Executive Summary
AIC (AI Credits) consumption could not be computed for the last 24 hours because no span, error, or log event in Sentry for the gh-aw project contains any recognized AIC field (gh-aw.aic, gh_aw.aic, agent_usage.aic, aic). Grafana Tempo could not be queried at all in this session because the MCP server only exposed list_datasources/get_datasource — the trace-query tools (tempo_get-attribute-names, tempo_traceql-search, tempo_get-trace) were not available. This is reported as an observability gap, not as zero usage.
Total AIC: unknown (not zero — instrumentation does not currently emit AIC on spans).
Key Metrics
| Metric |
Value |
| Events analyzed |
100 spans sampled (Sentry spans dataset, last 24h, paginated sample; gen_ai.chat sub-query returned 25 chat spans) |
| Events with AIC data |
0 |
| Events with AIC data (Sentry) |
0 |
| Events with AIC data (Grafana) |
unknown (tool unavailable) |
| Total AIC |
unknown |
| Unique workflows |
unknown (no github.workflow/gh_aw.workflow attribute present on sampled spans) |
| Avg AIC/event |
n/a |
| P95 AIC/event |
n/a |
Top 10 Workflows by AIC Consumption
Not computable — no workflow-attribution fields (github.workflow, github.workflow_ref, workflow.name, gh_aw.workflow) nor AIC fields were present on any sampled span. Spans did carry transaction values such as gh-aw.activation.setup, gh-aw.pre_activation.setup, gateway.request, and gen_ai.chat (with model names like claude-sonnet-5, copilot/gpt-5.4, copilot/claude-haiku-4.5, openai/gpt-5.6-sol), but these are span/transaction names, not workflow identifiers, and carry no numeric AIC value.
Grafana AIC Findings
Grafana AIC: not queryable in this run.
list_datasources succeeded and identified a Tempo datasource (uid: grafanacloud-traces, type tempo), confirming tracing data should exist for gh-aw.* services.- However, the Grafana MCP toolset enabled for this session did not include any Tempo query tools (
tempo_get-attribute-names, tempo_traceql-search, tempo_get-trace); only list_datasources/get_datasource were callable.
- As a result, no TraceQL query (e.g.
{.service.name =~ "gh-aw.*"}) could be executed and no trace-level verification of AIC attributes was possible via Grafana in this run.
- This is a tool-availability gap distinct from a data gap — it should be re-attempted with the Tempo query tools enabled/exposed.
Data Quality and Gaps
Data Quality and Gaps
- Events missing workflow identifiers: 100/100 sampled spans — no
github.workflow, github.workflow_ref, workflow.name, or gh_aw.workflow attribute observed in the returned fields for any span in spans, errors, or logs datasets.
- Events missing AIC attributes: 100/100 sampled spans (
errors dataset: 0 results found in 24h window; logs dataset: 10 sample log lines returned, all internal LLM-runtime trace logs with no AIC fields).
- Assumptions/fallback fields used: none could be applied — no field from the recognized AIC list (
gh-aw.aic, gh_aw.aic, agent_usage.aic, aic) appeared in any queried payload, so no zero-fill vs. absent distinction could even be exercised for Sentry spans.
- Sentry-specific caveats:
search_events (AI-powered) was not available in this MCP build; list_events (direct query syntax) was used instead. Query span.op:ai* returned 0 results; a broader empty-query pull of spans (sorted -timestamp, 24h) returned a representative sample of 100 spans across ~2 hours of activity (many workflow runs), dominated by gateway.request/MCP proxy spans (POST /mcp/*), gen_ai / gen_ai.chat spans, and gh-aw.*.setup activation spans — none carrying AIC fields. A further targeted query for span.op:gen_ai.chat (fields including gh-aw.aic, gh_aw.aic, agent_usage.aic, github.workflow, github.run_id) returned 25 chat spans with only id, span.op, span.description, timestamp populated — no AIC or workflow-identifier fields rendered.
- Grafana-specific caveats: Tempo datasource exists (
grafanacloud-traces) but no query tool was exposed to execute TraceQL or fetch attribute names/values in this session; treat as "Grafana data unavailable" per runbook, not as confirmed absence of AIC instrumentation.
Recommendations
- Instrument AIC emission on spans: Verify emit-side logic in
actions/setup/js/send_otlp_span.cjs to confirm whether gh-aw.aic (or an equivalent cost/usage attribute) is being set as a numeric OTLP span attribute at all; if the attribute is not emitted, add it so both Sentry and Grafana can aggregate AIC per workflow/run.
- Add workflow/run attribution: Ensure spans carry
gh-aw.workflow.name and gh-aw.run.status/run-id attributes consistently (per the OTLP emit-side semantics noted in the runbook) so future reports can group by workflow — currently no span in the sample carried any workflow-identifying field.
