Measurements
Block propagation
Every provider is subscribed at the same moment from one process. When each announces the same block or slot, the arrival is stamped by the same clock before any parsing. What is published is how far behind the earliest announcement each provider was, for that specific height.
BSC over WebSocket
BSC over WebSocket4,964,273 blocks · 2 providers · 4 forked
eth_subscribe newHeads — fires when the node accepts a new head.
| # | Provider | Median behind | p95 |
|---|---|---|---|
| 2 (2nd) | NodeReal | 305 ms | 390 ms |
| 1 (1st) | PublicNode | 0 ms | 0 ms |
4 heights were announced under more than one hash in this window. Those providers were on different forks at that moment, not merely slower.
Ethereum over WebSocket
Ethereum over WebSocket171,602 blocks · 3 providers
eth_subscribe newHeads — fires when the node accepts a new head.
| # | Provider | Median behind | p95 |
|---|---|---|---|
| 1 (1st) | PublicNode | 0 ms | 13 ms |
| 2 (2nd) | dRPC | 107 ms | 348 ms |
| 3 (3rd) | NodeReal | 285 ms | 391 ms |
Solana over WebSocket
Solana over WebSocket2,813,532 slots · 2 providers
slotSubscribe — fires when the node's bank forks a new slot.
| # | Provider | Median behind | p95 |
|---|---|---|---|
| 1 (1st) | Solana Foundation | 0 ms | 221 ms |
| 2 (2nd) | PublicNode | 4.5 ms | 225 ms |
What this does not measure
Each of these was tried and rejected on evidence, not on principle.
- Absolute time from subscribing to a first notification. It is dominated by where in the block interval the subscription landed — up to a full block of noise, measured at 10.2 seconds of spread on Ethereum between trials whose subscribe round trip differed by 13ms.
- Counts of which provider won most often. Two consecutive runs over the same providers scored 5-1 and then 7-0, because one block on which everyone stalled flips the headline. Lead share is shown alongside the distribution, never instead of it.
- Comparisons across transports. WebSocket slotSubscribe fires when a node's bank forks a slot; Yellowstone SLOT_PROCESSED fires once that slot has been executed, a median 575ms later. They are different events, so they are ranked separately.
- The first heights after a subscription opens. A node replays recent slots to a new subscriber, and that backlog drain is not a race: on Solana the median delta over the first 25 slots was 2,205ms against 72ms over the last 25 of the same session. Those rows are recorded and excluded from every figure here.
30d window · generated 2026-10-08 18:07:05 UTC · 3 cohorts