Methodology
What every number on the site means, where it comes from, and its limits.
Each validator gets a scheduler and a client for each day from the Firedancer validator reports and Trillium. Fees, tips and transaction order come from Dune. Block fill and block time come from Ghost shred data and Trillium.
Adoption
Stake and stake share
The active stake of a scheduler's validators on the day, as a share of all active stake.
sum of active_stake over the validators of the group / sum over every validator with an identity row
Stake is the active_stake field of the Firedancer validator report for the day: the stake the validator carries while it votes. The report also has an epoch_stake field, the stake fixed at the start of the epoch; reports.firedancer.io shows that one in its tables. The two differ by the stake that moved in the last epoch, and the active total leaves out validators that were delinquent on the day. So a share here is a few points larger than the same share on reports.firedancer.io (2026-11-03: 38.6 % against 36.9 % for Agave JitoBAM). Validators without a scheduler label count as Unknown and stay in the denominator.
Revenue
Block revenue
The value a leader realizes from its blocks.
base fees after the burn + priority fees + Jito tips + Rakurai TIN + other tips
Base fees count at the validator's half: the other half is burned. It is what the blocks earn, not what the validator keeps after its commission to stakers.
Revenue channels
The parts we split revenue into.
base fees · priority fees · Jito tips · Rakurai TIN · other tips
Every channel is net: the part the validator receives. Base fees are the validator's half (the other half is burned). Jito tips are 94 % of each tip, because Jito keeps 6 %. Rakurai TIN is 80 % of the Rakurai income of the block, because Rakurai keeps 20 % (see below). Other tips are the full tips to the other tip accounts we track. Priority fees do not include Rakurai payout transactions.
Rakurai TIN
The validator's part of all Rakurai income: tips, backrun share and subscriptions.
80 % × Rakurai tips when the block lands, for each block; + 80 % × (backrun share + subscriptions of each epoch) × the part of the epoch in the period, for each period
Rakurai TIN is net: the part the validator receives, like Jito tips. Rakurai keeps 20 % of each tip, backrun share and subscription (Rakurai's on-chain configuration, read on 2026-10-13), so we count 80 %. We count the Rakurai tips of each block when the block lands. Rakurai's on-chain ledgers give the backrun share and the subscriptions of each validator for each epoch. These two amounts are not paid in a block, so no block holds them. We divide the amount of each epoch over its UTC days by the part of the epoch's time in each day. The totals and the means of a period (the composition, the history, the weeks and the Validators table) add the days of the period. The distributions of single blocks and revenue per second show only the Rakurai tips that a block holds. Rakurai claims these amounts for the validator, usually in the next epoch. Then the validator pays its part to itself as the priority fee of a payout transaction in its own next block. We remove these payouts from priority fees, because Rakurai TIN already counts the same money. From 2026-10-13, a daily query finds the payout transactions. Before that date, we take each payout from Rakurai's claim transaction on Solana: the amount is exact. We put it in the validator's first block after the claim, or in a later block when the payout is larger than the fees of that first block. On a day without payout data, priority fees keep these payouts, and the coverage note says so.
Revenue per second
What a block earns for each second of slot time, so longer slots do not look like more revenue.
block revenue / leader's mean slot duration that day (s)
The slot duration is the leader's mean Ghost slot duration on the day of the block: first shred of a slot to the first shred of the next slot. A block has no value per second when its leader has no Ghost slot duration that day, so the history starts on 2026-02-11. The directory shows the mean over the leader's blocks.
Ordering
Footrule distance
How far a block's order is from a perfect sort. 0 is sorted, 1 is reversed, about 0.64 is random.
Σ |landed position − ideal position|, normalized to [0, 1]
We use Spearman's footrule. 0.667 is the theoretical random for independent and complete rankings; our data is finite and not fully independent, so the empirical random baseline is closer to 0.64. The ideal order comes from a ranking key (below); the default is (priority fees + all tips) / CU. Blocks need at least 5 non-vote transactions. From 2026-10-13, a block with a Rakurai payout transaction is not in the footrule or VOR: the payout is a large priority fee that the validator pays to itself, not a market order. Older payout blocks stay in: fewer than 1 in 1,000 Rakurai blocks. The method follows the initial work by @MostlyData_.
