4.65%

At the top end, well-run validators on good hardware perform almost identically. At the bottom end are validators that rarely include user transactions, build slow blocks, cheat on commission or can't stay online. Sorting by APY alone hides all of this, so the Compass Score looks at how each validator actually serves the network.

What we measure

Most metrics cover the last 11 epochs, about 12 days. That smooths out luck for smaller validators, which get fewer blocks to produce. Each validator gets a percentage of the best value for each metric, weighted as follows:

• Block time (31%): how quickly its blocks are built this epoch, compared with the median validator running the same client family (Agave-based or Firedancer). Faster than median earns a small bonus. Validators without enough measured blocks get full marks here rather than a penalty.
• Skip rate (19%): the share of its leader slots where it produced no block.
• Compute units per block (19%): how much work its blocks carry.
• User transactions per block (19%): non-vote transactions it includes for real users.
• Uptime (12.5%): how much of the time it has been online and voting.

For compute units and user transactions, the benchmark is the 16th-best validator, so a few outliers don't set the bar. Validators with only a few blocks are pulled toward the network median, so a handful of lucky or unlucky blocks can't swing their score. Since Alpenglow, vote credits and root distance no longer say anything about performance, so they are no longer scored.

The score in the table is a percentage of the top validator's score, so 100% means joint best.

Why we mark down the largest validators

Concentrated stake is bad for decentralisation and network health. The superminority is the small group of validators that together hold a third of all stake. Since the Alpenglow upgrade, a group holding just over 40% of stake could halt Solana, so every extra share these few validators gain brings that line closer. Spreading stake across more, smaller validators makes the network harder to stop or censor.

So we cut the score of every superminority validator by 62.5%, however well it performs. If you're choosing between two good validators, pick the smaller one.

Other penalties

Penalties multiply together, so a validator with two issues loses more than one with a single issue.

• Score set to zero: commission above 10% (including 100% private validators), offline right now, a record of raising commission to take stakers' rewards, or a warning from us.
• Cut by 75%: block times slower than 98% of validators.
• Cut by 50%: flagged as a slow voter.
• Cut by 44%: no published name, website or details. Partial details cost 20%.
• Cut by 25%: commission above 7%, or outdated validator software.

0% commission is not penalised. Data-centre concentration isn't part of the score today, so check the location column yourself.

The scoring is a work in progress, and we welcome feedback.