Junior Exercise 2 - Bug Story From a Failing Example

Rating

• size: 2
• complexity: 2

Purpose

Practice using AI to generate and compare debugging hypotheses while still grounding the investigation in concrete evidence.

Scenario

A loyalty-points service is crediting points for returned items when it should be subtracting them.

Known failing examples:

• purchase of 100 earns 100 points
• return of 40 should reduce points by 40
• current behavior adds 40 instead

Pseudocode:

if event.type == "purchase":
  points = event.amount
else:
  points = abs(event.amount)

Task

Use AI to propose likely bug explanations and a minimal fix.

Then decide what you would verify before trusting the fix.

Expected output

• a hypothesis list ranked from most to least likely
• one minimal code or logic change
• a minimal reproduction or test idea that proves the fix

Good AI uses

• comparing alternative explanations
• asking what hidden assumptions exist in the current logic
• asking for a minimal reproduction idea

Verification focus

• prove the defect with a concrete example
• verify that the fix handles purchase and return paths
• check whether negative amounts or refund events create new edge cases

Anti-patterns to watch

• asking for a full rewrite when the defect is localized
• accepting the first plausible explanation
• fixing the symptom without verifying event semantics

Debrief prompts

• what made one hypothesis stronger than another
• what evidence would disprove your chosen explanation
• what would a weaker engineer be tempted to trust too quickly