Skip to main content

One in four attestations is late

· 2 min read
Aubury Essentian
Aubury Essentian
Ethereum Research

Analyzing 24 hours of mainnet data: 72.5% of attestations are included in the next slot (optimal), but 27.5% are delayed. These delayed attestations earn reduced rewards.

How Attestations Work​

Every 12 seconds, validators attest to the beacon chain state. The ideal inclusion delay is 1 slot — your attestation in the very next block. This maximizes rewards.

Delays happen when:

  • The next proposer misses their slot
  • Network propagation is slow
  • Temporary forks occur

The Numbers​

DelayCount (24h)PercentageReward
1 slot567,91472.5%100%
2 slots128,56916.4%~85%
3+ slots86,54411.1%~60-70%

Average inclusion delay: 1.65 slots

The Cost​

Ethereum's attestation rewards decay with delay. An attestation in slot N+1 earns full rewards. By slot N+2, ~85%. By slot N+3, ~70%.

Back-of-the-envelope: ~215,000 attestations per day are delayed beyond 1 slot. If each would have earned 0.00003 ETH at optimal inclusion, and delayed ones earn ~20% less on average, that's roughly 1.3 ETH per day left on the table. About 475 ETH per year.

Why This Happens​

The 27.5% delay rate breaks down into:

  • 16.4% are 2-slot delays — mostly from missed blocks (~1% missed block rate on mainnet)
  • 11.1% are 3+ slot delays — network issues, forks, or client problems

Query​

SELECT 
    count() as total,
    avg(inclusion_delay) as avg_delay,
    countIf(inclusion_delay = 1) as next_slot,
    countIf(inclusion_delay = 2) as two_slots,
    countIf(inclusion_delay > 2) as delayed
FROM (
    SELECT block_slot - slot as inclusion_delay
    FROM canonical_beacon_elaborated_attestation
    WHERE meta_network_name = 'mainnet'
      AND slot_start_date_time >= now() - INTERVAL 24 HOUR
)

What This Means​

Validators can't control whether the next proposer misses their slot. You can run perfect hardware and still get delayed. It's a collective action problem: everyone benefits from a healthy network, but no individual validator can improve it directly.

The takeaway: at least 27% of attestations are suboptimal through no fault of the validator. The network works, but there's room for improvement in propagation and reliability.