Overview

Test Generation Settings control what the platform generates for a single project: which requirement types the AI extracts from documents, which test strategies fire, and which severity tiers survive. They are distinct from the system-wide provider/model configuration in AI Settings. Use them to focus generation—for example, "critical + high only" for a release gate, or strategy fan-out paused while you iterate on requirements. Explore the capability at test strategy fan-out.

Before you begin

You must be able to edit the project (open its project editor).
Changes apply to the next Generate tests run—existing tests are unaffected.
Nav path: open the project editor for a project, then click the AI Generation tab. The panel heading reads AI Generation Settings.

Step 1 — Open the AI Generation tab

Open the project editor for your project.
In the tab row, click AI Generation. The panel loads with the heading AI Generation Settings and the summary "Control which requirement types the AI extracts from documents, which test strategies fire on those requirements, and which severity tiers are kept."
While it loads you'll briefly see "Loading test generation settings…". Projects that never customized these inherit server-supplied defaults, so you see what is currently running before changing anything.

Step 2 — Set the strategy fan-out master toggle

At the top is the checkbox Enable strategy fan-out for this project.

On (default): strategy-driven tests are generated according to the Strategies grid below.
Off: no strategy-driven tests are generated—this is the switch for "run no strategies at all." The help text notes that spec-driven tests still generate, and the Requirement types and Severity filters below still apply. Existing tests are unaffected.

If an administrator has disabled fan-out globally, a warning appears: "Strategy fan-out is globally disabled by an administrator (TEST_STRATEGY_FAN_OUT_ENABLED=false). The settings below will be saved but won't take effect until that flag is re-enabled."

Step 3 — Choose requirement types to extract

In the Requirement types to extract card, tick the types the AI should persist when it extracts requirements from uploaded documents (BRDs, specs, backlogs, OpenAPI). Quick actions (top-right): All, None, Field-only, Backlog-only.

Requirement type	What it covers
Field constraint	required, type, length, range, enum, format
Business rule	cross-field / conditional logic
State transition	e.g. pending → processing → completed
Cross-field dependency	depends_on / mutually_exclusive
Security constraint	auth, scopes, PII
Performance constraint	latency / throughput SLA
Data format	timestamps, currency, conventions
Error handling	status codes, error envelopes
Integration dependency	external service contracts
User story	backlog "As a … I want …"
Acceptance criterion	Given / When / Then
Capability	feature area / tag rollup
Use case	multi-endpoint journey

At least one type must be selected—otherwise the inline error "Select at least one requirement type to extract." appears and Save is blocked. Field-only selects field constraint, error handling, and data format; Backlog-only selects user story, acceptance criterion, capability, and use case.

Step 4 — Choose which strategies fire

In the Strategies card, tick the test families to generate. Quick actions: All, None, Defaults (the card lists the current defaults inline). These controls are disabled while fan-out is off.

Strategy	What it emits
Positive (happy path)	valid, expected-success requests
Negative (rule violation)	requests that break declared rules
Boundary (min/max edge values)	edge values around limits
Equivalence partitioning	one representative per input class
Decision table (conditional rules)	combinations of conditional rules
Pairwise (combinatorial)	pairwise parameter combinations
HTTP semantics (405/415/etc.)	wrong method / media-type handling
Auth depth (token/scope checks)	token and scope enforcement
Schema fuzz (type/format violations)	malformed types/formats
Security probes (SQLi/XSS/etc.)	injection / scripting probes
Performance / latency SLA	latency threshold checks

While fan-out is on, at least one strategy must be selected. If none are, the inline error reads "No strategies selected. Pick at least one, or turn off "Enable strategy fan-out" above to disable all strategy-driven tests." and Save is blocked.

Step 5 — Set the severity filter

In the Severity filter card, keep only the tiers you care about—Critical, High, Medium, Low (ISTQB S1–S4). Each option shows an ISTQB hint (for example, Critical = "system-down / breach class…"). Quick actions: All, Critical + High only. Tests outside the selected set are dropped before persistence across every generation path (strategy fan-out, spec-driven, gap/edge)—so this filter applies even when fan-out is off.

Step 6 — Save

Click Save settings (label shows Saving… during the request). Save is disabled while any validation error above is present; hovering the button shows the blocking reason.
On success a green notice appears: "Saved. New settings apply to the next "Generate tests" run." The panel reloads to show the server-canonical resolved view.
If the save fails, a red banner shows the server error.

Troubleshooting

Save disabled / error banner — resolve the inline validation errors first: at least one requirement type, and (while fan-out is on) at least one strategy.
Settings saved but nothing changes in generation — an administrator has globally disabled fan-out (see the warning banner). Ask an admin to re-enable TEST_STRATEGY_FAN_OUT_ENABLED.
New settings didn't affect existing tests — by design; they apply to the next Generate tests run only.

Test Generation Settings