tdqs / dimensions / behavioral-transparency
Behavioral Transparency
Does it disclose behavior beyond what annotations already declare?
Summary
- level
- tool definition
- weight
- 20% of the tool score
- key
- behavioral_transparency
How it is graded
Graded relative to annotations. When annotations exist, the bar is lower: the description earns credit for adding context annotations cannot carry (what gets destroyed, auth requirements, rate limits). Without annotations, the description carries the full disclosure burden. A description that contradicts its annotations scores 1 and raises the Annotation Contradiction flag.
Score anchors
- 1
- no disclosure, or the description contradicts its annotations
How to score well
- Declare the MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint). They lower the disclosure burden on the description.
- Spend the description on what annotations cannot express: what exactly gets destroyed, auth prerequisites, rate limits.
- Never let the description contradict the annotations. That is an automatic 1 here and a public Annotation Contradiction flag.
See also
The normative text is the specification; this page restates it. All dimensions: dimensions.