7. Measurement Systems, Data Quality and Analytics Readiness

Introduction and Context

Lean and Six Sigma decisions are only as reliable as the measurements on which they are based. A process may appear unstable because an instrument is poorly calibrated, or it may appear capable because operators record defects inconsistently. In both situations, statistical analysis can create confidence without creating truth. Measurement is therefore not a clerical activity that follows process improvement; it is part of the process-improvement system itself.

In Lean, measurement supports the identification of waste, flow interruption, variation, delay, rework, and customer-value loss. In Six Sigma, measurement establishes the operational baseline, provides evidence for root-cause analysis, and enables verification of improvement. The Define–Measure–Analyze–Improve–Control (DMAIC) logic assumes that data are sufficiently valid, reliable, traceable, and timely to support decisions. If this assumption is not tested, later analyses may be technically correct but operationally misleading.

This chapter examines how organizations convert vague requirements into operational definitions, design data-collection plans, evaluate measurement systems, assess data quality, and prepare data for analytics. Particular attention is given to variable and attribute measurement systems, gauge repeatability and reproducibility (Gage R&R), attribute agreement analysis, traceability, uncertainty, and digital readiness. The central argument is that analytics readiness begins before data are collected: it begins with a clear definition of what is to be measured and why.

Learning Objectives

After studying this chapter, students should be able to:

1. Explain the role of measurement systems in Lean and Six Sigma improvement.

2. Translate customer, regulatory, and process requirements into operational definitions.

3. Construct a practical and statistically credible data-collection plan.

4. Distinguish accuracy, precision, repeatability, reproducibility, stability, resolution, and bias.

5. Apply the logic of variable Gage R&R and attribute agreement analysis.

6. Evaluate data quality using validity, completeness, consistency, timeliness, uniqueness, and traceability.

7. Identify risks associated with poor measurement and inappropriate analytics.

8. Design measurement systems that support digital dashboards, predictive analytics, and responsible decision-making.

Measurement as a System of Evidence

A measurement system is the complete set of people, instruments, procedures, software, environmental conditions, and standards used to assign values or classifications to a characteristic. It is broader than a gauge. For example, a hospital medication-error measure includes the definition of an error, the reporting process, the staff who identify events, the information system, rules for duplicate records, and the time period used for reporting.

A useful conceptual model is:

Observed value = true process value + measurement error.

Measurement error can be random or systematic. Random error causes readings to vary unpredictably. Systematic error, often called bias, shifts readings consistently away from the accepted or reference value. A system can therefore be precise but inaccurate, or accurate on average but highly variable. Both forms of error affect improvement decisions.

In process capability studies, excessive measurement variation inflates the apparent process variation and can make a capable process appear incapable. Conversely, a measurement system that fails to detect small but important differences can conceal real process problems. The practical question is not whether any measurement system is perfect, but whether its performance is adequate for the decision being made.

Important Measurement Concepts

Accuracy refers to closeness to an accepted reference value. Precision refers to the closeness of repeated readings to one another. Bias is the systematic difference between the average observed measurement and the reference value. Repeatability is variation when the same operator measures the same item repeatedly using the same equipment and method. Reproducibility is variation attributable to differences between operators, appraisers, laboratories, locations, or measurement conditions.

Stability describes whether the measurement system performs consistently over time. Resolution, or discrimination, is the smallest meaningful increment that the instrument or recording system can distinguish. Linearity concerns whether bias is consistent across the operating range. Sensitivity is the ability to detect meaningful changes in the characteristic. Uncertainty expresses the range associated with a measurement result within which the true value is expected to lie, subject to stated assumptions.

These concepts must be interpreted in relation to tolerance, specification, or decision requirements. A ruler that is adequate for measuring furniture may be inadequate for precision machining. Similarly, a customer-satisfaction scale may be adequate for identifying broad trends but inadequate for making individual employee-performance decisions.

From Requirements to Operational Definitions

Many measurement failures originate in ambiguous language. Terms such as “defect,” “on time,” “complete,” “clean,” “critical,” and “customer complaint” may have different meanings to different people. An operational definition converts an abstract requirement into a rule that two competent people can apply and obtain the same result.

A robust operational definition identifies:

• the characteristic or event being measured;
• the unit of measure and scale;
• the population, process step, or product scope;
• the inclusion and exclusion criteria;
• the observation point and time window;
• the decision rule or acceptance threshold;
• the responsible data collector; and
• the source of evidence or record.

