QAVE Studio Docs
QAVE Studio Docs
Sign inGet started — the Studio loop
Getting StartedFirst assurance runQuick assure
ConceptsLifecycleGlossaryHow QAVE Studio worksComing soon
Account & workspaceSignup and loginPeopleDemo creditsRolesSettings
Generate simulation suitesContext templatePII and field enhanceDimensionsAgents and plannerDraft review and Save
ExportExport formatsExport formats (detail)
Assurance SDKSDK overviewQAVE Studio suite fetch APIAgent and JSONL queueLangfuse bridge
Execute and tracesRecipe chatbot (reference SUT)TracingFetch tracing
AssureEvaluate APIRubrics and evidenceReading scores
History
WorkbenchGolden data
CoverageCoverage dashboardCoverage strategiesCoverage advisorTest case quality
RAG AssuranceRAG runsRAG datasetsRAG analytics
Project evalsError analysisUnderstanding scores
Studio guidesDimensionsTuplesConversation modesMulti-lingual testingWorkflowsHuman reviewBest practices
ReferenceAPI and packagesTroubleshooting
Studio guides

Tuples

Tuples

Tuples are specific combinations of dimension values that define individual test scenarios. This guide explains what tuples are, how they drive generation, and how to manage them.


What is a Tuple?

A tuple is a specific combination of dimension values. Each tuple represents one distinct test scenario that your test cases should cover.

Example tuple:

Intent Type = Complaint
Emotional Tone = Frustrated
User Type = New User

This tuple says: "Generate test cases for a new user who is frustrated and filing a complaint." QAVE Studio generates N test cases anchored to this exact scenario, varying only within the constraints of that combination.


Full Tuples vs. Partial Tuples

Full tuples specify a value for every approved dimension. They produce the most precise test cases.

Partial tuples leave some dimensions unspecified. They are useful when:

  • You want to test a specific combination without over-constraining the generation
  • A dimension is not relevant for a particular scenario (e.g., "Emotional Tone" for an informational inquiry)

How Tuples Drive Generation

Each test case in QAVE Studio traces back to a specific tuple. This traceability means:

  • You know exactly which dimension-value combination each test case covers
  • Coverage metrics are based on which tuples have test cases
  • Exports include tuple_id, tuple_label, and dimension_values for downstream analysis

The generation prompt includes the tuple's dimension-value constraints as explicit instructions to the LLM, ensuring the generated conversation reflects that specific scenario.


Tuple Quality Scores

Each generated tuple is scored by the TupleValueAssessor LLM agent (0.0–1.0):

ScoreMeaningAction
0.7–1.0High valueShown prominently; approve first
0.3–0.7Medium valueShown with quality indicator; review carefully
0.0–0.3Low valueAuto-rejected; not shown unless filter enabled

When to accept low scores:

  • A low score indicates the combination has lower assessed test value, not that it's invalid
  • If a low-quality tuple covers a specific risk area in your Known Limitations, it may still be worth generating
  • Re-enable low-quality tuples by adjusting the quality filter in the Tuples page

Managing Tuples

Approve / Reject

  • Click ✓ to approve a tuple for generation
  • Click ✗ to reject a tuple
  • Use Bulk Approve / Bulk Reject for efficiency

Create Manually

  1. Click Add Tuple in the top right
  2. Select a value for each approved dimension
  3. Add a label (optional; auto-generated if not provided)
  4. Save — the tuple is added with APPROVED status

When You're Done

Click Done Reviewing Tuples. This gates test case generation — you can now use the Generate page to select tuples and generate test cases.

⚠️ "Done Reviewing Tuples" is a gate-keeping step. Individual approve/reject actions do not enable the Generate page. You must explicitly click "Done Reviewing" to proceed.


How to Interpret Tuple Labels

QAVE Studio auto-generates a human-readable label for each tuple by concatenating its dimension-value pairs:

Intent=Complaint | Tone=Frustrated | UserType=New User

This label appears in:

  • The Tuples review table
  • Coverage heatmap tooltips
  • Test case metadata
  • Export filenames and Pytest function names

You can customize the label by editing the tuple detail.


Choosing a Coverage Strategy

When clicking "Done Reviewing Dimensions" or triggering manual regeneration, a Coverage Strategy panel lets you choose how tuples are built:

  • Pairwise (recommended): Every pair of dimension values appears in at least one tuple. Fast, 50–100 tuples typical. Best for most projects.
  • N-wise: Every combination of N dimension values is covered. Set N=3 for triple coverage, N=4 for quadruple. Use for high-risk domains.
  • Full Cartesian: Every possible combination. Only practical for very small dimension spaces (≤ 4 dimensions × ≤ 3 values).

Estimated Count Preview

Before confirming, the panel shows an estimated tuple count for the selected strategy. This updates live as you change strategy or N value. Use this to decide whether to use a max_tuples cap.

Max Tuples Cap (Optional)

Enter a number in the "Max tuples" field to limit output. When set:

  • The workflow generates all combinations first, then samples down to your limit
  • Sampling is deterministic — the same inputs always produce the same subset
  • Leave blank for fully autonomous generation (no limit)

Pagination

The Tuples review table and browse page now use server-side pagination (50 tuples per page). Use the ← → controls at the bottom to navigate. The total count is shown alongside the page number.

Filters (status, type) are applied server-side — the count reflects matching tuples only.


Expandable Row Details

Click any tuple row's label (or the chevron icon) to expand it and see:

  • All dimension values with their names
  • Rejection reason (if auto-rejected or rejected by reviewer)
  • Quality score percentage

Collapsed rows show just the label, dimension count, type, quality dots, and status for quick scanning.

Dimensions

Dimensions

Conversation modes

Conversation modes

On this page

What is a Tuple?Full Tuples vs. Partial TuplesHow Tuples Drive GenerationTuple Quality ScoresManaging TuplesApprove / RejectCreate ManuallyWhen You're DoneHow to Interpret Tuple LabelsChoosing a Coverage StrategyEstimated Count PreviewMax Tuples Cap (Optional)PaginationExpandable Row Details