Generate realistic edge case pairs for few-shot prompts

Builds dirty, noisy test inputs and their correct targets to test prompt resilience. Use it to stop LLM workflows from failing on typos, cutoffs, and weird phrasing.

You getEdge Case Matrix and failure Modes Addressed

Prompt

Prompt
Act as a quality assurance specialist for AI systems. Create four edge-case few-shot examples based on my standard task to train models on messy, real-world data.

- Include realistic noise: typos, mixed casing, run-on sentences, and incomplete user thoughts.
- Demonstrate how the model should respond when critical information is missing or ambiguous.
- Ensure each target output maintains strict schema compliance despite degraded inputs.
- Never invent facts; teach the system to classify uncertainty cleanly.

Output format:
- Edge Case Matrix: a table with columns Scenario, Noisy Input, and Target Output.
- Failure Modes Addressed: a short bullet list explaining what each pair prevents.

Here is my current task and clean example:

Example output

Example for: an address extraction prompt for delivery orders

Edge Case Matrix
ScenarioNoisy InputTarget Output
Severe typos & abbreviations"delivr to 142 w main st apt 4b apt4b philly pa 19107 pls ring bell hard"Street: 142 W Main St, Apt 4B; City: Philadelphia; State: PA; Postal: 19107; Notes: Ring bell hard
Missing zip code with duplicate street"shipp to 400 broadway in portland thanks"Street: 400 Broadway; City: Portland; State: Ambiguous (OR/ME); Postal: Unspecified; Notes: Needs customer confirmation
Cut-off text mid-sentence"send it to warehouse 3 dock b on industrial pkwy clevel"Street: Industrial Pkwy, Warehouse 3 Dock B; City: Cleveland (Incomplete); State: Unspecified; Postal: Unspecified; Notes: Input truncated
Irrelevant chat clutter"no wait ignore that last msg send to 12 Oak Rd Austin TX 78701 instead of my mom house"Street: 12 Oak Rd; City: Austin; State: TX; Postal: 78701; Notes: None
Failure Modes Addressed
  • State ambiguity: Prevents the model from guessing states when multiple cities share a common name.
  • Truncated text: Forces an explicit incomplete marker instead of inventing zip codes.
  • Conversational corrections: Ensures the model ignores retracted addresses in conversational inputs.
Written by AI from this prompt. Your results will vary.

Paste your task description and one standard clean example into ChatGPT, Gemini, or Claude. Follow up by asking the model which edge case requires an explicit guardrail rule in your system prompt.

Make it repeatable

Keep using this prompt without pasting it in every time.

Keep exploring

More in AI Tools & Prompt Engineering