Swarm Learning Systems FAQs

Question 1

How does the skill measure task success?

Accepted Answer

It uses a weighted calculation (rawScore) that combines task duration, error counts, retry attempts, and final success status to generate an implicit feedback signal without requiring manual user input.

Question 2

How does the system handle outdated performance data?

Accepted Answer

The system applies a confidence decay formula with a 90-day half-life, meaning older feedback events lose weight over time to ensure the swarm adapts to the current state of the codebase.

Question 3

What happens when a strategy fails frequently?

Accepted Answer

If a pattern has at least three observations and a failure rate over 60%, it is automatically inverted into an Anti-Pattern, which the swarm is then explicitly instructed to avoid in future prompts.

Question 4

What is the pattern maturity lifecycle?

Accepted Answer

Patterns progress from 'candidate' to 'established' and finally to 'proven' based on positive feedback. Proven patterns receive a 1.5x score multiplier during task planning to encourage their reuse.

Question 5

Can I see why a specific subtask failed?

Accepted Answer

Yes, the Error Accumulator tracks specific error types (validation, timeout, conflict, etc.) and provides a detailed error context that includes stack traces and tool names for better debugging.

Swarm Learning Systems

Key Features

Use Cases

Swarm Learning Systems

Key Features

Use Cases