SEA latency methodology

Use this methodology to measure SeaLink latency consistently from Southeast Asia. Repeatable p50 and p95 results are more useful than quoting one fastest request.

Measurement scope

Measure from a Southeast Asia client to the SeaLink public endpoint used by your application.
Health checks confirm service reachability only; measure model latency with real requests to the selected model.
For streaming, record both time to first byte and total completion time. User perception usually follows the first byte.

Metric definitions

Metric	Meaning
connect_ms	Time to establish TCP/TLS. This can be near zero when a connection is reused.
ttfb_ms	Time from request start to first byte. It shows how long users wait before the response begins.
total_ms	Time from request start until the response body is fully read. This is the primary non-streaming metric.
p50 / p95	Median and 95th percentile across repeated samples. Production decisions should use p95, not a single fastest sample.

Minimal repeatable check

Use the public health endpoint to confirm reachability, then send a low-cost model request with your API Key. Do not treat /api/health latency as model-generation latency.

curl

BASE="https://api.sealink.io"
curl -sS -o /dev/null -w 'connect=%{time_connect} ttfb=%{time_starttransfer} total=%{time_total}\n' \
  "$BASE/api/health"

Reporting rules

Reports must include region, URL, endpoint, model, streaming mode, sample count, and time window.
Report at least p50 and p95. A single request is only a spot check.
Call out model-service degradation, rate limits, or task queues separately from the measured end-to-end latency.
Use /api/health for public availability checks. Restricted diagnostic endpoints are not part of customer latency measurements.