Value Ordering Ratio (VOR)
How much of the best value-first order a block achieves. 1 is the best order, about 0.5 is random.
area under the block's value curve ÷ area under the best order's curve
Walk through the block in landed order and plot the share of value collected against the share of compute used. The best order sorts the same transactions by value / CU. Each VOR names its value. VOR of block revenue (vor_block_revenue) uses 50 % of the base fee (after the burn) + priority fees + Jito tips + other tracked tips. VOR of priority fees + all tips (vor) uses the same value without base fees. The Ordering chart shows block revenue by default. Block revenue data starts on 2026-09-08. Before that date, the chart shows vor. The Validators directory uses vor. Unlike the footrule, a misplaced large payment costs more than a misplaced small one. Landed transactions only. Blocks with a Rakurai payout transaction are not in VOR.
Block fill
Block fill
How fast each part of a block lands, from its first to its last shred.
share landed by each point of the block, averaged over blocks
Parts: non-vote transactions, vote transactions, compute units and so on. The x-axis is either % of the block's own duration (50 bins) or ms since the first shred (10 ms bins). Complete Ghost blocks only. The curve of each part averages only the blocks where that part is more than zero: a block with no Jito transactions does not count in the Jito curve.
Block time
Slot duration
Time between the first shreds of two consecutive blocks, against the slot target of the epoch.
first shred of the next block − first shred of this block
Counted for the earlier leader, only when the next slot follows directly. Target per epoch: 400 ms at first, then 350 ms from epoch 1020, 300 ms from epoch 1024, 250 ms from epoch 1037, 200 ms from epoch 1053 (SIMD-0525). Ghost measures each block; Trillium gives one value per validator and epoch. We never mix the two.
Shred span
Time from a block's first shred to its last shred.
last shred − first shred
No target. Ghost only.
How block time is aggregated
Charts over time use every block; box plots and the validator table use whole epochs only.
Times outside 0 to 5,000 ms are left out. Histograms use 5 ms bins. Box plots, the epoch history and the validator table use only epochs fully inside the range, each against its own target.
Oracles
Oracle update position
Where in a block an oracle update lands.
tx position ÷ tx count, or ms after the first shred
The % view needs complete transaction positions, which exist for the pilot days listed under Coverage.
Ranking keys
- Priority fees per CU: sort by priority fees / CU. The block follows the fee market, like the stock Agave and Firedancer schedulers.
- Priority fees + Jito tips per CU: sort by (priority fees + Jito tips) / CU. The order of the original work by @MostlyData_.
- Priority fees + all tips per CU: sort by (priority fees + all tips) / CU. The block earns the most fees and tips per unit of block space. The default.
- Jito tips: sort by Jito tip. The largest Jito tips go first.
- Non-Jito tips: sort by non-Jito tip. The largest Rakurai and other tracked tips go first.
- All tips: sort by all tips. The largest tips go first.
Schedulers and clients
Schedulers
- Vanilla
- The stock scheduler of the Agave, Frankendancer or Firedancer client.
- Jito
- A modified version of the client with a built-in Maximum Extractable Value (MEV) auction and block engine, to increase revenue.
- JitoBAM
- A block-building extension for the Jito client that offloads transaction sequencing to external schedulers that run in trusted execution environments.
- Rakurai
- A performance-focused fork of Agave with its own scheduling and transaction processing.
- Harmonic
- An open block-building system. The validator runs Harmonic's client (Salsa, an Agave fork, or Samba, a Firedancer fork) and picks a scheduling strategy; a remote block builder that runs that strategy builds the blocks, and the validator streams them to the network. Harmonic tips are paid as priority fees.
- Raiku
- A sidecar that runs next to the validator client and sells guaranteed inclusion in upcoming blocks, booked ahead of time (AOT) or just in time (JIT).
Clients
- Agave
- The main Solana validator client, maintained by Anza. It continues the work of the Solana Labs client and is the reference implementation for the network.
- Frankendancer
- A hybrid client and the step towards full Firedancer. It combines the consensus and runtime of Agave with Firedancer's networking, signature verification and block packing.
- Firedancer
- A validator client written in C by Jump Crypto.
Harmonic strategies
- FBA (Frequent Batch Auction)
- Harmonic collects transactions in 50 ms batches and sorts each batch by priority fees and tips, highest first. Arrival time inside a batch does not matter.
