[amilandi]

Summary

• classify every FlightCheck run invoked with --invocation-source adk as declarative_agent
• preserve custom_agent attribution for runs from main-ca
• emit telemetry from the early native-DA maker-diagnostics path, which previously returned before both telemetry families
• map diagnostic outcomes to the existing ADK pass / partial / fail taxonomy
• keep existing branch-based attribution for CLI, installer, and connect runs

Root cause

ADK /flightcheck runs had two independent gaps:

1. Runs from topic branches emitted agentType=unknown, so DA dashboards filtered them out.
2. Native DA diagnostics exited before the legacy FlightCheck and additive ADK telemetry blocks, so no event was sent at all.

The ADK invocation path is authoritative for DA attribution; only the CA shipping branch overrides it.

Live debugging

An instrumented hybrid/local run posted 52 PROD events and received HTTP 204, confirming the collector path is healthy. Raw FlightCheck Kusto remains delayed, with its newest visible row currently dated October 26.

Validation

• Manually triggered a flightcheck from this branch and validated that it correctly propagates to aria
• python -m pytest -q tests/flightcheck/test_telemetry.py tests/flightcheck/test_cli.py (154 passed)
• python -m ruff check solutions/ess-maker-skills/scripts/flightcheck/telemetry.py solutions/ess-maker-skills/scripts/flightcheck/cli.py tests/flightcheck/test_telemetry.py tests/flightcheck/test_cli.py
• git diff --check

[Dawn Jeong (daeunJe0ng)]

[copilot-review] Request changes: the diagnostic outcome mapping is sound, but the new telemetry path can misattribute the selected agent, tenant, and connector, and the invocation-aware agent classification reaches only the legacy event family. Static review of f6dd640; tests and live dashboard behavior were not independently verified.

[Dawn Jeong (daeunJe0ng)]

[copilot-review] Request changes at 7a3d6aae7160f3e9fd9c75ffa89ca1e1853a1204.
The shared invocation-aware classification is sound, and all four previous attribution findings are fixed for their original triggers.
I withdraw those findings.
One P2 integration issue remains: newly emitted native diagnostics can be reported as Ready by the legacy projection, and the checked-in readiness queries misclassify diagnostic outcomes.
Details and the suggested fix are in the inline comment.
Static review only; tests, builds, live telemetry, and deployed dashboards were not independently exercised.

[Dawn Jeong (daeunJe0ng)]

[copilot-review] Approve at 53285b75c0fef722bf0b36e8e3a5756621e8a539. No high-risk blocker identified. Previous attribution findings and the original diagnostic-outcome defects are fixed. The remaining connector success-rate issue is non-blocking and recorded in the existing thread for a separate follow-up PR. Static review only; tests, live telemetry, and deployed dashboards were not independently verified.