KaziHub Step 3: Domain & Guideline Mastery Exam

This is an open-book exam testing strict guideline compliance. Use the KaziHub Operations Manual in the left pane to evaluate each scenario. Answers must reflect guideline criteria, not personal preference.

Task 1: Negative Constraint Adherence

Prompt Given to Model: "Write a 3-bullet action plan for database indexing. Do not use code blocks, and do not use the words 'optimize' or 'improve'."

Model Completion: • Adjust cache parameters to accelerate read latency. • Execute table re-indexing during scheduled quiet windows using:

SQL

REINDEX TABLE operations_ledger;

• Monitor query execution plans for full-table scans.

Question: "Audit this completion against the prompt constraints and Guideline Section 2. Which rating is correct?"

The completion is fully compliant because the SQL snippet clarifies the instruction.

The completion fails Pillar 2 (Negative Constraints) because it wraps the SQL statement in a markdown code block despite explicit negative instructions.

The completion fails Pillar 4 because it uses an overly technical tone.

The completion is compliant because the prompt did not specify the language for the code block.

Task 2: Code Evaluation & Rule Adherence

Prompt Given to Model: "Write a Python function named filter_active_nodes(nodes) that returns nodes with is_active == True.

Constraint: Use a standard for loop. Do not use list comprehensions or lambda functions."

Model Completion:

Python

def filter_active_nodes(nodes): return [node for node in nodes if node.get("is_active") is True]

Question: "How should an evaluator score this output against the project rubric?"

Score 5 (Flawless) because the list comprehension is cleaner and faster than a for loop.

Score 1 (Critical Failure) because .get() cannot be called on dictionaries in Python.

Score 2 (Substandard) because the model completely ignored the negative constraint prohibiting list comprehensions, despite the code executing cleanly.

Score 4 (Good) because negative constraints are treated as optional recommendations.

Task 3: Factual Grounding & Hallucination Check

Prompt Given to Model: "Briefly state the 4 core object types Git stores inside .git/objects."

Model Completion: "Git stores all project data internally using four primary object types: Blobs (raw file contents), Trees (directory listings), Commits (metadata and tree pointers), and Checkpoints (tracking index staging states)."

Question: "What is the factual accuracy status of this model completion?"

Fully accurate; Checkpoints represent the staging index in modern Git versions.

Hallucination detected; Git stores Blobs, Trees, Commits, and Annotated Tags. 'Checkpoints' does not exist as an internal object typ

The model confused Blobs with Trees.

Git only stores two object types: Commits and Deltas.

Task 4: Evidence-Based Rationale Writing

Scenario: "Draft the exact audit justification you would submit for Task 1 above. Your justification must strictly follow the High-Scoring Formula from Section 4 of the manual: [Claim of Superiority / Rating] + [Guideline Citation] + [Direct Evidence / Quote]."