I85PREP
⌘ K
Certified Pega Data Scientist study materials
Certified Pega Data Scientist study resources

Certified Pega Data Scientist practice questions

One question at a time. Build your Certified Pega Data Scientist understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Pega Platform

In Pega Platform, which component of the Prediction Studio workflow is responsible for producing the probability score that a predictive model assigns to a case?

  1. The data-decisions that consume the model output
  2. The predictor mapping defined in the model configuration
  3. The outcome definition configured for the model
  4. The machine-learning algorithm selected for the model Correct answer.

Explanation

The algorithm (for example a neural network or scorecard) is what actually learns from the historical data and produces the probability for the outcome. Predictors only supply the input fields, and outcome definitions only declare what is being predicted.

Question 2Pega Platform

A data scientist wants to confirm how well a binary propensity model separates positive from negative cases across all thresholds. Which Prediction Studio performance metric should be examined?

  1. The predictor importance ranking
  2. The AUC (area under the ROC curve) Correct answer.
  3. The model's response time
  4. The number of predictors used

Explanation

AUC measures the model's ability to rank positive cases above negative ones across every possible threshold, which is exactly the separability question. Predictor importance and predictor count describe inputs, and response time is a runtime concern.

Question 3Pega Platform

Which Pega Platform capability lets a data scientist run a predictive model against a dataset outside of case processing to inspect its scores?

  1. The model's simulation or test run in Prediction Studio Correct answer.
  2. The data page's load activity
  3. The decision data rule's test option
  4. The case type's process flow tracing

Explanation

Prediction Studio provides a way to run the model against a sample or uploaded dataset and review the resulting scores before deploying. Process flow tracing and data page loading are runtime diagnostics, not model evaluation tools.

Question 4Pega Platform

In a Pega adaptive model, what does the performance value reported after each update represent?

  1. The average latency of the scoring call
  2. The count of predictors currently active
  3. The percentage of predictions that matched the observed outcome Correct answer.
  4. The number of cases scored since deployment

Explanation

Adaptive model performance is the proportion of predictions that agreed with the actual outcome recorded for scored cases. It is a correctness ratio, not a timing or volume statistic.

Question 5Pega Platform

Which statement describes how predictor importance is used when refining a predictive model in Prediction Studio?

  1. It records the order in which predictors were added to the model
  2. It shows which predictors contribute most to the model's predictions so weak ones can be removed Correct answer.
  3. It guarantees the model will improve if the top predictor is removed
  4. It measures how often each predictor field is null in the sample

Explanation

Predictor importance ranks inputs by their contribution to predictions, helping a data scientist prune noisy or redundant fields. Removing a top predictor does not guarantee improvement, and null frequency is a separate data-quality check.

Question 6Pega Platform

A model predicts a binary outcome, and the data scientist wants to choose a threshold that balances false positives against false negatives. Which artifact supports that decision?

  1. The ROC curve with its trade-off points Correct answer.
  2. The predictor importance chart
  3. The number of outcome values configured
  4. The model's last update timestamp

Explanation

The ROC curve plots true-positive rate against false-positive rate at each threshold, so it directly supports choosing an operating point. Importance, timestamps, and outcome counts do not express that trade-off.

Question 7Pega Platform

When configuring a predictive model, what is the purpose of the outcome definition?

  1. It identifies the target field and its positive value that the model learns to predict Correct answer.
  2. It sets the schedule for model retraining
  3. It lists the input fields the model may use as predictors
  4. It defines the service level for scoring requests

Explanation

The outcome definition names the target property and which value counts as the positive outcome, giving the algorithm its learning objective. Predictors are configured separately, and retraining schedules and SLAs are unrelated settings.

Question 8Pega Platform

A data scientist notices that a model performs very well on the training sample but poorly on new cases in production. Which explanation is most consistent with this pattern?

  1. The adaptive model has not received enough negative outcomes
  2. The model's predictors were all removed before deployment
  3. The model has overfit the training data and does not generalize Correct answer.
  4. The production data page returns fields in a different order

Explanation

Strong training performance with weak production performance is the classic signature of overfitting, where the model captures noise specific to the sample. Predictor removal would degrade both, and field ordering does not affect scoring.

Question 9Pega Platform

In Pega Platform, which data preparation step most directly reduces the risk of a model learning from a field that leaks the outcome?