For example, “late delivery” is weakly defined. A stronger definition might state: “A delivery is late when the shipment is not physically received at the customer’s designated location by 17:00 local time on the confirmed promise date, excluding delays formally approved by the customer before dispatch.” This definition clarifies the event, timing, location, and exclusions.

Operational definitions should also distinguish units. A defect is a nonconformity; a defective unit is a product or service unit containing one or more defects. If ten defects occur on three units, the defect count and defective-unit count are different measures. Confusing them produces incorrect rates and misleading Pareto analyses.

Definitions should be tested through a small pilot. Independent users apply the rule to borderline cases, discuss disagreement, and revise the wording. Borderline examples should be retained in a decision guide, especially for attribute inspection and service-quality classification.

Designing a Data-Collection Plan

A data-collection plan connects a management question to a reproducible method of obtaining evidence. It should be prepared before broad data collection begins. A practical plan usually includes the following fields:

| Planning element | Question addressed |

|---|---|

| Purpose | What decision will the data support? |

| Measure | What characteristic or event is being measured? |

| Operational definition | How will the measure be recognized and classified? |

| Data type | Is the data variable, attribute, count, ordinal, or categorical? |

| Source | Where will the data originate? |

| Sampling unit | What is one observation? |

| Population | To which process, product, customer, or period does it apply? |

| Sampling method | How will observations be selected? |

| Sample size and frequency | How much data are needed and when? |

| Collector and method | Who collects it and using what procedure? |

| Measurement system | Which instrument, form, or application is used? |

| Stratification | Which shifts, products, machines, locations, or suppliers matter? |

| Data controls | How are errors, missing values, and changes managed? |

The first question should always be, “What decision will this measurement support?” Data collected without a decision purpose often create large databases with little improvement value. The plan should also specify whether the purpose is estimation, comparison, monitoring, diagnosis, prediction, or compliance.

Sampling deserves particular care. A convenience sample may be easy to obtain but systematically exclude night-shift problems, difficult customers, high-mix products, or unusual operating conditions. Random, systematic, stratified, or rational subgroup sampling should be selected according to the process question. Time order must normally be preserved because mixing observations can conceal trends and shifts.

The plan should distinguish a measurement unit from a sampling unit. A production measurement may be taken from a part, while the sampling unit is a batch or time interval. In service operations, a transaction may be the measurement unit, while a customer journey may be the unit relevant to experience. Failure to define these units can lead to pseudoreplication, in which repeated observations are incorrectly treated as independent evidence.

Evaluating Variable Measurement Systems

Variable measurement systems produce numerical values such as length, temperature, cycle time, mass, or waiting time. Their evaluation commonly considers bias, repeatability, reproducibility, stability, and linearity.

Bias and Reference Standards

Bias is evaluated by comparing observed measurements with a known reference value. The reference must itself be credible and traceable. Several repeated readings are taken under defined conditions, and the average is compared with the reference. Bias should be interpreted against the tolerance or process decision threshold rather than against an arbitrary percentage.

A reference standard is not necessarily an absolute truth. It may be a calibrated artifact, a certified material, a master sample, or an agreed laboratory method. The organization should document the source, calibration status, uncertainty, and validity period of the reference.

Repeatability and Reproducibility

A Gage R&R study separates observed measurement variation into components associated with the measurement system and the parts being measured. In a crossed design, several operators measure the same set of parts more than once. The design should include parts that represent the actual process range, not only ideal or easily measured items.

The key components are:

• Equipment variation, primarily repeatability: variation when one operator repeats measurements.
• Appraiser variation, primarily reproducibility: variation between operators.
• Part-to-part variation: genuine differences among the items measured.
• Total variation: combined measurement-system and part variation.

A common summary is the percentage contribution of measurement variation to total variation, or the percentage of study variation. Rules of thumb often classify a system below approximately 10% as generally acceptable, between approximately 10% and 30% as potentially acceptable depending on risk and use, and above approximately 30% as requiring improvement. These are guidelines, not universal laws. A critical safety characteristic may require a much stricter standard, while a coarse screening process may tolerate more variation.

The number of distinct categories (ndc) indicates how many statistically distinguishable groups the measurement system can separate. A system may produce repeatable readings but still fail to discriminate among process levels. In capability analysis, a low ndc undermines the usefulness of conclusions about process variation.

