Domain IV: Manage AI Model Development and Evaluation — Study Game

How to Play

Pick a game mode and test yourself. Cover answers and try to recall before peeking.


The Two Gates (IV.5 + IV.6)

Card 1 — Front: What's the IV.5 gate question?

Answer: Is the prepared data quality sufficient to train on? End of Phase III.

Card 2 — Front: What's the IV.6 gate question?

Answer: Is the model ready to operate in production? End of Phase V.

Card 3 — Front: Distinguish III.8 from IV.5.

Answer: III.8 = end of Phase II ("do we have what we need?"). IV.5 = end of Phase III ("is prepared data sufficient to train?"). Different artifacts, different boundaries.

Card 3b — Front: When are you ready to move past Phase III (the IV.5 gate)?

Answer: All data preparation issues addressed and the plan executed — Phase III is about actually doing the prep. Workbook steps complete with a well-defined answer, set of steps, and execution. No roadblocks. "You are not ready to build your model until the data is exactly how you need it, in the right shape, and in the right format."

Card 3c — Front: What does PMI ask to decide model readiness at the IV.6 gate?

Answer: Have you developed a model that aligns with the goals and objectives you set out to achieve? Does it perform well on the data that was carefully selected and prepared? Have you completed all the necessary steps of model evaluation — testing, validation, metrics analysis? And do you have adequate responses to the Phase V workbook questions?

Card 4 — Front: Outcomes at the IV.5 and IV.6 gates?

Answer: GO or NO-GO. NO-GO means pause and iterate back to the phase that owns the root cause — for IV.6, PMI names Phase I, II, III, or IV. Do not progress with roadblocks or warning signs.

Card 5 — Front: What does IV.6 GO authorize?

Answer: Domain V (deployment) work to begin.

Technique Selection (IV.1)

Card 6 — Front: What's the PM's role in IV.1?

Answer: Oversee — ensure technique is documented, justified against AI pattern + success criteria, aligned with operational constraints. PM doesn't pick the technique.

Card 7 — Front: Three ML categories?

Answer: Supervised, Unsupervised, Reinforcement.

Card 8 — Front: Difference between algorithm and model?

Answer: Algorithm = procedure. Model = trained artifact. You train an algorithm to produce a model.

Card 9 — Front: Three patterns of pretrained AI?

Answer: Pretrained model (adapt for task), Foundation model (very large pretrained), GenAI (generates new content).

Card 10 — Front: What's transfer learning?

Answer: Pretrained + fine-tune on your task data.

Card 11 — Front: What's RAG?

Answer: Retrieval-Augmented Generation — retrieve relevant context + generate from foundation model.

Training (IV.3)

Card 11b — Front: Four root-cause areas to review when training overruns?

Answer: Data, Technique, Hardware, Results — review when training overruns.

Card 12 — Front: Overfit vs underfit?

Answer: Overfit = memorizes training data, fails on new. Underfit = doesn't learn even on training.

Card 13 — Front: Typical train/validation/test split?

Answer: ~70%/15%/15%.

Card 14 — Front: What does generalization mean?

Answer: Model performs well on data it hasn't seen — the goal of training.

Data Preparation (IV.4)

Card 14b — Front: Four core data preparation activities (IV.4)?

Answer: Transform formats, Reconcile inconsistencies, Impute missing values, Map fields.

Card 15 — Front: How much of an AI project is data engineering?

Answer: About 80% — data identification and data aggregation. Over 80% of project time goes to data wrangling. Model training, tuning, and implementation are the small remainder.

QA/QC (IV.2)

Card 16 — Front: What does IV.2 QA/QC cover?

Answer: Configuration management + performance verification + bias measurement + documentation throughout development.

Card 17 — Front: Three transparency dimensions?

Answer: Systemic (how built), Decision (why this prediction), Algorithmic (algorithm-level).

Card 18 — Front: XAI vs Interpretability?

Answer: XAI = post-hoc explain any model. Interpretability = inherently understandable models. High-stakes prefers interpretability.

GAME MODE 2: Scenario Showdown — What Should the PM Do?

Scenario 1: The Training Overrun

Reveal

Pause training. Conduct (Data/Technique/Hardware/Results) root-cause review. Document decision: continue, change approach, or escalate. 2.5x overrun = project event, not technical hiccup.

Scenario 2: The Black-Box Healthcare Decision

Reveal

Document technique selection; ensure trade-off between performance and explainability is presented to stakeholders for decision; consider interpretable-by-design alternatives. IV.1 + Domain I.2 cross-pull.

