Outcome
Set up an A/B test. kind broadcast: on a DRAFT broadcast, 2-5 arms { key, weight, templateVersion?, subject? } (an arm without overrides sends the broadcast as-is), optional holdoutPercent (0-50, receives nothing) and testPercent (1-99: only that share is tested at launch; the winner goes to the rest on promote). kind journey: journeyId and the nodeId of a split node in the draft; arms and holdout come from the node. Each contact's arm is hash(experiment id, contact id), recorded once. controlArm defaults to the first arm. Conversion, within goal.windowHours after exposure (send, split or withheld holdout): goal.kind event (default) = the lifecycle eventType (e.g. purchase); click = a tracked-link click on the experiment's own sends, only URLs starting with urlPrefix if set.
Before you start
- Verify identity and pin the intended organization and environment before discovering the action.
- Discover a granted Nurture workspace. The application permission is workspace:write; required workspace access is read_write. A permission alone does not supply a workspace grant.
- Organization and workspace come from verified context. Select resourceType/resourceId separately from action input; body fields cannot grant a different organization or workspace. The first authorized product action can initialize an identity-only workspace mirror.
- Obtain authorization for the documented effect or personal-data export before supplying CLI --confirm or MCP confirm:true.
Discover the target
topolo whoami --json
topolo actions get app_topolo_nurture.experiments.create --detail --json
topolo resources topolo-nurture --json
topolo workspaces list topolo-nurture --json
Request
POST /api/experiments
Save action input as payload.json. The reviewed plan separates path identifiers, query fields and HTTP body. Set NURTURE_WORKSPACE_ID from granted resource discovery; it is separate from action input.
Inspect and test this action in the signed action console. A test response still requires the verification below.
Input fields
Use an empty JSON object; verified credential and resource context are separate.
Exact input schema:
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"oneOf": [
{
"type": "object",
"properties": {
"kind": {
"type": "string",
"const": "broadcast"
},
"broadcastId": {
"type": "string",
"minLength": 1,
"maxLength": 64
},
"arms": {
"minItems": 2,
"maxItems": 5,
"type": "array",
"items": {
"type": "object",
"properties": {
"key": {
"type": "string",
"pattern": "^[a-z0-9][a-z0-9_-]{0,31}$"
},
"label": {
"type": "string",
"minLength": 1,
"maxLength": 60
},
"weight": {
"type": "integer",
"minimum": 1,
"maximum": 100
},
"templateVersion": {
"type": "integer",
"minimum": 1,
"maximum": 100000
},
"subject": {
"type": "string",
"minLength": 1,
"maxLength": 200
}
},
"required": [
"key",
"weight"
],
"additionalProperties": false
}
},
"holdoutPercent": {
"type": "integer",
"minimum": 0,
"maximum": 50
},
"testPercent": {
"type": "integer",
"minimum": 1,
"maximum": 99
},
"name": {
"type": "string",
"minLength": 1,
"maxLength": 120
},
"goal": {
"anyOf": [
{
"type": "object",
"properties": {
"kind": {
"type": "string",
"const": "event"
},
"eventType": {
"type": "string",
"enum": [
"signup",
"trial_started",
"trial_ending",
"form_submitted",
"purchase",
"subscription_changed",
"feature_used",
"inactive",
"mql",
"engage_not_now"
]
},
"windowHours": {
"type": "integer",
"minimum": 1,
"maximum": 2160
}
},
"required": [
"eventType",
"windowHours"
],
"additionalProperties": false
},
{
"type": "object",
"properties": {
"kind": {
"type": "string",
"const": "click"
},
"urlPrefix": {
"type": "string",
"minLength": 1,
"maxLength": 500
},
"windowHours": {
"type": "integer",
"minimum": 1,
"maximum": 2160
}
},
"required": [
"kind",
"windowHours"
],
"additionalProperties": false
}
]
},
"controlArm": {
"type": "string",
"pattern": "^[a-z0-9][a-z0-9_-]{0,31}$"
},
"minSamplePerArm": {
"type": "integer",
"minimum": 10,
"maximum": 1000000
}
},
"required": [
"kind",
"broadcastId",
"arms",
"name",
"goal"
],
