Dimensions
Dimensions
Dimensions are the axes of variation that define your test space. This guide explains what dimensions are, how to work with them, and how they feed into tuple generation.
What is a Dimension?
A dimension is a meaningful attribute along which test cases should vary. When you change the value of a dimension, you should get a meaningfully different conversation or test scenario.
Example dimensions for a customer service bot:
| Dimension | Values |
|---|---|
| Intent Type | complaint, inquiry, cancellation request, praise |
| Emotional Tone | frustrated, calm, confused, aggressive |
| User Type | new user, returning user, premium member |
| Communication Style | terse, verbose, using slang, formal |
A good dimension produces different bot responses or requires different handling when its value changes.
A poor dimension is too granular (e.g., "User's exact age"), too narrow (applies to one test case), or doesn't produce different bot behaviors.
System Default vs. Project Dimensions
System Default Dimensions are maintained by QAVE Studio administrators and apply to all projects. They represent universally relevant test axes (e.g., Emotional Tone, Communication Style). Admins can promote project dimensions to system defaults.
Project Dimensions are generated specifically for your project, based on your Context Template. They reflect the nuances of your specific AI system (e.g., "Appointment Type" for a dental clinic bot).
Both types appear in the Dimensions page. System defaults are seeded on first use and cannot be deleted by project members.
Dimension Values and Translations
Each dimension has one or more values. Values are the specific instantiations of the dimension (e.g., for "Emotional Tone": frustrated, calm, confused, aggressive).
Adding translated values: For multi-lingual projects, you can add translated versions of values (e.g., "frustrado" for Spanish, "निराश" for Hindi). These translations are used when generating test cases in non-English languages.
How to Review Dimensions
After dimension generation completes, you'll see all generated dimensions in the Dimensions page with a status badge (PENDING, APPROVED, REJECTED).
To approve a dimension:
- Click the ✓ (checkmark) button next to it
- It will be used in tuple generation
To reject a dimension:
- Click the ✗ (X) button next to it
- Rejected dimensions will not be used in tuple generation
To add a dimension manually:
- Click Add Dimension in the top right
- Fill in the name, description, and values
- The new dimension will be added with APPROVED status
Dedup confidence score: If a dimension shows a confidence score, it indicates how likely it is to be a duplicate of an existing dimension:
- 0.0–0.5: Not a duplicate (proceed)
- 0.5–0.85: Uncertain — shown for human review
- 0.85–1.0: High confidence duplicate — auto-rejected
When you're done: Click Done Reviewing Dimensions. This triggers tuple generation with all approved dimensions.
⚠️ The "Done Reviewing Dimensions" action is a gate-keeping step — individual approve/reject actions do not trigger downstream generation. You must explicitly click "Done Reviewing" to proceed.
Promoting a Dimension to System Default
Admin-only. If a dimension is valuable across multiple projects:
- Open the dimension detail
- Click Promote to System Default
- The dimension will appear in the default library for all future projects
How Dimensions Affect Tuple Generation
After you click "Done Reviewing Dimensions", the approved dimensions and their values become the inputs to the tuple generation algorithm.
- More dimensions = more combinations = more tuples
- More values per dimension = more combinations
- The coverage strategy (pairwise, n-wise, full Cartesian) determines which subset of all combinations is generated
See the Coverage Strategy Guide for details.