Dependency Health-Sweep Prompt: Is It Us or Upstream?

You are an SRE triaging a failing service. The fastest MTTR win right now is answering one question correctly: is this our code, or a dependency? Design and walk me through a rapid, read-only health sweep of everything this service depends on and everything that depends on it.

Give me:
- The failing service and symptom: [SERVICE, ERROR/LATENCY PATTERN]
- Its dependencies: [DATABASES, CACHES, QUEUES, INTERNAL SERVICES, THIRD-PARTY APIs]
- Its dependents: [WHO CALLS THIS SERVICE]
- Available signals: [METRICS, DASHBOARDS, STATUS PAGES YOU CAN CHECK]

Work through this:

1. **List the dependency graph tersely.** Upstream (what this service needs) and downstream (what needs it), so the sweep is complete and nothing is forgotten under pressure.

2. **Order the sweep by likelihood and cheapness.** Prioritize the dependencies most likely to be at fault and fastest to check first (a shared database or a flaky third-party API before a rock-solid internal service). Justify the order briefly.

3. **Give a read-only check per dependency.** For each, the specific read-only signal or command that reveals its health from this service's perspective — connection pool saturation, upstream error rate, third-party status plus your own call latency to it, queue depth. Prefer signals that reflect YOUR calls, not just the dependency's global status.

4. **Interpret the pattern.** State what each result would tell you: "if the DB connection pool is exhausted, the fault is likely ours (leak) or the DB's (slow queries) — here's how to tell them apart."

5. **Reach a verdict.** After the sweep logic, state the most probable location of the fault (our code / a specific dependency / a dependent feeding back) with confidence, and the single check that best confirms it.

Output format: a "DEPENDENCY SWEEP" — a table of DEPENDENCY, DIRECTION, READ-ONLY CHECK, WHAT-A-BAD-RESULT-MEANS, ordered by check priority — then a VERDICT (+confidence) and NEXT CONFIRMATION. Every check must be non-mutating. Rank the fault-location hypotheses explicitly.