LOCAL MATRIX
15/15
validated rows across 500–10,000 requested EPS
// measured evidence, not marketing targets
SlimStream benchmark results.
A transparent view of the validated local Data Plane and end-to-end test runs completed on 21 September 2026. Every accepted result is tied to raw, checksummed artifacts.
local: 10,000 EPS requestedE2E: 2,500 EPS requested50,000 EPS: not demonstrated
E2E MATRIX
9/9
validated rows across 500–2,500 requested EPS
DLQ RECOVERY
PASS
full drain and stable zero-state verified for all E2E matrix rows
01 / local Data Plane
HTTP acceptance under load
The local level measures the benchmark HTTP ingestor, bounded queue behavior and request latency. It does not measure downstream emitter or sink delivery.
| Requested EPS | Average achieved | Rows |
|---|---|---|
| 500 | 496.646 | 3/3 |
| 1,000 | 989.985 | 3/3 |
| 2,500 | 2,459.968 | 3/3 |
| 5,000 | 4,847.743 | 3/3 |
| 10,000 | 9,512.090 | 3/3 |
One 1,000 EPS repetition recorded one transport error. No rejected events were recorded in the accepted matrix.
02 / end-to-end
Delivery, failure and recovery
The E2E level includes the router, real http_json emitter, controlled HTTP 503 sink, retry behavior, DLQ persistence, replay and full drain verification.
| Requested EPS | Average achieved | Full drain |
|---|---|---|
| 500 | 494.976 | 3/3 |
| 1,000 | 989.964 | 3/3 |
| 2,500 | 2,454.933 | 3/3 |
All nine rows had sink request progression, failure/DLQ evidence, replay_verified=true and drain_verified=true.
03 / evidence boundary
What these results do—and do not—show.
Included:
raw load-generator JSON and logs, Prometheus snapshots, sink counters, DLQ snapshots, resource samples and SHA-256 manifests.
Not claimed:
50,000 EPS capacity, absence of memory leaks, or exact routing-stage p99 without dedicated profiles and instrumentation.
Scope:
local numbers are acceptance observations; E2E numbers cover the tested local fixture and controlled sink.
Historical run:
the earlier partial-replay run remains separate and retained residual DLQ entries; it is not presented as a full-drain result.
audit trail
Inspect the methodology and raw evidence.
The public summary is intentionally concise. The repository documentation describes the acceptance contract, validator rules, drain semantics and reproducibility commands.