Scenario 3: The Operational Mismatch

Reveal

NO-GO — a model that can't run in its destination isn't ready to operationalize. Inference efficiency and operational efficiency are part of Phase V evaluation. Loop back to V.1 (infrastructure) or IV.1 (technique change) with stakeholder decision.

Scenario 4: The Bias Discovery During QA

Reveal

Treat as IV.2 + I.3 issue: document, escalate per accountability, engage stakeholders for remediation, do not authorize IV.6 GO until bias within tolerance.

Scenario 5: The Parallel Work Request

Reveal

Confirm IV.6 must complete before Domain V work begins — sequential, not parallel. Gate authorizes the transition.

Scenario 6: The Plateau Concern

Reveal

Coordinate investigation of early plateau (data quality, technique fit, hyperparameter tuning). Engage IV.2 QA/QC. Don't blindly accept "acceptable" without root-cause.

Scenario 7: The Iteration Trap

Reveal

Pause and review iteration trajectory: are improvements converging or plateauing? Is the technique a fit? Is data sufficient? Document decision: continue, change technique, rescope, or escalate.

Scenario 8: The Quality Gate with Bias

Reveal

NO-GO — an unresolved bias finding is a roadblock, and the data isn't "exactly how you need it." Loop back to IV.4 to remediate or III.1 to redefine. Trustworthy AI (I.3) cross-pull.


GAME MODE 3: Pattern Match Challenge

#ScenarioECO Task
1Overseeing model technique selectionIV.1
2Overseeing model QA/QCIV.2
3Managing model training executionIV.3
4Managing data transformationIV.4
5Verifying data quality (gate)IV.5
6Verifying model ready for ops (gate)IV.6
7Confirming the model can run where it's going before authorizing deploymentIV.6
8Confirming the prepared data is in the right shape and format before trainingIV.5
9Root-cause triage when training overruns the planned windowIV.3
10Coordinating the categories of data prep work (transform, reconcile, impute, map)IV.4

Scoring: 9-10 = Expert | 7-8 = Solid | Below 7 = Review


GAME MODE 4: Fill-in-the-Blank Speed Round

  1. At IV.5 you are not ready to build your model until the data is exactly how you need it, in the right ________ and the right ________.
  2. At IV.6, PMI asks whether the model aligns with the ________ and objectives you set out to achieve.
  3. III.8 = "do we have what we need?" IV.5 = "is the ________ data sufficient to train?"
  4. AutoML automates the technical decision but doesn't replace ________ documentation (IV.1).
  5. The PM doesn't pick the technique — the ________ does.
  6. Training overrun by 2.5x = project event. Apply review: Data, Technique, ________, Results.
  7. About ________% of an AI project is data engineering work.
  8. IV.6 GO authorizes ________ ________ work to begin.
  9. RAG = ________-Augmented Generation.
  10. Reproducibility means same data + same pipeline = ________ output.

Reveal answers
  1. shape / format
  2. goals
  3. prepared
  4. governance
  5. data scientist
  6. Hardware
  7. 80
  8. Domain V
  9. Retrieval
  10. same


GAME MODE 5: True or False Lightning Round

#StatementCorrect
1III.8 and IV.5 are the same gateFALSE — different gates at adjacent boundaries
2The PM picks the model techniqueFALSE — data scientist picks; PM oversees governance
3At IV.6 you can proceed as long as the model meets its requirements, even if key Phase V workbook questions are still unansweredFALSE — PMI requires adequate workbook responses too; do not progress with no answers, roadblocks, or warning signs
4Domain V work can begin in parallel with IV.6 gateFALSE — sequential
5AutoML bypasses IV.1 documentation requirementFALSE — automation ≠ governance
6Whether the model can actually run in its destination is part of Phase V evaluationTRUE — model size, inference efficiency, operational efficiency, and edge-device demands are all evaluated
7Production validation substitutes for IV.6 evaluationFALSE — gate is pre-deployment
8Reproducibility means inference reproducibility onlyFALSE — training reproducibility too
9The PM declares IV.6 GO unilaterallyFALSE — multi-stakeholder sign-off
10A failed contingency test still satisfies V.7 if documentedFALSE — V.7 requires tested plans
11A model that hits its technical metrics but misses the business KPIs still passes evaluationFALSE — iterate, refine, or start over so it aligns with the business objectives
12"Reflecting real-world differences" excuses biasFALSE — amplification/perpetuation matter

Scoring: 11-12 = Exam ready | 9-10 = Almost | <9 = Review


Scoring Summary

ModeScoreMax
Flashcards___/2222
Scenarios___/88
Pattern Match___/1010
Fill-in___/1010
True/False___/1212
TOTAL___/6262

Rating: 53+ = mastered · 40-52 = strong · 27-39 = review · <27 = re-study.