"additionalProperties": false
},
{
"type": "object",
"properties": {
"kind": {
"type": "string",
"const": "journey"
},
"journeyId": {
"type": "string",
"minLength": 1,
"maxLength": 64
},
"nodeId": {
"type": "string",
"pattern": "^[a-z0-9][a-z0-9_-]{0,31}$"
},
"name": {
"type": "string",
"minLength": 1,
"maxLength": 120
},
"goal": {
"anyOf": [
{
"type": "object",
"properties": {
"kind": {
"type": "string",
"const": "event"
},
"eventType": {
"type": "string",
"enum": [
"signup",
"trial_started",
"trial_ending",
"form_submitted",
"purchase",
"subscription_changed",
"feature_used",
"inactive",
"mql",
"engage_not_now"
]
},
"windowHours": {
"type": "integer",
"minimum": 1,
"maximum": 2160
}
},
"required": [
"eventType",
"windowHours"
],
"additionalProperties": false
},
{
"type": "object",
"properties": {
"kind": {
"type": "string",
"const": "click"
},
"urlPrefix": {
"type": "string",
"minLength": 1,
"maxLength": 500
},
"windowHours": {
"type": "integer",
"minimum": 1,
"maximum": 2160
}
},
"required": [
"kind",
"windowHours"
],
"additionalProperties": false
}
]
},
"controlArm": {
"type": "string",
"pattern": "^[a-z0-9][a-z0-9_-]{0,31}$"
},
"minSamplePerArm": {
"type": "integer",
"minimum": 10,
"maximum": 1000000
}
},
"required": [
"kind",
"journeyId",
"nodeId",
"name",
"goal"
],
"additionalProperties": false
}
]
}
Effects
- Creates one active experiment. Nothing is sent: a broadcast experiment applies when the broadcast launches; a journey experiment applies to contacts reaching the split node of a published version.
Interpret the response
The returned identifier path is $.experiment.id for experiment. Empty lists do not discover a target.
Call app_topolo_nurture.experiments.get with $.experiment.id and confirm status active.
Exact success schema:
{
"type": "object",
"required": [
"experiment"
],
"properties": {
"experiment": {
"type": "object",
"required": [
"id",
"kind",
"status",
"arms",
"controlArm",
"goal"
],
"properties": {
"id": {
"type": "string"
},
"name": {
"type": "string"
},
"kind": {
"type": "string",
"enum": [
"broadcast",
"journey"
]
},
"broadcastId": {
"type": [
"string",
"null"
]
},
"journeyId": {
"type": [
"string",
"null"
]
},
"nodeId": {
"type": [
"string",
"null"
]
},
"arms": {
"type": "array",
"items": {
"type": "object",
"required": [
"key",
"weight"
],
"properties": {
"key": {
"type": "string"
},
"label": {
"type": [
"string",
"null"
]
},
"weight": {
"type": "integer"
},
"variant": {
"type": "object"
}
}
}
},
"controlArm": {
"type": "string"
},
"holdoutPercent": {
"type": "integer"
},
"testPercent": {
"type": [
"integer",
"null"
]
},
"goal": {
"type": "object",
"required": [
"kind",
"windowHours"
],
"properties": {
"kind": {
"type": "string",
"enum": [
"event",
"click"
]
},
"eventType": {
"type": "string"
},
"urlPrefix": {
"type": [
"string",
"null"
]
},
"windowHours": {
"type": "integer"
}
}
},
"minSamplePerArm": {
"type": "integer"
},
"confidenceLevel": {
"type": "number"
},
"status": {
"type": "string",
"enum": [
"active",
"stopped",
"promoted"
]
},
"winnerArm": {
"type": [
"string",
"null"
]
},
"promotedBy": {
"type": [
"string",
"null"
]
},
"promotedAt": {
"type": [
"string",
"null"
]
},
"promotionOverride": {
"type": "boolean"
},
"overrideReason": {
"type": [
"string",
"null"
]
},
"remainderRecipients": {
"type": [
"integer",
"null"
]
},
"stoppedBy": {
"type": [
"string",
"null"
]
},
"stoppedAt": {
"type": [
"string",
"null"
]
},
"stopReason": {
"type": [
"string",
"null"
]
},
"createdBy": {
"type": "string"
},
"createdAt": {
"type": "string"
},
"updatedAt": {
"type": "string"
}
}
}
}
}
Worked examples
Actual dispatcher and migration-created SQLite or Worker handlers; identity/workspace and external service dependencies are simulated. Fixture identifiers and timestamps are synthetic. These responses establish no live authorization, paid generation, credential provisioning or message delivery. Replace synthetic identifiers using credential-scoped discovery. A local success is not evidence of a deployed effect, configured provider or delivered message.