Gage R&R results should never be interpreted mechanically. A high appraiser component may indicate inadequate training, unclear instructions, different measurement force, poor fixture design, or a genuinely subjective characteristic. High repeatability variation may result from worn equipment, environmental vibration, unstable parts, insufficient resolution, or inconsistent positioning.

Nested and Destructive Studies

A crossed study is inappropriate when the same part cannot be measured repeatedly, as in destructive testing or some biological assays. A nested design assigns different parts to operators and estimates variation through an appropriate experimental structure. Destructive systems require special attention to sample homogeneity, test order, material differences, and laboratory conditions. The design must ensure that apparent operator differences are not simply differences in the specimens assigned to operators.

Resolution, Stability, and Linearity

Instrument resolution should be small relative to the tolerance or decision interval. A digital display with many decimal places does not guarantee meaningful resolution; the entire system, including fixture and method, must support that discrimination.

Stability is examined using control charts or periodic checks with a standard. Drift may be caused by temperature, aging components, software changes, calibration loss, or maintenance conditions. Linearity studies evaluate bias at multiple points across the operating range. A gauge that is accurate at the centre of its range may be biased at the lower or upper limit.

Attribute Agreement and Classification Systems

Attribute systems classify observations into categories such as pass/fail, conforming/nonconforming, defect type, severity level, or complaint category. They are common in visual inspection, auditing, medical diagnosis, customer-contact coding, and document review.

Attribute agreement analysis evaluates agreement within an appraiser, between appraisers, and against a known standard when one exists. Percent agreement is intuitive but can be misleading when one category is very common. Chance-corrected statistics such as kappa may provide additional insight, although kappa itself is affected by prevalence and category distribution. The choice of statistic should be explained rather than treated as a universal pass/fail rule.

A useful study includes representative samples: clear conformances, clear nonconformances, and borderline cases. Each appraiser should assess samples independently and, where possible, without seeing earlier decisions. Training before the study is appropriate, but coaching during the assessment can hide the true performance of the system.

Low agreement may arise from ambiguous standards, poor lighting, inadequate images, fatigue, insufficient training, overlapping categories, or an intrinsically subjective characteristic. Improvement may require a defect catalogue, reference samples, better inspection conditions, simplified categories, automated image assistance, or escalation rules for uncertain cases.

Data Quality and Analytics Readiness

Data quality is multidimensional. A dataset is not “good” merely because it contains many records or passes a software import. Important dimensions include:

• Validity: values conform to defined formats, ranges, and business rules.
• Accuracy: values represent the intended real-world condition.
• Completeness: required records and fields are present.
• Consistency: the same concept is represented similarly across systems and time.
• Timeliness: data are available at a useful point in the decision cycle.
• Uniqueness: duplicate records are controlled.
• Integrity: relationships among records remain correct.
• Traceability: the origin, transformation, and ownership of data can be reconstructed.
• Accessibility: authorized users can obtain and interpret the data.

Analytics readiness means more than storing data in a database. It requires clear metadata, stable identifiers, defined units, controlled versions, documented transformations, and known limitations. A dashboard may display a precise-looking defect rate while mixing production dates with shipment dates, excluding cancelled orders inconsistently, or counting rework as new production. Visual polish cannot compensate for weak data definitions.

A data dictionary should identify field names, definitions, data types, units, allowed values, source systems, owners, update frequency, and rules for missing values. Codes such as “0,” “N/A,” blank, and “unknown” must not be treated as interchangeable. Missingness may itself contain information; for example, a field may be missing because a process step was skipped.

Data lineage records where data originate, how they are transformed, and where they are consumed. Version control is important when definitions, algorithms, forms, or software change. When a KPI changes, users should be able to determine whether the process changed or merely the measurement rule.

A Practical Readiness Workflow

An organization can evaluate measurement and data readiness through the following sequence:

1. Define the decision, customer requirement, or risk being addressed.

2. Identify the characteristic, event, population, and unit of analysis.

3. Write and test the operational definition.

4. Map the data-generating process, including handoffs and opportunities for error.

5. Select the measurement method and reference standard.

6. Pilot the collection form, instrument, or digital workflow.

7. Assess variable or attribute measurement performance.

8. Establish sampling, storage, validation, and escalation rules.

9. Create metadata, lineage, ownership, and access controls.

10. Analyze baseline data while preserving time order and relevant stratification.

11. Review results with process owners and measurement experts.

12. Monitor the measurement system after implementation.

This workflow is iterative. Discovery during a measurement study may reveal that the original operational definition is not useful or that the process contains multiple populations. Revising the definition is not failure; it is a normal part of building valid evidence.