  1. Increasing the size of the training sample
  2. Shortening the model's retraining interval
  3. Excluding predictors that are populated only after the outcome is known Correct answer.
  4. Adding more predictors to the model

Explanation

Leakage occurs when a predictor is only available after the outcome is determined, so the model learns a shortcut. Larger samples, extra predictors, and faster retraining do not remove leaked information.

Question 10Pega Platform

What role does a data set play when creating a predictive model in Prediction Studio?

  1. It holds the decision logic that calls the model
  2. It stores the model's runtime scores for auditing
  3. It supplies the historical records used to train and test the model Correct answer.
  4. It defines the user interface for reviewing predictions

Explanation

A data set provides the historical population from which predictors and outcomes are drawn for training and evaluation. It is not a score store, a UI definition, or the calling logic.

Question 11Pega Platform

Which Pega Platform feature would a data scientist use to compare the predictive power of two candidate models before promoting one to production?

  1. The case type's life cycle configuration
  2. The model comparison view showing performance metrics side by side Correct answer.
  3. The decision table's otherwise condition
  4. The access group's role assignments

Explanation

Prediction Studio offers a comparison capability that places candidate models' metrics next to each other to support selection. Life cycle, decision table, and access settings are unrelated to model quality.

Question 12Pega Platform

A propensity model's output is consumed by a decision strategy. What must be configured so the strategy can reference the model's score?

  1. A predictive model component that maps the model to a score property Correct answer.
  2. A when condition that filters cases by age
  3. A data transform that renames the case ID
  4. A correspondence fragment that renders the score

Explanation

The strategy references the model through a predictive model component, which exposes the score as a property for downstream logic. Filtering conditions, renaming transforms, and correspondence do not surface model output.

Question 13Pega Platform

In an adaptive model, what happens to predictor weights as new case outcomes are recorded?

  1. They are frozen at the values set during initial training
  2. They are reset to zero after each scored case
  3. They are recalculated only when the model is redeployed
  4. They are updated incrementally to reflect recent outcomes Correct answer.

Explanation

Adaptive models learn continuously, adjusting predictor weights as outcomes arrive so the model tracks changing behavior. They are not frozen, zeroed, or limited to redeployment cycles.

Question 14Pega Platform

A data scientist observes that a model's performance metric in Prediction Studio is based on a very small number of scored cases. What is the most reasonable interpretation?

  1. The outcome definition is invalid
  2. The model is definitely performing well in production
  3. The predictors must all be replaced immediately
  4. The metric is unreliable and more scored cases are needed before drawing conclusions Correct answer.

Explanation

Performance estimates from very few scored cases carry wide uncertainty, so conclusions should wait for a larger sample. Small samples do not prove good performance or invalidate the outcome definition.

Question 15Pega Platform

Which consideration most strongly argues for using a scorecard model rather than a neural network in a regulated decisioning context?

  1. Scorecards require no training data
  2. Scorecards offer greater transparency into how each predictor affects the score Correct answer.
  3. Scorecards always achieve higher AUC than neural networks
  4. Scorecards cannot be monitored after deployment

Explanation

Scorecards express additive, inspectable contributions per predictor, which supports explainability requirements. They do not automatically outperform neural networks and still require training data and monitoring.

Question 16Pega Platform

What is the effect of including a predictor that is highly correlated with another predictor in a predictive model?

  1. It removes the need for an outcome definition
  2. It can add redundancy without improving predictive power Correct answer.
  3. It forces the model to become adaptive
  4. It guarantees a higher AUC

Explanation

Highly correlated predictors carry overlapping information, so adding one often contributes little and can complicate interpretation. It does not guarantee better performance or change model type requirements.

Question 17Pega Platform

Which Pega Platform artifact records the predictions and outcomes needed to evaluate an adaptive model over time?

  1. The case type's data transform
  2. The user portal's landing page
  3. The application's access group
  4. The adaptive model's performance tracking of scored cases and their outcomes Correct answer.

Explanation

Adaptive models track each scored case together with its eventual outcome, enabling ongoing performance measurement. Data transforms, access groups, and portals do not capture prediction-outcome pairs.

Question 18Pega Platform

A data scientist must decide whether a model's positive predictions should trigger a retention offer. Which factor should most influence the threshold choice?