Successful request
Actual dispatcher and migration-created SQLite; verified identity/workspace and external CRM, Engage, Notify, Consent, Nexus, Brand, Generator, Voice and delivery dependencies are simulated. No live authorization, paid generation or external delivery is established.
Input · payload.json:
{
"kind": "broadcast",
"broadcastId": "bc_9194b079231e4246b56ba9a260d4b399",
"arms": [
{
"key": "control",
"weight": 50
},
{
"key": "question",
"label": "Question subject",
"weight": 50,
"templateVersion": 1,
"subject": "Ready?"
}
],
"holdoutPercent": 20,
"testPercent": 50,
"name": "Subject test",
"goal": {
"eventType": "purchase",
"windowHours": 72
}
}
CLI:
topolo actions validate app_topolo_nurture.experiments.create --data @payload.json --json
topolo actions plan app_topolo_nurture.experiments.create --resource-type workspace --resource-id "$NURTURE_WORKSPACE_ID" --data @payload.json --json
topolo actions call app_topolo_nurture.experiments.create --resource-type workspace --resource-id "$NURTURE_WORKSPACE_ID" --data @payload.json --confirm --json
MCP tool and arguments:
{
"tool": "topolo_call_action",
"arguments": {
"actionId": "app_topolo_nurture.experiments.create",
"input": {
"kind": "broadcast",
"broadcastId": "bc_9194b079231e4246b56ba9a260d4b399",
"arms": [
{
"key": "control",
"weight": 50
},
{
"key": "question",
"label": "Question subject",
"weight": 50,
"templateVersion": 1,
"subject": "Ready?"
}
],
"holdoutPercent": 20,
"testPercent": 50,
"name": "Subject test",
"goal": {
"eventType": "purchase",
"windowHours": 72
}
},
"resourceType": "workspace",
"resourceId": "<identifier returned by resource discovery>",
"confirm": true
}
}
Captured local response · HTTP 201:
{
"experiment": {
"id": "exp_e6a4ff5a74cb4129922a9c41aca79ca9",
"name": "Subject test",
"kind": "broadcast",
"broadcastId": "bc_9194b079231e4246b56ba9a260d4b399",
"journeyId": null,
"nodeId": null,
"arms": [
{
"key": "control",
"label": null,
"weight": 50,
"variant": {}
},
{
"key": "question",
"label": "Question subject",
"weight": 50,
"variant": {
"templateVersion": 1,
"subject": "Ready?"
}
}
],
"controlArm": "control",
"holdoutPercent": 20,
"testPercent": 50,
"goal": {
"kind": "event",
"eventType": "purchase",
"windowHours": 72
},
"minSamplePerArm": 100,
"confidenceLevel": 0.95,
"status": "active",
"winnerArm": null,
"promotedBy": null,
"promotedAt": null,
"promotionOverride": false,
"overrideReason": null,
"remainderRecipients": null,
"stoppedBy": null,
"stoppedAt": null,
"stopReason": null,
"createdBy": "u1",
"createdAt": "2026-10-04T00:32:35.068Z",
"updatedAt": "2026-10-04T00:32:35.068Z"
}
}
HTTP 201. Call app_topolo_nurture.experiments.get with $.experiment.id and confirm status active.
Another supported request
Actual dispatcher and migration-created SQLite; verified identity/workspace and external CRM, Engage, Notify, Consent, Nexus, Brand, Generator, Voice and delivery dependencies are simulated. No live authorization, paid generation or external delivery is established.
Input · payload.json:
{
"kind": "broadcast",
"broadcastId": "bc_982e00778f594bae9cc4b6483ff552a6",
"arms": [
{
"key": "control",
"weight": 50
},
{
"key": "question",
"label": "Question subject",
"weight": 50,
"templateVersion": 1,
"subject": "Ready?"
