Collectors

Decides which metrics get collected on which conversations (the config form of a collector). Its metrics: list references both custom metrics you defined and Roark’s system metrics, all by slug.

# yaml-language-server: $schema=https://roark.ai/roark-config.schema.json
kind: collector
name: consent-on-frontdesk
modality: call
status: ACTIVE
metrics:
  - consent_collection_consent_obtained
  - refund_policy_accuracy
filters:
  - conditions:
      - type: AGENT
        key: frontdesk
        operator: EQUALS

name is the slug config identity. displayName (optional) is the collector’s name as it appears in Roark, where spaces and capitals are allowed; it defaults to name. Set it to a collector’s exact existing name to bring one you created in the app under config (apply takes it over instead of creating a duplicate), for example displayName: "Credit Card Apr" with name: credit-card-apr.
modality is call or chat and is immutable once created (make a new collector to switch).
Metrics are referenced by their stable slug, not a UUID, and must be visible to the project and support the collector’s modality.
Filters are condition groups: groups OR together, conditions within a group AND. An AGENT condition’s key is a config-managed agent name (resolved on apply); every other type (CALL_SOURCE, CALL_PROPERTY, INTEGRATION) uses its key/value verbatim. Omit filters to score every matching conversation.

Per-collector metric settings

Some metrics take a setting that applies only in this collector, the same setting the app’s inline metric config sets. Write the metric as an object with metric (its slug) and its setting, next to plain slugs:

kind: collector
name: credit-card-apr
displayName: Credit Card Apr
modality: call
metrics:
  - metric: property_transcript_mismatch
    verifiedProperties: [purchase_apr_total, extracted_purchase_apr_total]
  - apr_retrieved

Setting	Metrics	What it does
`verifiedProperties`	Property Verification (`property_transcript_mismatch`, `property_transcript_mismatch_count`, `stt_property_capture_rate`)	Checks only these call properties against the transcript. Omit to check every custom property.
`excludedPhrases`	Voice consistency	Agent lines containing one of these phrases are not judged, such as a recorded prompt in a different voice.
`voicemailScript`	Voicemail	The message the agent is expected to leave on voicemail.
`screeningBrief`	Call screening	What the agent should say when a screener asks who is calling and why.

A metric takes at most one setting, and only the one listed for it. Anything else fails the apply with an error naming the metric and the setting it takes. A misspelled setting fails validation rather than being ignored.
The file is the source of truth. A metric listed without a setting runs with its default, so a setting made in the app is replaced on the next apply that changes the collector. Keep the setting in the file.
Tool and knowledge base scoping differ per environment (they reference ids), so they are set in the app or through the metric policies API (metrics[].settings) instead.

Re-scoring past calls with the collector’s settings

Changing a setting applies to conversations scored afterwards. To re-score earlier calls with the collector’s settings, create a metric collection job with the collector’s id as policyId:

curl -X POST https://api.roark.ai/v1/metric/collection-jobs \
  -H "Authorization: Bearer $ROARK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "callIds": ["3f0c2a9e-6b1d-4c3e-9a51-0d2b8e7f4c11"], "policyId": "<collector id>" }'

Each metric runs exactly as the collector runs it, and metrics defaults to all of the collector’s metrics. The collector’s filters are not applied: the calls you list are scored as given.

For the full field reference of every kind, see the Config DSL reference schema.

Getting Started

Simulations

Observability

Metrics

Config as Code

Autoimprove

Recipes

Integrations

Enterprise

SDKs & Developer Tools

Resources

Per-collector metric settings

Re-scoring past calls with the collector’s settings

​Per-collector metric settings

​Re-scoring past calls with the collector’s settings

Per-collector metric settings

Re-scoring past calls with the collector’s settings