How FirstLight measures feed latency

Every figure on the FirstLight home page is a per-block measurement of Robinhood Chain feed latency. FirstLight and each comparator are received on one host and stamped by one clock, so the difference between two stamps is exact.

The principle

For each Robinhood Chain block, a subscriber process on one host records when FirstLight delivered the block and when each comparator delivered the same block. Both times come from the same clock. The lead for that block is the difference between the two times. The home page shows these per-block leads, their distribution and their quantiles.

No figure on the home page is a model or an estimate. A comparator without measured blocks shows "measuring", not a number.

Definitions

Arrival: For a comparator, a block arrives when a subscriber process has read the block's complete sequencer message from the socket. For FirstLight, a block arrives when a subscriber process has read the block's complete Header record from the /headers stream. FirstLight makes that record when the block's sequencer message is complete.
Lead: For each block, the comparator's arrival minus FirstLight's arrival. A positive lead means that FirstLight wins the block. Zero means that the two arrived together. A negative lead means that the comparator won the block.
Matching: Blocks are matched by the feed's sequence number. A pair whose block hashes disagree is dropped and counted.
Share of blocks: The percentage of matched blocks that FirstLight wins.

Comparators

Fastest private provider: the fastest private sequencer stream that we measured, received live until its feed ended on 2026-10-29. Its figures are now a fixed benchmark from that measurement.
Typical provider: a node provider's newHeads subscription. This is a header notification that the provider publishes after the block has executed. A lead over it is a lead over a node's execution, not over the sequencer.

The figures

Windows

The home page shows each live comparator over its last ten minutes, from at least 20 paired blocks. With fewer, it shows the comparator's fixed record. A fixed record states its period beside its figures. The fixed records were measured before 2026-10-17, and they do not take FirstLight's arrival from the Header record that the live figures use. The record of the fastest private provider (2026-10-16, 05:00 to 13:00 UTC) takes it from the complete block on the feed (/), as a subscriber on the measuring host received it. The record of the typical provider (2026-09-21 22:36 to 2026-09-22 11:04 UTC) takes it from the first of FirstLight's own input connections to deliver the block, on an earlier measuring host. Both points come before the Header record.

The continuous record keeps every block's lead over the typical provider from the Header record, and over the fastest private provider until 2026-10-29. The page shows the figures of each UTC day.

Statistics

For each comparator the home page gives the median, the mean, p90, p99, p99.9 and the maximum lead, the share of blocks that FirstLight wins and a histogram of the leads. The distributions are heavy-tailed, so the quantiles carry more information than the mean.

Units and rounding

The stamps are integer nanoseconds. The home page shows a lead in milliseconds with one decimal below one second, in seconds with two decimals from one second, and in whole seconds from 100 s. The board's medians show hundredths of a millisecond. Every lead is rounded down, so a lead is never shown larger than measured. A share is a percentage with two decimals, rounded down.

The chart

For each block, the chart draws FirstLight's lead over each live comparator, in that comparator's colour, with its running median. A block that a comparator won is not drawn. The figures under the chart count it.

What this is not

FirstLight does not change how the sequencer orders or includes transactions. It changes when you learn what the sequencer did.
The figures are measured on one host. Your lead depends on where your client runs and on your network path. Measure it from your own host.

Measure it yourself

Ask for a free key on Telegram.
Connect to the /headers stream and to the feed that you use today, from the same host.
Stamp each arrival with one clock. Match the blocks by block hash: every Header record and every feed carries it.
Subtract the two stamps for each block, and compare the quantiles, not only the mean.

On /headers, a block's Lead arrives at about the time of the block's first raw transaction on the transactions stream (Ultra), and its Footer within about 0.1 ms of the complete block on the feed (Pro). A block can have no Lead. The live figures on the home page use this stream's Header record, so a free key lets you repeat the measurement from your own host. See the plans.