  1. The number of predictors in the model
  2. The model's creation date
  3. The relative business cost of a false positive versus a false negative Correct answer.
  4. The color scheme of the review dashboard

Explanation

Threshold selection should reflect the asymmetric cost of wrong decisions, such as wasted offers versus lost customers. Predictor count, creation date, and presentation styling do not encode business cost.

Question 19Pega Platform

What distinguishes a predictive model from an adaptive model in Pega Platform?

  1. A predictive model only handles text, while an adaptive model only handles numbers
  2. A predictive model needs no data, while an adaptive model requires a data set
  3. A predictive model runs in batch, while an adaptive model runs only at design time
  4. A predictive model is trained on historical data, while an adaptive model updates continuously as outcomes arrive Correct answer.

Explanation

Predictive models learn from a prepared historical data set and are refreshed through retraining, whereas adaptive models adjust incrementally with each recorded outcome. Both can handle varied data types and both operate at runtime.

Question 20Pega Platform

Which symptom most suggests that a model's predictors include a field populated by the very process the model is meant to predict?

  1. Gradual performance drift over several months
  2. A steady but modest AUC around 0.65
  3. Near-perfect performance during training that collapses in production Correct answer.
  4. Frequent retraining requests from business users

Explanation

A predictor populated after the outcome is known creates leakage, producing unrealistically strong training results that fail in production. Gradual drift and modest AUC are normal patterns rather than leakage signals.

Question 21Pega Platform

In Prediction Studio, what does the model's positive outcome value signify?

  1. The target value the model treats as the event of interest when scoring Correct answer.
  2. The predictor with the highest importance
  3. The threshold above which cases are rejected
  4. The sample size used during training

Explanation

The positive outcome identifies which target value represents the event being predicted, such as 'churn' or 'accepts offer'. It is not a predictor ranking, a rejection threshold, or a sample size.

Question 22Pega Platform

A data scientist is preparing training data and finds a predictor that is null for most records. What is the most appropriate action?

  1. Keep it because null values always improve accuracy
  2. Convert the model to an adaptive model automatically
  3. Remove the outcome definition to compensate
  4. Evaluate whether the predictor should be excluded or imputed before training Correct answer.

Explanation

A mostly null predictor carries little reliable signal and may need exclusion or imputation, decided through data-quality review. Nulls do not improve accuracy, and neither removing the outcome nor switching model type addresses the issue.

Question 23Pega Platform

Which statement best describes how a predictive model score is typically used inside Pega decisioning?

  1. The score replaces the need for any decision logic
  2. The score is compared against thresholds or used in prioritization logic within a decision strategy Correct answer.
  3. The score is written directly into the case's audit trail as the final decision
  4. The score is used only for reporting after the case closes

Explanation

Model scores feed decision strategies, where they are compared to thresholds or combined with other factors to prioritize actions. They do not replace decision logic or serve merely as post-hoc reporting.

Question 24Pega Platform

What is the primary reason to monitor a deployed predictive model's performance over time?

  1. Monitoring converts the model into an adaptive model
  2. Monitoring eliminates the need for an outcome definition
  3. Data relationships shift, so a model that was accurate can degrade Correct answer.
  4. Monitoring automatically retrains the model each day

Explanation

Model performance decays as population behavior and data relationships change, so ongoing monitoring detects degradation and triggers action. Monitoring does not itself retrain, remove outcome definitions, or change model type.

Question 25Pega Platform

A team wants the model to prioritize cases most likely to accept an offer, but the historical data contains far more non-acceptors than acceptors. Which concern is most relevant?

  1. Class imbalance may bias the model toward the majority outcome Correct answer.
  2. The model will refuse to train without a scorecard algorithm
  3. The outcome definition must be deleted
  4. The predictors must all be numeric

Explanation

When one outcome dominates, models can favor the majority class and under-detect the minority event, so imbalance handling matters. Algorithm choice, outcome definition, and predictor data types are separate concerns.

Question 26Pega Platform

In Pega Platform, which component of a predictive model configuration specifies the exact business result the model is trained to estimate?

  1. The score threshold
  2. The outcome definition Correct answer.
  3. The sampling method
  4. The predictor set

Explanation

The outcome definition names the field and the positive value the model learns to estimate, so it anchors both training and scoring. Predictors are only the inputs, and thresholds and sampling shape behavior around that defined outcome.