Applied Case: Measurement Failure in a Precision Assembly Process

Background

The following is an illustrative case. A manufacturer of precision control modules experienced an increase in final-test failures. The production team believed that a fastening operation was producing excessive torque variation. A project team proposed adjusting the fastening tool and replacing several components.

Situation and Problem

The recorded torque values showed a standard deviation that appeared large relative to specification. However, operators used two different electronic torque testers, and the work instruction did not specify the angle or seating condition for the measurement. Three shifts also used different methods for positioning the assembly. Management needed to determine whether the process or the measurement system was responsible for the apparent variation.

Analysis

The team first wrote an operational definition: torque would be measured at a specified joint, after a defined seating sequence, using a named fixture and a calibrated tester. They selected ten assemblies spanning the observed operating range. Three operators measured each assembly twice on the same instrument under controlled conditions.

The study indicated substantial repeatability variation and a smaller but meaningful operator component. The instrument resolution was adequate, but the fixture allowed slight angular movement. A second review found that the final-test data combined first-pass results with measurements taken after rework. Consequently, the original defect rate did not represent first-pass process performance.

Approach and Solution

The organization introduced a rigid fixture, standardized the seating procedure, separated first-pass and post-rework records, and added an automated range check. Operators were retrained using reference assemblies that represented low, nominal, and high torque. The data system recorded instrument identification, operator, timestamp, product revision, and rework status. A short-term stability check was added at the start of each shift.

A repeat measurement study was conducted after implementation. Measurement variation was reduced sufficiently for the process team to distinguish meaningful part-to-part differences. The team then repeated the process capability analysis using only valid first-pass records and preserved the time order of observations.

Outcome

The apparent process problem was found to be partly a measurement problem and partly a data-definition problem. Tool adjustment was still required, but it was targeted using more trustworthy evidence. The revised dashboard showed separate indicators for first-pass yield, rework yield, and measurement-system status. Supervisors could now determine whether a change in results reflected production performance, inspection conditions, or data capture.

Learning Points

The case demonstrates four principles. First, measurement-system analysis should precede strong conclusions about process capability. Second, operational definitions must identify the status and context of an observation. Third, traceability makes it possible to investigate changes rather than merely report them. Fourth, analytics readiness is achieved through disciplined process design, not by adding a dashboard after data collection.

Technology, Automation, and Artificial Intelligence

Digital instruments and manufacturing execution systems can improve timeliness, reduce transcription errors, and attach context automatically. Barcode or RFID identification, sensor capture, electronic signatures, and automated calibration reminders strengthen traceability. Statistical process-control software can issue alerts when a measurement shows unusual behaviour, but the alert is meaningful only if the measurement system is stable.

Computer vision can support attribute inspection, and machine-learning models can identify patterns in images, sound, or process signals. These technologies introduce new measurement questions: What is the labelled reference standard? How representative are training data? How are false positives and false negatives balanced? Does performance change across products, lighting conditions, suppliers, or demographic groups? Model accuracy should be evaluated against the operational decision and risk, not only against an aggregate test score.

Automation can also amplify poor definitions. If a vague defect category is coded at high speed, the organization obtains more inconsistent data, not better data. Digital systems should therefore include validation rules, exception handling, audit trails, controlled master data, and human review for uncertain cases.

Governance, Ethics, and Professional Responsibility

Measurement systems influence resource allocation, employee evaluation, supplier approval, customer treatment, and regulatory reporting. Professionals must not manipulate definitions, sampling windows, exclusions, or missing values to create favourable results. Changes in KPI definitions should be documented and communicated.

Privacy and confidentiality are relevant when data involve employees, patients, customers, or identifiable transactions. Access should follow legitimate business need, and retention should comply with applicable law and organizational policy. Automated classification should be monitored for systematic error, especially when decisions affect individuals. A measurement system should make uncertainty visible rather than presenting estimates as unquestionable facts.