- Re-enable Grafana Tempo query tools: The Grafana MCP integration for this workflow should expose
tempo_get-attribute-names, tempo_traceql-search, and tempo_get-trace (not just datasource listing) so Tempo-side AIC/cost data can be cross-checked against Sentry.
- Re-run with AI-powered
search_events: If an embedded LLM provider becomes available in the Sentry MCP build, re-run using search_events for potentially richer field discovery/aggregation than the direct list_events query syntax used here.
References
Generated by 📊 Daily AIC Consumption Report (Sentry + Grafana OTel) · goose · sonnet50 · 34.2 AIC · ⌖ 10.4 AIC · ⊞ 11.2K · ◷
Executive Summary
AIC (AI Credits) consumption could not be computed for the last 24 hours because no span, error, or log event in Sentry for the
gh-awproject contains any recognized AIC field (gh-aw.aic,gh_aw.aic,agent_usage.aic,aic). Grafana Tempo could not be queried at all in this session because the MCP server only exposedlist_datasources/get_datasource— the trace-query tools (tempo_get-attribute-names,tempo_traceql-search,tempo_get-trace) were not available. This is reported as an observability gap, not as zero usage.Total AIC: unknown (not zero — instrumentation does not currently emit AIC on spans).
Key Metrics
spansdataset, last 24h, paginated sample;gen_ai.chatsub-query returned 25 chat spans)github.workflow/gh_aw.workflowattribute present on sampled spans)Top 10 Workflows by AIC Consumption
Not computable — no workflow-attribution fields (
github.workflow,github.workflow_ref,workflow.name,gh_aw.workflow) nor AIC fields were present on any sampled span. Spans did carrytransactionvalues such asgh-aw.activation.setup,gh-aw.pre_activation.setup,gateway.request, andgen_ai.chat(with model names likeclaude-sonnet-5,copilot/gpt-5.4,copilot/claude-haiku-4.5,openai/gpt-5.6-sol), but these are span/transaction names, not workflow identifiers, and carry no numeric AIC value.Grafana AIC Findings
Grafana AIC: not queryable in this run.
list_datasourcessucceeded and identified a Tempo datasource (uid: grafanacloud-traces, typetempo), confirming tracing data should exist forgh-aw.*services.tempo_get-attribute-names,tempo_traceql-search,tempo_get-trace); onlylist_datasources/get_datasourcewere callable.{.service.name =~ "gh-aw.*"}) could be executed and no trace-level verification of AIC attributes was possible via Grafana in this run.Data Quality and Gaps
Data Quality and Gaps
github.workflow,github.workflow_ref,workflow.name, orgh_aw.workflowattribute observed in the returned fields for any span inspans,errors, orlogsdatasets.errorsdataset: 0 results found in 24h window;logsdataset: 10 sample log lines returned, all internal LLM-runtime trace logs with no AIC fields).gh-aw.aic,gh_aw.aic,agent_usage.aic,aic) appeared in any queried payload, so no zero-fill vs. absent distinction could even be exercised for Sentry spans.search_events(AI-powered) was not available in this MCP build;list_events(direct query syntax) was used instead. Queryspan.op:ai*returned 0 results; a broader empty-query pull ofspans(sorted-timestamp, 24h) returned a representative sample of 100 spans across ~2 hours of activity (many workflow runs), dominated bygateway.request/MCP proxy spans (POST /mcp/*),gen_ai/gen_ai.chatspans, andgh-aw.*.setupactivation spans — none carrying AIC fields. A further targeted query forspan.op:gen_ai.chat(fields includinggh-aw.aic,gh_aw.aic,agent_usage.aic,github.workflow,github.run_id) returned 25 chat spans with onlyid,span.op,span.description,timestamppopulated — no AIC or workflow-identifier fields rendered.grafanacloud-traces) but no query tool was exposed to execute TraceQL or fetch attribute names/values in this session; treat as "Grafana data unavailable" per runbook, not as confirmed absence of AIC instrumentation.Recommendations
actions/setup/js/send_otlp_span.cjsto confirm whethergh-aw.aic(or an equivalent cost/usage attribute) is being set as a numeric OTLP span attribute at all; if the attribute is not emitted, add it so both Sentry and Grafana can aggregate AIC per workflow/run.gh-aw.workflow.nameandgh-aw.run.status/run-id attributes consistently (per the OTLP emit-side semantics noted in the runbook) so future reports can group by workflow — currently no span in the sample carried any workflow-identifying field.tempo_get-attribute-names,tempo_traceql-search, andtempo_get-trace(not just datasource listing) so Tempo-side AIC/cost data can be cross-checked against Sentry.search_events: If an embedded LLM provider becomes available in the Sentry MCP build, re-run usingsearch_eventsfor potentially richer field discovery/aggregation than the directlist_eventsquery syntax used here.References
grafanacloud-traces