Question 27Pega Platform

A data scientist wants a predictive model to learn from historical case data stored in Pega Platform. Which artifact supplies the labeled records used for training?

  1. A section rule
  2. A report definition
  3. A decision table
  4. A data set Correct answer.

Explanation

A data set defines and delivers the population of records, including predictors and outcome values, used to train and evaluate a model. Decision tables and sections are runtime UI or logic artifacts, and report definitions are for display rather than model training input.

Question 28Pega Platform

In Prediction Studio, what is the main purpose of reviewing a model's performance metrics before deploying it?

  1. To check that the model's decision strategy is already built
  2. To ensure the model was trained on the largest available data set
  3. To verify that every predictor field is populated in the database
  4. To confirm the model discriminates outcomes well enough to justify production use Correct answer.

Explanation

Performance metrics such as AUC or lift summarize how well the model separates positive from negative outcomes, which is the evidence for deployment readiness. Data completeness, strategy construction, and data volume are separate concerns that metrics do not certify.

Question 29Pega Platform

Which Pega Platform capability lets a data scientist compare several candidate models and promote the best performer to production?

  1. Data transform rules
  2. Case type inheritance
  3. Access control policies
  4. Prediction Studio model management Correct answer.

Explanation

Prediction Studio provides model management where candidate models are reviewed, compared, and promoted or replaced. Inheritance, access control, and data transforms are platform features unrelated to model lifecycle comparison.

Question 30Pega Platform

What does an adaptive model in Pega Platform do as new outcomes are recorded for scored cases?

  1. It freezes its weights after the first deployment
  2. It retrains from scratch each night using the full history
  3. It discards outcomes that contradict earlier predictions
  4. It updates its predictor weights continuously to reflect recent behavior Correct answer.

Explanation

Adaptive models learn incrementally from each recorded outcome, adjusting predictor weights so the model tracks changing patterns. They do not require full retraining, do not freeze after deployment, and do not filter inconvenient outcomes.

Question 31Pega Platform

Which Pega Platform artifact stores the predictions made and the outcomes later observed so an adaptive model's quality can be assessed over time?

  1. The data set's schema definition
  2. The case type's work history
  3. The adaptive model's learning records Correct answer.
  4. The decision strategy's test run log

Explanation

Adaptive models keep learning records pairing each prediction with the eventual outcome, which is the basis for evaluating ongoing performance. Work history, test logs, and schema definitions do not capture that prediction-outcome pairing.

Question 32Pega Platform

In Pega decisioning, how is a predictive model score typically consumed at runtime?

  1. As a trigger that restarts the case workflow
  2. As a hard block that prevents case creation
  3. As an input to a decision strategy that ranks or selects actions Correct answer.
  4. As a replacement for the case's service-level agreement

Explanation

Model scores flow into decision strategies, where they inform prioritization and selection of actions or offers. They do not block case creation, replace service levels, or restart workflows.

Question 33Pega Platform

A data scientist wants to know which predictors a Pega predictive model actually relies on. Where is that information reported?

  1. In the operator's access role
  2. In the case type's flow diagram
  3. In the application's ruleset stack
  4. In the model's predictor performance or importance view Correct answer.

Explanation

Prediction Studio reports predictor importance so the data scientist can see which fields drive the model's scores. Flows, roles, and ruleset stacks describe process, security, and code organization rather than model reliance.

Question 34Pega Platform

A data scientist must decide how to handle a predictor field that is missing for a large share of the training records. Which Pega Platform approach is most defensible?

  1. Duplicate the field so the model can learn from both copies
  2. Include it unchanged because models tolerate missing values automatically
  3. Replace all missing values with the positive outcome value
  4. Assess whether the field's coverage justifies inclusion and consider excluding or imputing it Correct answer.

Explanation

Low coverage weakens a predictor's reliability, so the data scientist should evaluate coverage and either exclude the field or apply a defensible imputation. Including it blindly, duplicating it, or filling gaps with the outcome value distorts learning.

Question 35Pega Platform

Two predictors in a Pega predictive model carry almost identical information. What is the most likely consequence for the model?

