Workload: High-volume at-least-once order pipeline

[dexcompiler]

Epic: #58

Type: Scenario | Signal area: Correctness, Reliability, Performance | Runtime mode: Both

Deliver

Scale the existing outbox/inbox + idempotency + HLC/vector-clock demo into a large order-processing workload with drops, duplicates, reordering, retries, and delayed acknowledgements.

Signals to capture

Duplicate suppression failures, causal-order violations, retry storms, backlog growth, throughput collapse.

Acceptance criteria

Scenario runs in simulated and system time modes; simulated mode yields stable, deterministic results for the same seed and parameters.

[liuyingxuvka]

Your acceptance criteria mentions simulated mode determinism (same seed/params) and catching duplicate suppression failures, retry storms, and backlog growth. I built Flow-Guard after getting bitten by at-least-once workflows where subtle state transitions + retries produce duplicate side effects.

Flow-Guard is a free/open-source workflow/logic/architecture simulator that uses an explicit finite-state mathematical model (not a prompt-only wrapper). It can help you model the outbox/inbox + idempotency rules + retry/backoff and then explore traces/counterexamples.

Repo: https://github.com/liuyingxuvka/FlowGuard
Feedback thread: liuyingxuvka/FlowGuard#1

If you try it, I'd love feedback on whether the modeling primitives fit this kind of workload, and what feels missing/confusing.