}
],
"name": "Subject test",
"goal": {
"eventType": "purchase",
"windowHours": 72
}
}
CLI:
topolo actions validate app_topolo_nurture.experiments.create --data @payload.json --json
topolo actions plan app_topolo_nurture.experiments.create --resource-type workspace --resource-id "$NURTURE_WORKSPACE_ID" --data @payload.json --json
topolo actions call app_topolo_nurture.experiments.create --resource-type workspace --resource-id "$NURTURE_WORKSPACE_ID" --data @payload.json --confirm --json
MCP tool and arguments:
{
"tool": "topolo_call_action",
"arguments": {
"actionId": "app_topolo_nurture.experiments.create",
"input": {
"kind": "broadcast",
"broadcastId": "bc_982e00778f594bae9cc4b6483ff552a6",
"arms": [
{
"key": "control",
"weight": 50
},
{
"key": "question",
"label": "Question subject",
"weight": 50,
"templateVersion": 1,
"subject": "Ready?"
}
],
"name": "Subject test",
"goal": {
"eventType": "purchase",
"windowHours": 72
}
},
"resourceType": "workspace",
"resourceId": "<identifier returned by resource discovery>",
"confirm": true
}
}
Captured local response · HTTP 201:
{
"experiment": {
"id": "exp_41afd28d70724d678a329ee6bb478a54",
"name": "Subject test",
"kind": "broadcast",
"broadcastId": "bc_982e00778f594bae9cc4b6483ff552a6",
"journeyId": null,
"nodeId": null,
"arms": [
{
"key": "control",
"label": null,
"weight": 50,
"variant": {}
},
{
"key": "question",
"label": "Question subject",
"weight": 50,
"variant": {
"templateVersion": 1,
"subject": "Ready?"
}
}
],
"controlArm": "control",
"holdoutPercent": 0,
"testPercent": null,
"goal": {
"kind": "event",
"eventType": "purchase",
"windowHours": 72
},
"minSamplePerArm": 100,
"confidenceLevel": 0.95,
"status": "active",
"winnerArm": null,
"promotedBy": null,
"promotedAt": null,
"promotionOverride": false,
"overrideReason": null,
"remainderRecipients": null,
"stoppedBy": null,
"stoppedAt": null,
"stopReason": null,
"createdBy": "u1",
"createdAt": "2026-10-04T00:32:35.384Z",
"updatedAt": "2026-10-04T00:32:35.384Z"
}
}
HTTP 201. Call app_topolo_nurture.experiments.get with $.experiment.id and confirm status active.
Refused request · invalid_input
Actual dispatcher and migration-created SQLite; verified identity/workspace and external CRM, Engage, Notify, Consent, Nexus, Brand, Generator, Voice and delivery dependencies are simulated. No live authorization, paid generation or external delivery is established.
Input · payload.json:
{
"kind": "broadcast",
"name": "T",
"broadcastId": "bc_eb04f6f7c31d482d9cad2a0c6d9a2a7a",
"arms": [
{
"key": "a",
"weight": 50
},
{
"key": "b",
"weight": 50
}
],
"goal": {
"eventType": "purchase",
"windowHours": 72
},
"testPercent": 100
}
CLI:
topolo actions validate app_topolo_nurture.experiments.create --data @payload.json --json
topolo actions plan app_topolo_nurture.experiments.create --resource-type workspace --resource-id "$NURTURE_WORKSPACE_ID" --data @payload.json --json
Captured local response · HTTP 422:
{
"error": {
"code": "invalid_input",
"message": "The input does not match the action schema.",
"details": [
{
"origin": "number",
"code": "too_big",
"maximum": 99,
"inclusive": true,
"path": [
"testPercent"
],
"message": "Too big: expected number to be <=99"
}
]
}
}
HTTP 422. Preserve the refusal details and do not claim success.
Refused request · experiment_exists
Actual dispatcher and migration-created SQLite; verified identity/workspace and external CRM, Engage, Notify, Consent, Nexus, Brand, Generator, Voice and delivery dependencies are simulated. No live authorization, paid generation or external delivery is established.
Input · payload.json:
{
"kind": "broadcast",
"broadcastId": "bc_eb04f6f7c31d482d9cad2a0c6d9a2a7a",
"arms": [
{
"key": "a",
"weight": 50
},
{
"key": "b",
"weight": 50
}
],
"name": "T",
"goal": {
"eventType": "purchase",
"windowHours": 72
}
}
CLI:
topolo actions validate app_topolo_nurture.experiments.create --data @payload.json --json
topolo actions plan app_topolo_nurture.experiments.create --resource-type workspace --resource-id "$NURTURE_WORKSPACE_ID" --data @payload.json --json
Captured local response · HTTP 409:
{
"error": {
"code": "experiment_exists",
"message": "An active experiment already tests this; stop or promote it first."
