Download the recorded evidence
← Humraahi

ENGINEERING HUMRAAHI

One order.
One rightful claim.

A simple delivery needs careful engineering underneath. Here is what we tested, what we measured and what those results mean.

Correctness under contention.

Concurrent vendors
100
Successful claim
1
Expected conflicts
99

All requests targeted the same order through the real HTTP handler. One returned success; the other 99 returned HTTP 409. Expected conflicts are part of this test’s correctness condition.

Measured, with the context attached.

Recorded 2026-11-04 UTC. These are local closed-loop observations, not production capacity guarantees. Percentiles include the full HTTP response after dispatch, including database-pool wait, and exclude generator queue time.

All responses, including expected conflicts. Latency in milliseconds.
WorkloadIn flightRequestsSuccess / conflictp95p99
Unique order placement1200200 / 010.511.0
Independent order claims1200200 / 07.39.6
Unique order placement10200200 / 038.985.1
Independent order claims10200200 / 024.127.6
Unique order placement25200200 / 083.190.5
Independent order claims25200200 / 053.961.8
Same idempotency key50100100 / 0115.4121.9
100 vendors, one order1001001 / 99259.9268.1
One vendor, 25 orders25251 / 2442.943.7

Normal workloads use 200 observations each; p99 selects the 198th sorted observation. Race workloads have different sample counts. A single successful race winner cannot establish a useful success-tail estimate.

Test conditions and limits

Database verification recorded 647 orders and 622 claims, with 0 duplicate orders and 0 vendors holding multiple active orders.

Explore the tracking engineSource repository