distributed-failure-cataloguelisted

# Distributed Failure Catalogue ## Purpose Turn a symptom into a name, and a name into the skill that owns the fix. Distributed failures recur in a small number of shapes; an engineer who can name the shape reaches the right mechanism in minutes instead of rediscovering it during an incident. This skill deliberately **does not teach the fixes**. Every entry ends at an owner, because a catalogue that also explained the remedies would drift out of step with the skills that own them. Destructive cleanup is the explicit exception, with bounded guard guidance below. Its value is recognition and routing — and, for a design review, a list of concrete failures to argue a design against rather than a general appeal to robustness. ## Workflow This is a protocol/operations catalogue, with no Java language minimum. The scheduled-task reference uses the Java 25 API documentation for a long-standing contract. Inspect the target runtime, scheduler/framework, client retry defaults and deployment configuration before applying version-sensitive claims; this skill does not authorize upgrades or fault injection. 1. **Write down the observation, not the theory.** "Inbound rate rose while success rate fell", "the queue is empty and no alert fired", "duplicates 1.5 s apart". The index below is keyed on observations. 2. **Build a shared timeline and denominator.** Record clock domains, timestamp uncertainty and sampling gaps; do not infer causal ordering from cross-host timestamps alo