Governance responsibilities should be assigned. Process owners define business meaning; measurement specialists evaluate method performance; data stewards maintain definitions and lineage; information-technology teams protect systems; and managers ensure that decisions do not exceed the evidence available.

Common Mistakes and Practical Safeguards

A frequent mistake is beginning with available data rather than with the decision. Another is treating calibration as proof of total measurement-system adequacy. Calibration may establish instrument conformity, but it does not guarantee correct method, operator consistency, fixture control, or data integration.

Other common errors include using too narrow a part range in Gage R&R, hiding borderline samples in attribute studies, ignoring time order, mixing different populations, counting duplicate events, treating missing values as zero, and changing definitions without historical reconciliation. Organizations also sometimes use capability indices before verifying distributional assumptions, stability, and measurement adequacy.

Practical safeguards include piloting definitions, using representative samples, maintaining reference standards, documenting exceptions, separating first-pass from rework data, reviewing control charts for measurement stability, conducting periodic agreement studies, and linking every important metric to an owner and decision rule.

Current Developments

Measurement systems are increasingly distributed across connected equipment, cloud applications, laboratories, suppliers, and customer channels. This creates opportunities for real-time monitoring but increases the importance of interoperability, metadata, cybersecurity, and change control. Edge analytics can detect abnormal conditions near the source, while centralized platforms support cross-site analysis.

Digital twins and advanced process models depend on synchronized and traceable data. Synthetic data may assist model development, but it cannot replace representative real-world validation. Organizations are also moving toward risk-based measurement: critical characteristics receive stricter controls, more frequent verification, and stronger traceability than low-risk characteristics.

The emerging professional role is therefore broader than statistical analysis. Lean Six Sigma practitioners must understand measurement science, information governance, process behaviour, and the social consequences of metrics. Analytics maturity depends on the integration of these capabilities.

Best-Practice Principles

1. Begin with the decision and customer or regulatory requirement.

2. Define the measure so independent users can apply it consistently.

3. Evaluate the complete measurement system, not only the instrument.

4. Use representative parts, cases, operators, shifts, and conditions.

5. Separate measurement variation from genuine process variation.

6. Preserve time order, population boundaries, and relevant stratification.

7. Document units, codes, missing values, transformations, and ownership.

8. Treat traceability and version control as design requirements.

9. Match measurement-system risk to the consequence of a wrong decision.

10. Reassess the system after changes in product, process, software, personnel, or environment.

Chapter Summary

Measurement systems provide the evidence base for Lean and Six Sigma decisions. Their performance depends on instruments, methods, people, environment, software, and definitions. Operational definitions remove ambiguity by specifying what is measured, how it is classified, and under which conditions. Data-collection plans connect measures to decisions and protect against biased sampling.

Variable systems are evaluated through bias, repeatability, reproducibility, stability, linearity, resolution, and uncertainty. Gage R&R separates measurement-system variation from part-to-part variation. Attribute agreement studies assess consistency in classification and are especially important when inspection involves judgement. Data quality includes validity, accuracy, completeness, consistency, timeliness, uniqueness, integrity, and traceability.

Analytics readiness requires a data dictionary, controlled definitions, lineage, ownership, validation, and appropriate access. Digital tools and AI can strengthen measurement, but they cannot correct ambiguous definitions or biased data automatically. Ethical governance requires transparency about uncertainty, responsible use of personal information, and honest reporting of changes. The most reliable improvement systems treat measurement as a managed technical and organizational process.

Key Terms

Accuracy: Closeness of a measurement to an accepted reference value.

Attribute data: Data that classify observations into categories or counts rather than continuous numerical values.

Bias: Systematic difference between the average observed value and a reference value.

Data lineage: A record of data origin, movement, transformation, and use.

Data quality: The degree to which data are fit for their intended purpose.

Gage R&R: A study that estimates repeatability and reproducibility of a measurement system.

Linearity: Consistency of measurement bias across the operating range.

Measurement error: Difference between an observed value and the underlying true or accepted value.

Measurement system: People, equipment, methods, environment, software, and standards used to obtain measurements.

Operational definition: A precise rule specifying how a characteristic or event is recognized and recorded.

Precision: Closeness of repeated measurements to one another.

Repeatability: Variation under the same operator, equipment, method, and conditions.

Reproducibility: Variation between operators, appraisers, locations, or conditions.

Resolution: Smallest meaningful increment distinguishable by a measurement system.

