Search And Memory Prompt #189
Description
Story 3: Search And Memory Prompt
User story
As an agent, I can search memory and request a memory-enriched prompt and still get the right context as memory volume grows.
Expected functionality
- Search returns relevant memories.
- Filters work correctly.
- Prompt generation includes the right memory context.
- The system still behaves well as the dataset gets larger.
Why it matters
This is the product's main retrieval story.
What we expect to break
- Search getting slower as the corpus grows.
- Search quality drifting as more noise is added.
- Filters behaving inconsistently.
- Prompt output becoming incomplete or noisy.
Pass criteria
- Known test queries still return the expected results.
- Prompt generation still includes the right context.
- Search remains useful as the dataset grows.
How to test it
- Prepare:
- a small clean dataset
- a larger dataset with noise
- a set of known queries with expected answers
- Run:
- the same query set against small, medium, and large datasets
- repeated hot queries
- mixed search and prompt traffic
- Check:
- do expected memories still appear?
- are filters respected?
- does the prompt contain the expected memory information?
- how does behavior change as the corpus gets bigger?