docs: expose hosted diagnosis execution status in the quickstart

README.md

@@ -82,6 +82,10 @@ result = post("/diagnose/why-failed", {
     "content": json.dumps(trace), "format": "otel", "include_fixes": False,
 }, token)
 del token
+execution = result.get("execution")
+if not isinstance(execution, dict):
+    execution = {}
+print("Detector execution:", execution.get("status", "unavailable"))
 print("Failure signals:", result.get("failure_count"))
 for finding in result.get("all_detections", []):
     print(json.dumps({name: finding.get(name) for name in (
@@ -91,7 +95,14 @@ for finding in result.get("all_detections", []):
 
 Ingestion accepts work asynchronously. The diagnosis request above analyzes the
 submitted content; it does not prove that background analysis of the stored run
-has completed. Review the reported agent, spans, evidence and next action against
+has completed. Execution metadata is not available on every deployment. Missing
+metadata means coverage is unavailable, not complete. `partial`, `failed`, or
+`unavailable` execution must not be interpreted as a clean diagnosis, even with
+zero findings. Keep any partial findings for review. A `complete` execution
+status describes reported detector operations, not universal detector coverage,
+calibration validity, or task success.
+
+Review the reported agent, spans, evidence and next action against
 your run. Zero signals is not proof of success, and a suggested fix is not proof
 that the task will work after a change. `include_fixes=False` avoids requesting
 optional generated fixes; it does not promise that every hosted detector is free

docs/plans/hosted-execution-docs.md

@@ -0,0 +1,33 @@
+# Hosted diagnosis execution reporting
+
+**Status:** Active
+**As of:** 2026-09-10
+**Owner:** engineering
+**Evidence scope:** documentation-only candidate
+
+The canonical hosted example currently prints findings without displaying
+detector execution coverage. A separately tested backend candidate exposes a
+sanitized execution summary, but it is not yet deployed. Display that summary
+when present and explicitly report unavailable when absent. Explain incomplete
+and partial results without claiming detector accuracy or a completed workflow.
+
+Do not change SDK runtime, dependencies, version, authentication, ingestion,
+package publishing or deployment. Preserve the existing hosted example and
+its warning that ingestion is asynchronous. This update is safe before the
+backend change: absent metadata means unavailable, not an assumed clean run.
+
+Verify Markdown Python blocks compile and execute only the extracted reporting
+statements against local complete/partial/failed/missing inputs; no network,
+credential reads or customer files. Review the exact diff before publication.
+DONE: reviewed docs and local example checks; no SDK/runtime score credit.
+
+## Local evidence
+
+All five Python documentation fragments compile with top-level await permitted
+for the pre-existing async examples; the first ordinary-script compile attempt
+failed on their top-level await and is not recorded as a pass. The hosted
+example itself parses as ordinary Python. Executed only its three new reporting
+statements against eight local inputs: missing, null, non-object and empty
+metadata, then complete/partial/failed/unavailable. All eight passed. No network,
+credential prompt, file ingestion or SDK method was executed. The example
+reports server-supplied status; it is not an independent coverage verifier.