Stability: Consistency of measurement-system performance over time.

Traceability: Ability to reconstruct the origin, history, and transformations of a measurement or data record.

Variable data: Numerical measurements on a continuous or discrete scale.

Uncertainty: Quantified doubt or interval associated with a measurement result.

Review Questions

Short Answer Questions

1. Why must a measurement system be evaluated before process capability is interpreted?

2. Distinguish repeatability from reproducibility.

3. What information should an operational definition contain?

4. Why can percent agreement be insufficient in an attribute study?

5. Define data lineage and explain its value in analytics.

Long Answer Questions

1. Explain the components of a variable measurement system and describe how a Gage R&R study evaluates them.

2. Discuss the design of a data-collection plan for a Six Sigma project, including sampling and stratification.

3. Explain how validity, accuracy, completeness, consistency, timeliness, uniqueness, and traceability contribute to data quality.

4. Compare variable measurement-system analysis with attribute agreement analysis.

5. Discuss how digital technologies and AI can improve or weaken measurement-system performance.

Analytical / Application Questions

1. A service organization reports that “on-time resolution” improved from 82% to 94%, but it changed the definition from calendar days to business hours and excluded reopened cases. Analyze the measurement and governance problems and propose a corrective plan.

2. Three inspectors disagree frequently when classifying surface defects, although each inspector agrees with their own earlier decisions. Diagnose possible causes and design an attribute agreement study and improvement strategy.

3. A factory’s capability index declines after a new automated gauge is installed. Develop a structured investigation that separates actual process deterioration from bias, resolution, stability, repeatability, reproducibility, and data-integration problems.

22. Multiple Choice Questions with Answers

Question 1

Which statement best describes an operational definition?

A. A general statement of management intent

B. A precise rule for recognizing and recording a characteristic or event

C. A software specification for storing data

D. A statistical estimate of process capability

Correct Answer: B. A precise rule for recognizing and recording a characteristic or event

Question 2

In a Gage R&R study, variation caused when one operator repeatedly measures the same part is primarily called:

A. Reproducibility

B. Bias

C. Repeatability

D. Linearity

Correct Answer: C. Repeatability

Question 3

A measurement system gives readings that are very close to one another but consistently higher than the reference value. It is best described as:

A. Precise but biased

B. Accurate but unstable

C. Unrepeatable but unbiased

D. Low in resolution but linear

Correct Answer: A. Precise but biased

Question 4

Which sampling practice is most likely to conceal a shift-related process problem?

A. Stratified sampling by shift

B. Random sampling across all operating periods

C. Convenience sampling during the day shift only

D. Systematic sampling with time order preserved

Correct Answer: C. Convenience sampling during the day shift only

Question 5

In attribute agreement analysis, representative samples should include:

A. Only obvious conforming units

B. Only previously rejected units

C. Clear cases and borderline cases

D. Samples selected by the most experienced appraiser

Correct Answer: C. Clear cases and borderline cases

Question 6

Which data-quality dimension concerns whether required records and fields are present?

A. Completeness

B. Uniqueness

C. Timeliness

D. Linearity

Correct Answer: A. Completeness

Question 7

A process team combines first-pass failures with results after rework in one defect-rate measure. The primary problem is:

A. Insufficient instrument resolution only

B. Failure to define and separate different process populations

C. Excessive reproducibility

D. Use of a reference standard

Correct Answer: B. Failure to define and separate different process populations

Question 8

A gauge is calibrated but operators use different fixtures and positioning methods. Which conclusion is most appropriate?

A. Calibration proves the entire measurement system is adequate

B. The gauge must be biased

C. Method and operator effects still require evaluation

D. The process cannot be measured

Correct Answer: C. Method and operator effects still require evaluation

Question 9

Which feature most directly supports analytics readiness?

A. A visually attractive dashboard without metadata

B. A large volume of unvalidated records

C. Documented definitions, lineage, units, ownership, and transformations

D. Removing all missing values before analysis

Correct Answer: C. Documented definitions, lineage, units, ownership, and transformations

Question 10

When an AI inspection model is introduced, which question is most important for measurement validity?

A. Whether the interface uses modern colours

B. Whether training and validation data represent relevant products and conditions

C. Whether the model eliminates all human review

D. Whether the database contains the maximum number of fields

Correct Answer: B. Whether training and validation data represent relevant products and conditions