  1. Automatic removal of one predictor by Prediction Studio
  2. A guaranteed improvement in model discrimination
  3. Failure of the model to score new cases
  4. Unstable or redundant predictor contributions that complicate interpretation Correct answer.

Explanation

Highly correlated predictors split influence between them, producing unstable weights and redundant information without adding predictive power. The platform does not automatically drop one, discrimination does not reliably improve, and scoring still functions.

Question 36Pega Platform

A model's predictors include a field that is updated only after a retention offer is accepted. What is the most likely effect on the model?

  1. The model automatically ignores the field during training
  2. The model's score threshold adjusts to compensate
  3. The model learns from information unavailable at scoring time, inflating apparent performance Correct answer.
  4. The model becomes more accurate in production because the field is current

Explanation

A predictor populated by the outcome's own process is target leakage: it is unavailable when the prediction is needed, so offline performance looks strong but production performance collapses. The platform does not automatically detect or exclude such fields, and thresholds cannot compensate.

Question 37Pega Platform

A data scientist wants the model to rank customers by likelihood of accepting an offer, but only a small fraction of historical customers accepted. Which training consideration matters most?

  1. The size of the decision strategy that consumes the score
  2. The class imbalance between positive and negative outcomes Correct answer.
  3. The number of predictors in the model
  4. The naming convention of the outcome field

Explanation

When positives are rare, the model can favor the majority class and lose sensitivity to acceptors, so imbalance handling is the key concern. Predictor count, field naming, and strategy size do not address that skew.

Question 38Pega Platform

Why should a data scientist monitor a deployed predictive model's performance after go-live?

  1. Because predictor fields stop being populated once a model is deployed
  2. Because decision strategies ignore model scores after the first month
  3. Because Pega Platform deletes models that are not reviewed monthly
  4. Because relationships between predictors and outcomes can drift as business conditions change Correct answer.

Explanation

Model quality degrades when the population or behavior shifts, so ongoing monitoring detects drift and triggers retraining or replacement. There is no automatic monthly deletion, fields do not stop populating, and strategies do not stop reading scores.

Question 39Pega Platform

A decision strategy must use a predictive model's output to select an offer. What must exist for the strategy to reference that output?

  1. A separate report definition summarizing past scores
  2. A case-level attachment holding the model file
  3. A service-level agreement tied to the model
  4. A configured model or prediction component whose output property the strategy can read Correct answer.

Explanation

The model must be configured and connected to the decisioning logic so its output property is available for the strategy to evaluate. Reports, attachments, and service levels do not expose model output to strategy logic.

Question 40Pega Platform

A data scientist is selecting predictors for a churn model in Pega Platform and finds a field recorded after the customer already churned. What is the best action?

  1. Exclude the field because it is not available at the time the prediction is made Correct answer.
  2. Increase the model's sample size to offset the field
  3. Keep the field because it improves training accuracy
  4. Transform the field into a categorical predictor

Explanation

Predictors must be available at scoring time; a post-outcome field leaks the target and inflates training results without production value. Keeping, transforming, or offsetting the field does not fix its unavailability at prediction time.

Question 41Pega Platform

Which statement correctly distinguishes a predictive model from an adaptive model in Pega Platform?

  1. A predictive model is trained on historical data and updated through retraining, while an adaptive model updates continuously from outcomes Correct answer.
  2. A predictive model requires no outcome data, while an adaptive model requires it
  3. A predictive model updates continuously, while an adaptive model is trained once and frozen
  4. Both models update only when an operator manually edits the predictor list

Explanation

Predictive models learn from a historical sample and improve through periodic retraining, whereas adaptive models adjust weights incrementally as outcomes arrive. Neither model updates only through manual edits, and both rely on outcome data.

Question 42Pega Platform

A data scientist wants a model whose decisions can be explained to regulators case by case. Which model choice best supports that requirement?

  1. A model using hundreds of engineered interaction terms
  2. An ensemble of hundreds of decision trees
  3. A scorecard model with transparent, weighted predictors Correct answer.
  4. A deep neural network with many hidden layers

Explanation

Scorecards expose each predictor's contribution as an explicit weight, making individual decisions explainable. Neural networks, large ensembles, and heavy interaction engineering achieve accuracy at the cost of transparency.

Question 43Pega Platform

A data scientist observes strong training performance but weak results on new cases. Which Pega Platform action best addresses this?

