[token-consumption] Daily AIC Consumption Report - 2026-07-07

[github-actions]

Executive Summary

Total AIC (last 24h): not available — observability gap in both backends. No span in either Sentry or Grafana exposed a queryable numeric gh-aw.aic value during the window, so AIC consumption cannot be quantified. This is a telemetry gap, not zero usage: 12 distinct agentic workflows were demonstrably active in Sentry (attributed via gh-aw.workflow.name), and gen_ai LLM spans were emitted throughout the day — but the AIC-bearing spans never surfaced.

• Sentry AIC queryable? No. has:gh-aw.aic returned 0 results over 24h. The gh-aw.aic attribute is emitted (per instrumentation) only on the conclusion spans of token-owning jobs (agent/detection/engine); none of those conclusion spans appear in-window. Only setup/activation/pre_activation span types and gen_ai proxy spans are present, and none carry AIC.
• Grafana AIC queryable? No — hard tooling gap. The Grafana MCP build in this run exposes only list_datasources / get_datasource. The Tempo query tools (tempo_traceql-search, tempo_get-attribute-names, tempo_get-trace) are absent, so no TraceQL query could be executed against the Tempo datasource at all.
• Trend note: Activity is healthy (12 workflows, continuous gen_ai spans), but AIC attribution to those workflows is currently invisible in the telemetry query surface.

Key Metrics

Metric	Value
Events analyzed (Sentry, capped sample)	≥100 spans/query (see note)
Events with AIC data	0
Events with AIC data (Sentry)	0 (has:gh-aw.aic → 0)
Events with AIC data (Grafana)	unknown (Tempo not queryable)
Total AIC	unavailable (no numeric AIC present)
Unique workflows	12
Avg AIC/event	undefined (no AIC values)
P95 AIC/event	undefined (no AIC values)

Note: this MCP build has no search_events and no server-side aggregation; list_events caps at 100 rows/query with no offset/cursor, so exact 24h span totals cannot be computed. Counts below are a lower-bound activity proxy from a capped 100-span sample of workflow-attributed spans.

Top 10 Workflows by AIC Consumption

AIC is unrankable (attribute absent on all spans). The table below ranks by observed workflow-attributed span count as an activity proxy — it is not AIC and must not be read as cost.

Workflow	Events (proxy)	Total AIC	Avg AIC/Event
Package Specification Enforcer	13	n/a	n/a
Daily Go Function Namer	13	n/a	n/a
Constraint Solving — Problem of the Day	13	n/a	n/a
Test Quality Sentinel	11	n/a	n/a
Matt Pocock Skills Reviewer	9	n/a	n/a
Impeccable Skills Reviewer	9	n/a	n/a
Design Decision Gate 🏗️	8	n/a	n/a
Typist - Go Type Analysis	7	n/a	n/a
PR Code Quality Reviewer	6	n/a	n/a
PR Sous Chef	5	n/a	n/a

(Also observed: GitHub MCP Structural Analysis ×3, Daily AIC Consumption Report ×3.) Highest-AIC event per workflow: not determinable — no AIC values emitted.

Grafana AIC Findings

Grafana AIC: NOT queryable in this run (tooling gap, not a data gap).

• The Grafana MCP server exposed only 2 tools this run: list_datasources and get_datasource. The Tempo query tools required by the query loop — tempo_traceql-search, tempo_get-attribute-names, tempo_get-attribute-values, tempo_get-trace — are absent from the tool surface. No TraceQL query could be issued.
• The Tempo datasource itself is healthy and reachable: get_datasource(uid=grafanacloud-traces) → name grafanacloud-ghaw-traces, type tempo, url tempo-prod-10-prod-eu-west-2.grafana.net. So the gap is the MCP tool build, not the backend.
• Evidence line: grafana get_datasource --uid grafanacloud-traces → {"type":"tempo","uid":"grafanacloud-traces",...} succeeded; but there is no available tool to run { .service.name =~ "gh-aw.*" && span."gh-aw.aic" > 0 } against it. Per instrumentation notes (send_otlp_span.cjs), gh-aw.aic is emitted as a doubleValue and Tempo is expected to index it for { span."gh-aw.aic" > 0 }; that expectation could not be validated here.
• Grafana trace-explore deep links are omitted (base URL not available / redacted); the datasource UID grafanacloud-traces is provided as a relative reference.

