Data sources & requirements
StudioLens requirements are path-scoped. The Dataverse (Direct) build reads transcripts live over the Dataverse Web API and needs an org-data CSV; the Fabric build reads Delta tables from a Lakehouse that its notebooks populate. Both share the same transcript contract.
Inputs by path
- Dataverse (Direct) —
conversationtranscriptsover the Dataverse Web API (OData.Feed,/api/data/v9.2/) plus an Org Data CSV for user and department context. - Fabric — Delta tables in a Lakehouse, populated by the StudioLens notebooks, read over the Lakehouse SQL endpoint.
Permissions
| Path | Source | Permission |
|---|---|---|
| Dataverse (Direct) | Conversation Transcript table | Read via an Organizational login (System Customizer / Environment Maker) |
| Dataverse (Direct) | Org Data CSV | Read access to the CSV (SharePoint, local/synced, or UNC — gateway where required) |
| Fabric | Dataverse ConversationTranscript |
Graph application permissions + Dataverse Read on ConversationTranscript |
| Fabric | PPAC MCSMessages (message credits) | Global or Billing Administrator to export the reports |
| Fabric | Lakehouse | Fabric capacity (F2+) and a Lakehouse |
Transcript contract (both paths)
The transcript-analysis model produces a consistent set of tables:
| Table | Holds |
|---|---|
agent_sessions |
One row per conversation session. |
agent_turns |
One row per turn. |
agent_errors |
Error events. |
agent_subagents |
Sub-agent calls. |
agent_catalogue |
Agents seen in the transcripts. |
agent_performance |
Latency / performance signals. |
Consumption contract (Fabric only)
| Table | Holds |
|---|---|
credit_consumption_tenant |
Tenant-level message-credit consumption. |
credit_consumption_agent |
Per-agent message-credit consumption. |
credit_consumption_user |
Per-user message-credit consumption. |
The PPAC message-credit pages are Fabric-only. On the Dataverse (Direct) build the Credit Consumption page ships hidden and unpopulated — do not expect credit analysis on that path.
Retention
Dataverse retains conversation transcripts for approximately 30 days. For historical analysis beyond that window, use the Fabric build, which lands transcripts into the Lakehouse before the retention window closes.
