Skip to content
chainkit
§ scoreboard last 24 hours · production traffic live · polling every 30s

Real production traffic. Real provider performance.

Every datapoint is a real chainkit SDK call we made ourselves through the public tools page. No synthetic load, no cherry-picked windows. If a provider has a bad hour, that bad hour shows up here.

loading
§ 01 provider scoreboard
Loading…
requests per minute
§ 02 traffic · 24h
fig. 01 · stacked by provider 0 providers · 0 bucket-rows
weighted by request count
§ 03 p95 latency · capability × provider
fig. 02 · capability heatmap empty cells = not called this window
Loading…
how we compute these numbers
§ 04 methodology
01

Source

Every row in events_raw behind this page is a real SDK call we made ourselves through the public tools. No synthetic load, no shadow traffic.

02

Strategy

The SDK runs the same provider rotation a chainkit customer would. Currently a single mempool.space leg for read operations + the local Metal validator for address checks — more providers join the rotation as their credentials are wired in.

03

Aggregation

p50 / p95 / p99 are exact percentiles computed by TimescaleDB over 1-minute buckets, averaged across the displayed window. Success rate = count(ok=true) ÷ count(*) per bucket.

04

No cherry-picking

There's no filter in our SQL that omits failing windows. If a provider has a bad hour, that hour shows up — and the score arrow points down.