Data Quality and Gaps

Sentry:

• has:gh-aw.aic → 0 results over 24h. Filterability confirmed as working via has:gh-aw.workflow.name → 3+ results, so the 0 is a true absence, not an unfilterable-field artifact.
• AIC-bearing spans are the conclusion spans of token-owning jobs (gh-aw.conclusion.agent|detection|engine). None present in-window — only gh-aw.conclusion.setup, gh-aw.activation.setup, gh-aw.pre_activation.setup, and span.op:gen_ai spans. The gen_ai spans sampled carried no gen_ai.usage.* tokens either.
• errors dataset: 0 events in 24h (checked explicitly). logs dataset: 0 events in 24h (checked explicitly).
• transactions dataset is invalid in this MCP build (accepted datasets: spans, errors, logs, metrics, profiles, replays) — the specified transactions fallback was not executable.
• No search_events tool and no aggregation/pagination-cursor support → exact 24h totals, avg, and P95 are not computable from the tool surface; sample capped at 100 rows/query.
• Events missing workflow identifiers: many — http.server (MCP proxy) and most gen_ai spans lack gh-aw.workflow.name.
• Assumption/fallback: workflow ranking uses observed span count as an activity proxy since AIC is absent.

Grafana:

• Tempo query toolset absent (only datasource-metadata tools available) → AIC unknown, not zero.
• Tempo datasource confirmed present and healthy; the limitation is the MCP tool build, not the backend or the data.

Cross-backend: Spans export to both backends simultaneously; Sentry is treated as canonical to avoid double-counting. Both independently show no queryable AIC this window.

Recommendations

1. Fix the missing AIC-bearing conclusion spans (highest priority). The token-owning agent/detection/engine conclusion spans that carry gh-aw.aic never reached Sentry in the last 24h, despite 12 active workflows. Verify the OTLP export at conclusion time for those jobs (job completion firing the span, exporter flush before job teardown, and jobEmitsOwnTokenUsage gating in actions/setup/js/send_otlp_span.cjs:2124). Without these spans, AIC is invisible regardless of backend.
2. Restore the Grafana Tempo query toolset in the workflow's MCP build. This run's Grafana MCP lacked tempo_traceql-search / tempo_get-trace, making Grafana AIC unverifiable. Enable the Tempo tools so { span."gh-aw.aic" > 0 } can be validated and cross-checked against Sentry.
3. Add a queryable-AIC health check. Since has:gh-aw.aic = 0 is a strong "AIC pipeline broken" signal, add a lightweight daily assertion that at least one gh-aw.aic span exists per active workflow; alert when absent.
4. Ensure gh-aw.aic numeric typing survives ingestion. Confirm Sentry EAP infers gh-aw.aic as float (emitted as doubleValue) and Tempo indexes it, so sum()/avg()/percentile() aggregations work once the spans arrive — enabling real top-consumer ranking next cycle.

References

• Sentry — AIC presence check (0 results): https://github.sentry.io/explore/traces/?query=has:gh-aw.aic&project=4511347087179777&statsPeriod=24h&table=span
• Sentry — workflow-attributed spans (activity proxy): https://github.sentry.io/explore/traces/?query=has:gh-aw.workflow.name&project=4511347087179777&statsPeriod=24h&table=span
• Grafana — Tempo datasource (relative ref, base URL redacted): datasource uid grafanacloud-traces (grafanacloud-ghaw-traces)
• Workflow run: §28866697449

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• awmgmcpg

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "awmgmcpg"

See Network Configuration for more information.

Generated by 📊 Daily AIC Consumption Report (Sentry + Grafana OTel) · 174.1 AIC · ⌖ 36.8 AIC · ⊞ 7.3K · ◷

• expires on Jul 8, 2026, 4:49 AM UTC-08:00