Signal model
Read each signal for what it proves
Start with the workspace, project, run, and time window you are investigating. A signal is evidence about one layer, not a shortcut around the others.
| Signal | What it means | What it does not prove |
|---|---|---|
| Application/API health | The signed-in application or API responded at the observed time. | A workflow, agent, route, repository, provider, or telemetry pipeline is ready. |
| Workflow execution state | The durable run records its stage, owner, blocker, evidence, and completion state. | The assigned agent is live or every downstream prerequisite is healthy. |
| Agent heartbeat | Agents or Fleets received recent liveness and presence evidence. | Execution progress, inbox consumption, route reachability, repository access, or successful provider use. A fresh heartbeat is liveness evidence only. |
| Runtime-route readiness | Infrastructure reports redacted runtime, console, relay, tunnel, bundle, and preflight readiness. | Application-wide health, workflow success, or repository readiness. |
| Repository readiness | Projects > Repositories or Integrations shows project-scoped attachment, branch, access, and readiness state. | A runtime can clone, an agent is live, or a workflow is progressing. |
| Telemetry availability | Workflow Observability, Metrics, or Logs has current aggregate measurements for the selected scope and time window. | Health when data is missing, stale, partial, delayed, or outside the selected scope. Missing telemetry is unknown, not healthy. |
Triage path
Move from symptom to the owning record
- Confirm the workspace, project, selected time window, and customer-visible symptom before comparing signals.
- Check application or API reachability only to establish whether signed-in investigation can proceed.
- Open Workflows > Runs for durable run and stage state. Use Workflows > Observability for aggregate reliability, evidence coverage, and measurement gaps.
- Use Agents or Fleets for identity and heartbeat context without equating liveness with progress.
- Use Infrastructure for redacted runtime-route readiness. Check Projects > Repositories or Integrations separately for repository readiness.
- Use Metrics and Logs within the correct scope and time window. Treat absent or stale data as an investigation gap.
- Use Control for scoped intervention only after identifying the owning record and failing layer.
Continue with Workflows, Runtime operations, the Product surfaces map, or Troubleshooting for the next safe check.
Evidence safety
Share the minimum evidence needed to diagnose
- Prefer aggregate measurements and synthetic placeholders.
- Redact secrets and sensitive values before attaching evidence.
- Do not include email addresses, participant identities, customer names, credentials, or other PII.
- Do not share raw credentials, private addresses, deployment topology, or operator-only procedures.
- Do not paste raw logs if they can contain identities or sensitive payloads. Summarize the observed state instead.
- Escalate with timestamps and the relevant workspace, project, and run identifiers, using redacted evidence only.
Displayed objectives and measured product health are observational signals. They do not create a contractual SLA, SLO, RPO, or RTO.