  1. Remove the outcome definition to simplify the model
  2. Deploy the model immediately because training performance is high
  3. Increase the score threshold to one hundred percent
  4. Revisit predictor selection and regularization, then validate on a held-out sample Correct answer.

Explanation

The gap signals overfitting, so the data scientist should simplify or regularize the model and confirm results on data not used in training. Deploying, removing the outcome definition, or forcing the threshold does not address the underlying fit problem.

Question 44Pega Platform

In Pega Platform, which sampling consideration most affects the reliability of a model's reported performance?

  1. Whether the data set name follows naming standards
  2. Whether the case type has an approval workflow
  3. Whether the model uses more than five predictors
  4. Whether the evaluation records were held out from training Correct answer.

Explanation

Reported performance is trustworthy only when evaluated on records the model did not train on, which guards against optimistic estimates. Naming, predictor count, and workflow design do not determine evaluation validity.

Question 45Pega Platform

A model's positive outcome value is set to the value representing an accepted offer. What does the model's score represent?

  1. The estimated likelihood that a customer will accept the offer Correct answer.
  2. The estimated revenue from the offer
  3. The probability that the offer will be withdrawn
  4. The number of customers who received the offer

Explanation

The score estimates the probability of the defined positive outcome, here offer acceptance. It is not a revenue estimate, a count of recipients, or the chance of withdrawal.

Question 46Pega Platform

A data scientist must choose a probability threshold above which a retention offer is sent. Which factor should drive the choice?

  1. The size of the training data set
  2. The relative business cost of a wasted offer versus a missed retention Correct answer.
  3. The alphabetical order of the outcome values
  4. The number of predictors in the model

Explanation

Threshold selection is an economic trade-off: it balances the cost of false positives against the cost of false negatives. Predictor count, value ordering, and data volume are unrelated to that trade-off.

Question 47Pega Platform

A data scientist must confirm that a churn model's predictors would all be populated at the moment the score is requested in production. What is the most rigorous approach?

  1. Assume that any field on the case type is safe to use
  2. Trace each predictor to the point in the case lifecycle where it is captured and compare it with the scoring point Correct answer.
  3. Rely on the model's training accuracy as evidence of availability
  4. Test the model only against the historical sample

Explanation

Availability is a lifecycle question: each predictor must be captured before the score is needed, which requires tracing data capture points against the scoring point. Assumptions, training accuracy, and historical tests do not reveal timing gaps.

Question 48Pega Platform

An adaptive model in production begins to favor a predictor that previously had little influence. What is the most likely explanation?

  1. The predictor's data type was changed in the data set schema
  2. The decision strategy stopped sending outcomes to the model
  3. Recent outcomes show that predictor correlating more strongly with the positive outcome Correct answer.
  4. The model was retrained overnight on the full history

Explanation

Adaptive models shift weight toward predictors that recent outcomes show to be more predictive, so a rising influence reflects changing correlations. Schema changes, overnight retraining, and halted outcome feeds describe different mechanisms that would not produce this pattern.

Question 49Pega Platform

A team wants to deploy a model trained on two years of data, but a key predictor's meaning changed midway through that period because a field was repurposed. What is the best response?

  1. Increase the score threshold to offset the inconsistency
  2. Restrict training data to the period in which the predictor's meaning is consistent Correct answer.
  3. Keep all records because more history always improves the model
  4. Drop the outcome definition and rely on the remaining predictors

Explanation

A predictor whose meaning changed contributes inconsistent signal, so training should use only the period with stable semantics or the field should be reworked. More history, removing the outcome, or raising thresholds does not correct the inconsistency.

Question 50Pega Platform

A data scientist compares two candidate models in Prediction Studio: one has a higher AUC but was trained on a sample drawn only from a single region. What is the most important concern?

  1. The model with fewer predictors should always be preferred
  2. The sample may not represent the population the model will score in production Correct answer.
  3. AUC is irrelevant whenever a decision strategy is used
  4. The higher AUC guarantees the model will generalize everywhere

Explanation

A region-restricted sample can inflate apparent discrimination while failing to represent the deployment population, so representativeness matters more than the headline metric. High AUC does not guarantee generalization, predictor count is not decisive, and AUC remains relevant alongside strategy design.