Certified Pega Data Scientist study plan

Map the Pega Platform architecture layers (application stack, ruleset and ruleset stack, class hierarchy, data model) into a one-page diagram you can redraw from memory

Configure a Dev Studio application from scratch, defining the application record, organization, division, and two rulesets with versioning

Explain how inheritance and directed inheritance resolve a rule at runtime, and demonstrate it by placing the same rule name at two class levels

Set up role-based access with an access group, default ruleset, and portal so a test user lands in the intended landing page

Build a data model with embedded and linked data types, then verify how each behaves in a persisted vs. page-only context

Author and unit-test a data transform that copies, sets, and appends values across case and data pages

Create a decision table and a when rule, then trace which one fires using the Tracer during a test run

Apply guardrails correctly by resolving at least three warning-level findings in a sample application

Compare the debugging tools (Tracer, Clipboard, Live UI, AES-style log review) and pick the right one for four failure scenarios

Define Process AI in your own words, distinguishing it from classic decisioning and from rule-based automation, with two concrete use cases

Describe how predictive models are trained, scored, and monitored inside the platform, including what happens when model performance drifts

Explain the role of adaptive and self-learning models and when a static model is the better choice

List the data preparation steps a model needs (population, predictors, outcome labeling) and identify the two most common data quality failures

Draft an escalation path for a model that degrades in production, covering retraining, challenger models, and rollback

Configure a predictive model in Prediction Studio and score a small test population end to end

Interpret model output metrics (performance, AUC-style discrimination, score distribution) and explain what each tells a business stakeholder

Set up a simulation or comparison run to pit two candidate models against each other and justify the winner

Document the governance controls around model changes: who approves, what is versioned, and how outcomes are audited

Build a decisioning framework: proposition data, eligibility, applicability, and suitability rules wired into a single decision flow

Configure a next-best-action style arbitration with at least three actions and a ranking strategy, then inspect the output for each customer

Explain how adaptive analytics and interaction history feed back into arbitration and what cold-start looks like

Design a treatment or offer library with versioning so an expired offer cannot be selected

Trace a single decision from input customer data to final ranked action and annotate every rule that changed the outcome

Tune arbitration settings (weights, caps, suppression) and measure how the ranked output shifts when you change one lever at a time

Explain how decision results are captured for reporting and what you would check first if reported outcomes diverge from expected outcomes

Build a case type with at least three stages, two optional processes, and a decision-driven routing step

Configure a service level agreement with a goal and deadline, and verify escalation behavior fires in a timed test

Model a case lifecycle with subcases and dependency handling, then verify parent-child data propagation in both directions

Implement case-wide and stage-level validation plus a pre-processing data transform, and prove each fires at the intended moment

Configure assignment routing across work queues and skills, and test reassignment when no agent matches

Design a reporting view over case and decision data, and confirm the numbers reconcile with the case audit trail

Rehearse the four domains in exam proportion, spending time weighted to whichever domain scored weakest in your practice runs

Rewrite from memory the definitions that most often trip candidates: ruleset vs. ruleset stack, adaptive vs. predictive model, eligibility vs. suitability, stage vs. process

Complete a timed full-length practice set and review every wrong answer down to the underlying concept, not just the correct option

Build a one-page cheat sheet of formulas, thresholds, and configuration defaults you keep forgetting

Confirm logistics: exam booking, allowed reference material, and time allocation per domain during the test