}
}
HTTP 409. Preserve the refusal details and do not claim success.
Refused request · invalid_control_arm
Actual dispatcher and migration-created SQLite; verified identity/workspace and external CRM, Engage, Notify, Consent, Nexus, Brand, Generator, Voice and delivery dependencies are simulated. No live authorization, paid generation or external delivery is established.
Input · payload.json:
{
"kind": "broadcast",
"broadcastId": "bc_eb04f6f7c31d482d9cad2a0c6d9a2a7a",
"arms": [
{
"key": "a",
"weight": 50
},
{
"key": "b",
"weight": 50
}
],
"name": "T",
"goal": {
"eventType": "purchase",
"windowHours": 72
},
"controlArm": "zzz"
}
CLI:
topolo actions validate app_topolo_nurture.experiments.create --data @payload.json --json
topolo actions plan app_topolo_nurture.experiments.create --resource-type workspace --resource-id "$NURTURE_WORKSPACE_ID" --data @payload.json --json
Captured local response · HTTP 422:
{
"error": {
"code": "invalid_control_arm",
"message": "The control arm \"zzz\" is not one of the arms.",
"details": {
"arms": [
"a",
"b"
]
}
}
}
HTTP 422. Preserve the refusal details and do not claim success.
Verify the result
- Call app_topolo_nurture.experiments.get with $.experiment.id and confirm status active.
Errors and recovery
organization_required(HTTP 403): Select an organization available to the verified caller before choosing a Nurture workspace.permission_denied(HTTP 403): Grant the credential the required Nurture workspace permission.workspace_required(HTTP 409): Create a Nurture workspace, or pass the workspace resource context.invalid_input(HTTP 422): Correct the fields named in error.details using inputSchema, then retry.not_found(HTTP 404): List broadcasts or journeys to find a valid id in this workspace.broadcast_not_draft(HTTP 409): Experiments are set up while the broadcast is a draft; reject it back to draft first.journey_archived(HTTP 409): Use an active journey.experiment_exists(HTTP 409): Stop or promote the active experiment (app_topolo_nurture.experiments.list) first.split_node_not_found(HTTP 422): Add a split node to the journey draft with journeys.update, then pass its id.invalid_arms(HTTP 422): Use unique arm keys other than holdout and remainder.invalid_control_arm(HTTP 422): Pass controlArm as one of the arm keys in error.details.arms.template_required(HTTP 422): Set the broadcast templateId before pinning templateVersion on an arm.template_version_unavailable(HTTP 422): Call templates.get for the versions that exist.
Auth or gateway failures may precede this handler. Preserve the exact response shape, error string/code, request id and credential context; no synthetic fixture establishes live credential access.
Recovery boundary
Stop the experiment created in error; a stopped broadcast experiment is ignored at launch.
Related actions
- Get experiment results: Per arm: contacts assigned and exposed, conversions, conversion rate, uplift against the control (absolute and relative) and against the holdout, z and p-value. Statistics: a pooled two-proportion z-test of each variant against the control at a fixed 95% confidence level, alpha split across variants (Bonferroni); a comparison counts only when both arms reach minSamplePerArm exposures (default 100). decision is winner (recommendedWinner set), control_wins, insufficient_sample or no_significant_difference; a winner is never declared without significance. Conversion, within goal.windowHours after exposure (send, split or withheld holdout): goal.kind event (default) = the lifecycle eventType (e.g. purchase); click = a tracked-link click on the experiment's own sends, only URLs starting with urlPrefix if set.
- Stop experiment: Stop an active experiment without a winner. A broadcast experiment stopped before launch is ignored at launch; after launch, recipients already queued are still sent and the untested remainder is never sent. A journey keeps splitting by the node weights but stops recording assignments.
Evidence
Actual dispatcher and migration-created SQLite or Worker handlers; identity/workspace and external service dependencies are simulated. Fixture identifiers and timestamps are synthetic. These responses establish no live authorization, paid generation, credential provisioning or message delivery. Test scenario and capture digests are retained with each fixture.