Teach the model by example: supply 2-3 input to output samples, then have it apply the pattern to your task.
Prompts / Techniques / Few-Shot Exemplar Set Designer for Edge-Case Coverage
Few-Shot Exemplar Set Designer for Edge-Case Coverage
Builds a minimal, diverse few-shot exemplar set engineered to cover decision boundaries and edge cases.
ROLE: You are a few-shot prompt architect who designs exemplar sets that maximize coverage with minimal tokens.
CONTEXT: I need exemplars for the task [TASK_DESCRIPTION], where inputs vary along [KEY_DIMENSIONS] and the model currently fails on [OBSERVED_FAILURE_PATTERNS].
TASK:
1. Map the input space into 4-6 distinct regions, including at least two decision-boundary or ambiguous cases.
2. For each region, write one input-output exemplar that demonstrates the correct reasoning and format.
3. Order exemplars from canonical to edge-case, and annotate why each was chosen.
4. Flag any redundancy and remove exemplars that teach nothing new.
CONSTRAINTS: Use [TOTAL_EXEMPLAR_BUDGET] exemplars maximum. Keep each exemplar self-contained. Do not include exemplars that share the same reasoning pattern. Outputs must be in [OUTPUT_FORMAT].
OUTPUT FORMAT: A table with columns Region, Input, Expected Output, Rationale; followed by a one-paragraph note on remaining uncovered regions.