Raw IQ Glossary
Raw IQ is the underlying model’s general problem-solving ability before the harness, context, tools, memory, routing, and ownership layer determine what it can actually accomplish.
The phrase is useful because model intelligence is real. A stronger model can reason better, write better, code better, recover better, and notice patterns a weaker model misses. It would be foolish to pretend otherwise.
But raw IQ is not the work system.
A high-IQ model in a poor harness may fail because it lacks the right context, sees too many tools, cannot preserve state, or has no verification loop. A less capable model in a tight harness may perform well on routine, inspectable, center-of-distribution tasks. The practical question is therefore not only how smart the model is, but where that intelligence sits and how it is governed.
Raw IQ matters most when the work is novel, ambiguous, or hard to check. It matters less when the work is routine, bounded, and easy to verify. This is why Task Distribution and Routing Logic matter.
See also
Harness · Task Distribution · Routing Logic · Team Harness · Verification Gap