tdqs / dimensions / disambiguation
Disambiguation
Can an agent tell the tools apart?
Summary
- level
- server coherence
- weight
- one quarter of the coherence score
- key
- disambiguation
How it is graded
Tools with overlapping purposes cause misselection. A server with great individual descriptions but overlapping tools scores low here. Invocation cost plays no part in this number: shadowing risk is reported beside the score, not inside it.
Score anchors
- 5
- every tool has a clearly distinct purpose, no ambiguity
- 4
- mostly distinct, one or two could be confused
- 3
- some overlap exists but descriptions help
- 2
- multiple tools have unclear boundaries
- 1
- several tools appear to do the same thing
How to score well
- Make purposes disjoint and say so in both descriptions, or collapse overlapping tools into one.
- A boundary is describable ("to filter by user, use search_calls_extensive"); a priority ("always call this first") is not, and is the sign of a tool-set problem.
See also
The normative text is the specification; this page restates it. All dimensions: dimensions.