llm_request_completed: from owned boundary to trusted query
Treat llm_request_completed as a durable analytical contract: the owner emits one documented grain, a fixture proves the fields, and a reviewed query answers the stated question.
- 1
Outcome becomes final
Shared model gateway or provider adapter emits only after token usage, outcome, and provider response are final.
- 2
Contract is bounded
9 required fields preserve the declared grain: One completed model-provider request.
- 3
Fixture is verified
Check types, UTC time, alternate outcomes, idempotency, and every pseudonymous or review-classified field.
- 4
Question is answered
What does each successful AI workflow cost and how well does it perform?
Grain
One completed model-provider request.
Owner
Shared model gateway or provider adapter
Emit when
After token usage, outcome, and provider response are final.
Field contract
Typed fields with explicit privacy boundaries
Keep existing field names and types stable after production queries depend on them. Optional context should remain bounded, documented, and justified by a specific decision.
| Field | Type | Required | Privacy | Meaning |
|---|---|---|---|---|
| timestamp_utc | timestamp | yes | non-sensitive | UTC time at the outcome boundary. |
| event_id | string | yes | non-sensitive | Stable unique identifier used for deduplication. |
| release | string | yes | non-sensitive | Application or service version that emitted the event. |
| account_id | string | yes | pseudonymous | Stable internal account identifier, never an email or name. |
| workflow | string | yes | non-sensitive | Stable product workflow using the model. |
| model | string | yes | non-sensitive | Provider model identifier. |
| input_tokens | number | yes | non-sensitive | Provider-reported input tokens. |
| output_tokens | number | yes | non-sensitive | Provider-reported output tokens. |
| cost_usd | number | no | non-sensitive | Estimated allocation cost, reconciled separately. |
| status | string | yes | non-sensitive | Final provider or workflow outcome. |
Synthetic JSON event
{
"timestamp_utc": "2026-07-28T14:18:44Z",
"event_id": "evt_llm_01",
"account_id": "acct_8f31",
"release": "2026.07.2",
"workflow": "query_explanation",
"model": "model_family_a",
"input_tokens": 820,
"output_tokens": 244,
"cost_usd": 0.0124,
"status": "success"
}Privacy review
Review identifiers before ingestion
This example uses synthetic identifiers. Pseudonymous values can still be personal data, and review fields can expose business or provider context. Apply your own consent, retention, access, residency, and deletion requirements.
account_id: pseudonymous
Validation checklist
Prove the contract before building a dashboard
- Send one known llm_request_completed fixture after the documented outcome boundary.
- Verify all 9 required fields arrive with the documented types.
- Retry the same event identifier and confirm the chosen deduplication behavior.
- Send a controlled failure or alternate outcome when the workflow supports one.
- Run the related SQL over a fixed window and reconcile the result to the fixture.
Common mistakes
Keep one row equal to one durable outcome
- Emitting llm_request_completed before shared model gateway or provider adapter knows the final outcome.
- Mixing several grains in one table, which makes counts and rates ambiguous.
- Replacing controlled categories with raw URLs, payloads, prompts, or error text.
- Changing a field type in place after saved queries and dashboards depend on it.
- Adding identifiers without a documented investigation, access, and retention need.
Use the contract
Query and operationalize the event
Related contracts
Send a fixture before production traffic
Create a free API key, send the synthetic event, and inspect the inferred table before connecting a live workflow.