Robinhood Chain feed latency
FirstLight's lead per block over a typical provider and, until 2026-10-29, over the fastest private provider. One host receives every feed and one clock stamps every arrival. The record below shows the figures of every day.
The last hour, live
The lead of each block is the comparator's arrival minus FirstLight's arrival. A positive lead means that FirstLight delivered the block first. The figures cover the last hour of blocks.
| Comparator | Blocks compared | FirstLight first | Median | Mean | p90 | p99 | p99.9 | Max |
|---|---|---|---|---|---|---|---|---|
| A typical provider | 35,038 | 100.00 % | +47.6 ms | +52.1 ms | +63.4 ms | +141.5 ms | +363.5 ms | +633.4 ms |
A comparator with fewer than 20 compared blocks in the hour shows "measuring", not a figure.
The continuous record
What the record holds
The service records the lead of every block over the typical provider, and over the fastest private provider until its feed ended on 2026-10-29, on one host with one clock. It finalises each UTC day 10 minutes after the day ends. The tables below show the figures of each day. Coverage is the share of the day's sequence numbers that the record holds, from the first to the last recorded block.
The record starts on 2026-10-26. It holds 6,446,599 blocks: 8 finalised days and the days not yet finalised. Quantiles are per day; they cannot be combined across days.
Since 2026-10-26
| Comparator | Compared | FirstLight first | Mean | Max | Missed | Disagreed |
|---|---|---|---|---|---|---|
| The fastest private provider | 2,678,046 | 99.98 % | +2.7 ms | +878.7 ms | 552 | 0 |
| A typical provider | 6,444,910 | 99.97 % | +50.5 ms | +1.72 s | 1,689 | 0 |
Today so far (UTC)
2026-11-03: 335,879 blocks, coverage 100.00 %.
| Comparator | Compared | FirstLight first | Mean | Median | p90 | p99 | p99.9 | Max | Missed | Disagreed |
|---|---|---|---|---|---|---|---|---|---|---|
| A typical provider | 335,879 | 99.99 % | +50.1 ms | +46.7 ms | +61.2 ms | +104.0 ms | +314.0 ms | +633.4 ms | 0 | 0 |
The last 30 days
The fastest private provider
| Day (UTC) | Compared | FirstLight first | Mean | Median | p90 | p99 | p99.9 | Max | Missed | Disagreed | Coverage |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026-10-29 | 650,654 | 99.98 % | +2.4 ms | +2.0 ms | +4.1 ms | +9.7 ms | +15.8 ms | +878.7 ms | 10 | 0 | 99.99 % |
| 2026-10-28 | 831,523 | 99.99 % | +2.8 ms | +2.5 ms | +4.2 ms | +9.5 ms | +18.0 ms | +824.6 ms | 53 | 0 | 99.98 % |
| 2026-10-27 | 826,990 | 99.97 % | +2.8 ms | +2.7 ms | +3.9 ms | +6.0 ms | +16.6 ms | +537.0 ms | 489 | 0 | 99.99 % |
| 2026-10-26 | 368,879 | 99.96 % | +2.9 ms | +2.9 ms | +3.7 ms | +4.6 ms | +10.8 ms | +30.3 ms | 0 | 0 | 100.00 % |
A typical provider
| Day (UTC) | Compared | FirstLight first | Mean | Median | p90 | p99 | p99.9 | Max | Missed | Disagreed | Coverage |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026-11-02 | 836,301 | 99.99 % | +51.4 ms | +47.4 ms | +63.1 ms | +113.9 ms | +396.7 ms | +1.72 s | 892 | 0 | 99.67 % |
| 2026-11-01 | 830,287 | 99.99 % | +48.6 ms | +45.7 ms | +59.1 ms | +92.8 ms | +276.1 ms | +1.57 s | 0 | 0 | 99.99 % |
| 2026-10-31 | 823,278 | 99.98 % | +48.0 ms | +45.0 ms | +58.2 ms | +93.5 ms | +271.6 ms | +494.2 ms | 0 | 0 | 99.99 % |
| 2026-10-30 | 765,849 | 99.95 % | +50.1 ms | +45.8 ms | +62.5 ms | +135.0 ms | +414.7 ms | +1.03 s | 605 | 0 | 93.57 % |
| 2026-10-29 | 825,228 | 99.97 % | +51.0 ms | +45.2 ms | +64.2 ms | +155.6 ms | +496.5 ms | +1.14 s | 78 | 0 | 99.99 % |
| 2026-10-28 | 831,497 | 99.98 % | +50.2 ms | +44.9 ms | +62.1 ms | +149.3 ms | +451.6 ms | +1.66 s | 79 | 0 | 99.98 % |
| 2026-10-27 | 827,464 | 99.96 % | +52.9 ms | +48.5 ms | +64.6 ms | +140.1 ms | +410.8 ms | +816.4 ms | 15 | 0 | 99.99 % |
| 2026-10-26 | 368,859 | 99.98 % | +54.4 ms | +49.5 ms | +66.0 ms | +138.1 ms | +494.5 ms | +1.15 s | 20 | 0 | 100.00 % |
The method in short
- A subscriber process on one host receives FirstLight and the typical provider (and, until 2026-10-29, the fastest private provider). One clock stamps each arrival, so the difference between two stamps is exact.
- FirstLight's arrival is the read that completes the block's Header record on the
/headersstream. A comparator's arrival is the read that completes the block from that comparator. - Blocks are matched by sequence number. A block is compared when both sides delivered it and the block hashes agree. A block whose hashes differ is not compared. It is counted as disagreed.
- A block that a comparator did not deliver within 30 seconds is counted as missed.
- Each quantile is the lead at rank ceil(p × n) of the compared blocks. Every lead on this page is rounded down, so no lead shows larger than measured.
The full method is in the docs, under How Robinhood Chain feed latency is measured.
Questions
Who are the two comparators?
The fastest private provider is the fastest private sequencer stream that we measured, until its feed ended on 2026-10-29. A typical provider is a node provider's newHeads subscription, which comes after the block has executed.
Why are some blocks missed or disagreed?
A block is missed when the comparator did not deliver it within 30 seconds. A block is disagreed when the two block hashes differ. Neither kind of block has a lead, and the record counts both.
Can I measure the lead myself?
Yes. Ask for a free key, connect to the /headers stream and to your feed from one host, and subtract the two stamps of each block. The steps are in the measurement docs.