Known Gaps and Live-Tenant Verification Items¶
This solution is preview because several data points still require verification against a live tenant, and several upstream surfaces are themselves preview/beta. None of these block the architecture; they affect specific collectors or mappings.
Structural gaps (no current API closes these)¶
- Per-agent end-to-end USD is not attributable through supported public APIs. Azure Cost Management has no agent dimension; per-agent credit consumption is UI-export only. The report labels any per-agent figure as partial / heuristic / manual and never folds it into authoritative totals.
- Identity correlation between
copilot_package_id↔bot_id↔entra_agent_id↔ the PAYG CSVResource ID↔ the audit-log agent field is not exposed by any single supported API; cross-mapping is heuristic (display name / owner / environment). - Direct-tenant invoice line items at credit/meter granularity are not available via Graph
(
PartnerBilling.Read.Allis for CSP partner contexts).
Items to verify on a live tenant¶
| Item | Affects | What resolves it |
|---|---|---|
| Exact Azure meter names/IDs for Power Platform / Copilot Studio / Agent 365 | normalize_cost_facts.py meter filtering |
Run a Cost Management query grouped by MeterId/Meter over a subscription with known usage |
Whether capacityAllocations returns consumed vs allocated |
preview PP collector | Live call with nonzero usage; inspect payload |
Agent 365 skuPartNumber values ($15 standalone, E7 bundle) |
license mapping | GET /subscribedSkus from a tenant with the SKUs assigned |
Purview audit records-fetch endpoint + CopilotInteraction schema |
beta audit collector | Records-fetch + @odata.nextLink paging + field mapping are now implemented; the auditData field names/casing/availability when surfaced through Graph beta remain unverified — run a query to completion, enumerate records, and confirm the agent-identifying fields |
| Whether Agent 365 ACU consumption is a distinct Azure meter | cost coverage | Export Cost Management while generating known Agent 365 activity |
| Current GA api-version for PP licensing + Azure Cost Management | all collectors | Confirm against the published REST references at build time |
Unit semantics¶
Copilot Studio "credits" and "messages" are different units depending on context and agent complexity (a generative answer or agent action consumes more than one credit). The report documents the unit and the configured unit price it used; it does not assume a fixed credit↔message ratio.