The original title is "How to Black-Box Test MonkeyCode Task Reliability" which is 8 words. Let me check if it meets the criteria.

Short summary

This article presents a detailed black-box test plan for MonkeyCode's AI task management system, covering adapter design, invariant assertions, fault injection scenarios, and reliability metrics. It emphasizes testing distributed workflow edge cases—client disconnects, duplicate submits, worker termination, cancel races—rather than happy-path demos. The plan includes concrete fixtures, a fault injection matrix, and a four-axis scoring framework (availability, safety, liveness, explainability) that resists collapsing into a single misleading reliability score.

• •Defines a black-box adapter over MonkeyCode's public API surface to test task reliability without coupling to internals
• •Provides a fault injection matrix covering disconnects, duplicates, worker death, timeouts, cancel races, and storage interruptions
• •Proposes four-axis scoring (availability, safety, liveness, explainability) and distribution-based metrics like p95 recovery time

Generated with AI, which can make mistakes.