A plausible part isn't necessarily the right part.
A simulated hardware catalog task tested whether an assistant-selected item matched the actual specification rather than merely sounding reasonable.
Bounded product or illustrative example. This is not a customer result or a general reliability claim.
The story.
The founding task gym presented item choices and constraints, measuring wrong-item selections, search steps and completion. This is a cooperative task, not negotiation, and illustrates the kind of precise work assistants must do in real services.
The failure beneath the headline.
Correctness depends on available inventory, compatibility, model numbers, constraints and changing context. This is also a strong candidate for making merchants assistant-ready.
From failure to better performance.
Test part selection against compatibility and constraints so assistants can do more than give a plausible answer.
What this changes.
A near-match is not the right part, and placing an order is not completing a repair.
Try the organization example ↗