Validators

All-time is the share of every graded epoch the program marked Bonus or Baseline. Combinedis the mean of the two clusters, weighted equally, and is this site’s own figure: the program publishes no combined score. Epochs the program never assessed are gaps: they count on neither side of the division, so a validator with 200 graded epochs, 20 failures and 300 gaps reads 90%, not 39%. Bars run 0% to 100%; the percentage is always printed beside them.

The two clusters are judged against different published floors, so the columns are not a like-for-like comparison. Current thresholds for both are on the criteria page. Grades are the program’s own, counted here and never re-derived; a validator with no published grade for a cluster shows an em dash and sorts last under either direction, because an absence is not a low score.

The timing column and the IBRLscore are the figures here that are not the program’s own, and neither is a program threshold or has any bearing on a validator’s grades. The timing column takes one of two sources, and they are not the same measurement. Build time is the median time the leader took to build a block over a single epoch, from IBRL, and is the default. Slot time is the mean wall-clock duration of a slot over the last 10 epochs, from Trillium. Across the 437 validators carrying both they correlate at 0.57: the medians sit 7ms apart and the ranges overlap, yet half the validators differ by more than 15ms and one by 78ms. Switching the source therefore changes what is being measured, which is why the column header and its attribution change with it. Coverage differs too, and is not interchangeable: Trillium covers 449 validators on this roster, IBRL 441, so a validator can carry a figure in one mode and an em dash in the other. The per-validator page carries the same toggle, and its ten-epoch strip reads either source: IBRL’s endpoint answers per epoch, so that strip covers 463 validators against the 437 this column counts for one epoch.

Both modes are banded at the same 410 and 425ms, and the cut points are attributed differently in each. In slot-time mode they are Trillium’s own: 410ms is exactly where their slot_duration_is_lagging flag flips. In build-time mode they are this site’s reading, and they belong to neither vendor, because IBRL publishes no band for build time and Trillium never measured build time at all. The same marks are reused because the two distributions sit in the same range across the Approved set, 353 to 458ms of build time against 347 to 477ms of slot time, so a reader switching source does not have the scale move under them. The colour is blunter that way, and this is the cost rather than a hidden one: of the 359 Approved validators carrying a build time, 52% land at or under 410ms, 47% between 411 and 425ms and 2% above, where slot time spreads across all three bands. Neither set of cut points is a program threshold, neither affects a validator’s grades, and the millisecond value is printed beside the colour in both modes.

The score out of 100 is published by IBRL for epoch 1017, covering 441validators on this roster, and is stored exactly as published: this site derives no score, rank or composite of its own from it. The score and both timing sources measure mainnet block production, so where a validator funded on testnet only has no figure the cell reads “not on mainnet” rather than an em dash. A few of them do lead mainnet blocks, and those show the measured number.

Grade history covers the 513 validators the program has funded, including 36 in this view marked Departed. A departed record is frozen at the moment this site observed the removal; the program publishes no exit date and no reason, so the observation date is evidence of when, not a decision date.

Joined is the join epoch the program records. Many validators were graded before that epoch; where that is so the row says so, because no earlier join date is published.

Display names come from the program’s validator list, which carries them for 4,982 pubkeys, and a logo for 2,716. Unnamed records show a truncated pubkey and sort after named ones under the name sort. The logo is the operator’s own image as the program publishes it, shown beside the name to make a row identifiable at a glance; it is decoration and nothing is sorted, filtered or graded by it. Where a validator has no logo the space stays empty rather than showing a stand-in, because an absent logo is not missing data. Everything here is read from files committed to this repository; nothing is fetched while the page renders.