- FIFO (first in, first out)
- Harmonic orders transactions by arrival. Fees and tips do not change the order.
- MREV (maximum revenue)
- Harmonic selects transactions continuously to capture the most revenue per block. The selection logic is proprietary.
Firedancer submodes
- Revenue-optimized (BUN)
- A Firedancer or Frankendancer submode that puts the most MEV and transaction revenue first.
- Balanced (BAL)
- A Firedancer or Frankendancer submode that trades off revenue and performance.
- Performance-optimized (PRF)
- A Firedancer or Frankendancer submode that puts network performance and low latency first.
- Puffin mode (PUF)
- A Firedancer submode named after the Puffin validator, whose Pebble scheduler holds non-vote transactions for batch auctions about every 50 ms, ranked mainly by priority fees.
Classification
- Jito
- The day's Firedancer report names a Jito scheduler (for example Agave Jito).
- JitoBAM
- The day's report names JitoBAM or BAM (Jito's Block Assembly Marketplace).
- Harmonic
- The day's report names Harmonic.
- Rakurai
- The day's report names Rakurai, or Rakurai's list names the validator for that epoch.
- Raiku
- The day's report names Raiku.
- Vanilla
- The day's report names no block engine (for example Agave Vanilla). Check: if the validator earned Jito tips that day (a median of 0.0002 SOL per block or more), we use Trillium's scheduler for the day instead; if Trillium also says Vanilla or has none, the day is Unknown.
- Other
- Any other scheduler the report names.
- Unknown
- No report for the day, no scheduler in it, two reports that disagree, or a Vanilla day with Jito tips that Trillium does not confirm.
Data sources
- Firedancer validator reports
- reports.firedancer.io: the client, scheduler and stake of every validator, each day, and the self-reported validator names. The stake is the report's active_stake field, not its epoch_stake field (see Adoption).
- Trillium
- trillium.so: Firedancer submodes, a second scheduler label for checks, epoch boundaries, and slot timing per validator and epoch.
- Ghost
- Ghost's IBRL shred data: when each shred of a block arrived at one observer. It gives block timing, block fill and oracle timing. We use one complete block in 10, chosen by a hash of the slot.
- Dune
- Solana tables on dune.com: fees and tips per block, the transaction order behind the footrule and VOR, oracle updates, Rakurai payout transactions, and the daily SOL price in USD. Order and oracle data use up to 200 blocks per leader and day.
- Solana mainnet
- Harmonic's crank memos, read from the chain.
- Rakurai
- Rakurai's list of its validators per epoch, and Rakurai's revenue ledgers on Solana mainnet: the backrun share and the subscriptions of each validator for each epoch. We read the ledgers once a day.
CSV downloads
The full history of each section, grouped by scheduler and client. Each link gives the file of the newest release.
- Ordering quantiles
- Revenue breakdown per block
- Revenue distribution (per block and per second)
- Revenue per second
- Revenue over time
- Stake share
- Block anatomy totals
- Oracle update counts
- Validator directory
- Validator block time
- Revenue per group and day
- Stake and validators per group and day
- Footrule and VOR quantiles per group and day
- Block time per group and day (Ghost)
- Block time per group and epoch (Ghost and Trillium)
- Block fill curves per group
- Oracle update positions per group (% of block)
- Oracle update position per validator
Limitations
- USD values convert each day's SOL at that day's average SOL price (Dune prices.usd, wrapped SOL). Per-block distributions and the scatter stay in SOL.
- Samples: block timing and block fill use one complete Ghost block in 10; transaction order and oracle data use up to 200 blocks per leader and day. Ordering and Oracles use separate samples, so their block counts for one group can differ by a few percent.
- A group with fewer than 1,000 blocks in the range is marked as a small sample.
- Quantiles come from fixed bins, so a shown quantile is exact to its bin: 0.1% for SOL values, 0.001 for footrule and VOR, 1 ms for timing, 1 for transaction counts and 10,000 CU for compute units. Counts, means and totals are exact.
- Block time is when shreds arrived at one observer (Ghost), plus Trillium's aggregate. It shows when blocks arrived, not why.
- The Value Ordering Ratio and oracle positions as a share of the block exist only for the days listed under Coverage.
- Validator names are self-reported and nobody checks them.
Data from 2026-01-01 to 2026-11-07. The data updates each day. See the methodology for the definitions, sources and CSV downloads.