example: Every hour, a team measures five consecutive parts and plots their mean and range using the appropriate five-part constants. A range signal appears before the mean shifts, prompting investigation of inconsistent clamping rather than an immediate centering adjustment.
Common mistake: Reviewing only subgroup means can miss a growing spread that increases defect risk even when the average remains near target.
Official syllabus reference: [1]
55. Individuals and Moving-Range Charts
An individuals chart monitors single observations when rational multi-observation subgroups are unavailable. A moving range commonly uses the absolute difference between successive observations to estimate short-term variation. For two-point moving ranges under the conventional assumptions, sigma is estimated as average moving range divided by 1.128, and individuals limits are the average plus or minus three estimated sigmas. Chronological order is essential. Strong autocorrelation, changing exposure, or highly nonnormal observations can undermine conventional interpretation. The moving-range chart adds information about local changes that an individuals chart alone may not make obvious.
Worked example: A furnace produces one comparable batch per day, so the team charts each day's average hardness as an individual observation. It calculates moving ranges between consecutive days, preserving sequence rather than sorting hardness values before estimating variation.
Common mistake: Sorting observations before calculating moving ranges destroys the time relationship and can substantially understate the variation used to set limits.
Official syllabus reference: [1]
56. Choosing p, np, c, and u Charts
A p chart monitors the proportion of defective units and can accommodate varying sample sizes with appropriate limits. An np chart monitors the number of defective units when sample size is constant. A c chart monitors defect counts for effectively constant exposure, while a u chart monitors defects per unit when exposure varies. The distinction between defects and defective units is fundamental: a unit is classified once for p or np, but may contribute several defects for c or u. Conventional limits rely on suitable binomial or Poisson assumptions, so clustering and overdispersion warrant investigation.
Worked example: A team inspecting 100 identical boards daily can chart defective-board counts with np. If it instead records all solder defects and inspects varying board quantities, a u chart can normalize defect counts by the stated inspection exposure.
Common mistake: Using a p chart for multiple defects per unit confuses a binary unit classification with a count that can exceed one.
Official syllabus reference: [1]
57. Control Plans, Reaction Rules, and Document Control
A control plan specifies the characteristic or input to control, measurement method, sampling arrangement, limits or requirements, responsible owner, records, and reaction when conditions fail. It connects a signal with action rather than merely listing metrics. Reaction rules should distinguish product conformance issues from process-instability signals and define containment and investigation responsibilities. Document control maintains approved revisions, access, change authorization, and removal of obsolete instructions from use. Control limits should be revised only with a justified process basis and appropriate stable data, not automatically whenever an inconvenient signal appears.
Worked example: A torque control plan defines the calibrated tool, sampling frequency, engineering limits, control-chart rules, and owner response. When the fastening method changes, the approved instruction is revised and obsolete copies are removed before operators use the new method.
Common mistake: A dashboard without defined ownership and reaction rules reports deterioration but does not provide a dependable mechanism for controlling it.
Official syllabus reference: [1]
58. Training Competence, Audit Types, and PDCA
A training plan identifies required skills, learners, instruction, practice, competence assessment, and refresher needs. Attendance records show participation, while demonstrated performance shows whether the method can be executed correctly. First-party audits are internal, second-party audits evaluate a supplier or other party of direct interest, and third-party audits are performed independently for purposes such as certification. PDCA links planning a change, doing a controlled implementation, checking results against expectations, and acting to standardize or revise. Audit evidence and competence results can feed the next improvement cycle rather than merely produce completed forms.
Worked example: After training, operators demonstrate the new setup using an observable checklist. An internal audit later samples actual records and work practices. PDCA uses discrepancies to revise the instruction or training, then checks whether the correction is effective.
Common mistake: Training attendance and completed audit checklists do not establish competence or process effectiveness unless evidence shows the required method works in practice.
Official syllabus reference: [1]
59. TPM, Predictive Maintenance, and OEE
Total productive maintenance seeks reliable equipment performance through shared ownership, operator care, planned maintenance, focused improvement, and prevention of recurring losses. Predictive maintenance uses condition information such as vibration or temperature trends to anticipate deterioration, whereas preventive maintenance uses planned time or usage intervals. Overall equipment effectiveness commonly multiplies availability, performance, and quality, provided each term is clearly defined. The combined measure helps classify losses but should not conceal its components or encourage excess production. Maintenance actions should address demonstrated failure mechanisms and preserve safe boundaries between operator tasks and specialist work.
Worked example: A machine has availability 0.90, performance 0.95, and quality 0.98, giving OEE 0.8379, or 83.79%. Vibration trends identify bearing deterioration, while the component scores show whether the larger improvement opportunity lies in downtime, speed, or rejects.
Common mistake: Scheduling more maintenance without identifying equipment losses can consume capacity while leaving the actual causes of breakdowns or quality variation unchanged.
Official syllabus reference: [1]
60. Visual Factory, Andon, and Jidoka
A visual factory makes normal conditions, abnormalities, priorities, and required responses apparent where work occurs. Andon communicates status or calls for assistance through visible or audible signals. Jidoka builds in the ability to detect an abnormal condition and stop or prevent continuation so defects do not flow onward. A signal without ownership, response expectations, or recovery criteria provides visibility but weak control. Visual devices should distinguish actionable exceptions from ordinary variation and remain understandable to the people using them. Effective visual control connects detection, response, correction, and authorized restart.
Worked example: A missing-component sensor stops assembly release and activates an Andon signal. The designated responder checks the condition, corrects the cause, and verifies restart criteria. A nearby display distinguishes waiting for help from normal operation so escalation is unambiguous.
Common mistake: Installing colored lights alone does not create Jidoka when abnormalities neither prevent continuation nor trigger a defined corrective response.
Official syllabus reference: [1]
Official syllabus sources
- Six Sigma Green Belt Certification Body of Knowledge Brochure
- Six Sigma Green Belt Certification | Apply & Get CSSGB Certified