184,000 events per second, audited.

The throughput problem

Karavel operates a real-time tracking and customs reconciliation pipeline that ingests sensor and EDI data from over two hundred carriers, ports, and brokers. The peak production volume runs at 184,000 events per second.

The previous runtime was fast enough but had no usable lineage. Customs disputes routinely required forensic reconstruction by a small, exhausted team. A sustained shift toward provable, auditable supply chains made that posture untenable.

What we did

We deployed Anillion's orchestration and observability primitives into the existing data path. The work emphasised three properties:

1. 01Throughput. The runtime had to sustain peak load without downstream backpressure.
2. 02Lineage. Every event had to carry an unbroken chain of custody from ingestion to invoice.
3. 03Replay. Any window of traffic had to be re-runnable against a new policy or schema with full fidelity.

The deployment was incremental. We migrated one carrier integration at a time, in shadow, then in primary, with rollback windows that were never used.

What changed

• +Sustained throughput now reaches 184,000 events per second at p99.
• +Reliability against business SLOs has held at 99.998% for fourteen months.
• +Customs disputes are resolved against a queryable lineage, not a forensic project.
• +A previously six-person team now operates the runtime with three engineers.

The provable part

The Karavel pipeline is one of the largest provably-audited event streams we operate. The runtime gives the same answer to the regulator, the customer, and the carrier — every time, with a receipt.