Attestation committees still count validators, not stake

Large-balance validators have a split personality now.

The same >=1024 ETH cohort is 25.64% of active effective stake, but only 0.646% of ordinary attestation committee positions.

I got here by following the obvious follow-up to the sync committee post. Sync committees were clean: MaxEB-sized validators showed up close to their stake share, not their raw index count. Fine. That matches the spec.

Then I asked the dumber question: does the same thing happen in ordinary beacon committees, the ones validators use for attestations every epoch?

No. It disappears.

On June 20 UTC, the active set had:

• 877,423 plain 32 ETH validators, 70.50% of active effective stake
• 5,755 validators at >=1024 ETH, 25.64% of active effective stake
• only 0.65% of active validator indices in that >=1024 ETH bucket

Ordinary attestation committees followed the index count almost exactly. The large-balance cohort got 1,291,332 of 199,908,043 matched committee assignments: 0.646%.

Here is the query for the weird part:

WITH positions AS (
  SELECT arrayJoin(validators) AS validator_index
  FROM mainnet.int_beacon_committee_head FINAL
  WHERE slot_start_date_time >= toDateTime('2026-06-20 00:00:00')
    AND slot_start_date_time <  toDateTime('2026-06-21 00:00:00')
), bal AS (
  SELECT
    validator_index,
    effective_balance,
    multiIf(
      effective_balance >= 1024000000000, '1024-2048 ETH',
      effective_balance >=  256000000000, '256-1023 ETH',
      effective_balance >    32000000000, '33-255 ETH',
      '32 ETH'
    ) AS bucket
  FROM mainnet.fct_validator_balance_daily FINAL
  WHERE day_start_date = toDate('2026-06-20')
    AND startsWith(status, 'active')
)
SELECT
  bucket,
  count() AS committee_positions,
  uniqExact(positions.validator_index) AS unique_validators,
  round(100 * count() / sum(count()) OVER (), 4) AS position_share_pct
FROM positions
INNER JOIN bal USING validator_index
GROUP BY bucket
ORDER BY min(effective_balance);

Result:

Bucket	Committee positions	Position share
32 ETH	197,420,064	98.7554%
33-255 ETH	766,621	0.3835%
256-1023 ETH	430,026	0.2151%
1024-2048 ETH	1,291,332	0.6460%

This is not a small-sample wobble. Over the seven complete days from June 14 to June 20, the >=1024 ETH bucket sat between 0.634% and 0.646% of ordinary committee positions while its effective-stake share sat between 25.27% and 25.64%.

The contrast is the point.

Block proposer duties over May 22-June 20 landed at 24.82% for the same bucket. Sync committee slots landed at 24.46%. Both are close to stake share. Ordinary attestation committee positions are not even trying to be close.

That sounds strange until you look at the spec paths. get_beacon_committee passes get_active_validator_indices(...) into compute_committee(...), which shuffles indices. There is no effective-balance weighting in that committee assignment path.

Proposer selection and sync committee selection take a different path. After Electra, both use an effective-balance acceptance test against MAX_EFFECTIVE_BALANCE_ELECTRA. Balance is literally part of the random selection test there.

So Ethereum now has two surfaces that look similar from a distance but are not doing the same thing:

• ordinary attestation committee positions: index shuffle
• block proposer and sync committee selection: effective-balance weighted

Important caveat: I am counting committee positions, not final attestation weight or reward accounting. A 2,048 ETH validator is not magically reduced to a 32 ETH validator when rewards and consensus balances are computed. The odd part is the assignment surface itself: one validator index gets one ordinary attestation committee assignment per epoch, whether that index carries 32 ETH or 2,048 ETH.

That makes validator-count dashboards even more slippery after MaxEB. Sometimes count is the right denominator. Sometimes stake is. Sometimes the protocol quietly uses both, depending on which lottery you are looking at.

I don't love that. It is easy to say "validators" and accidentally mean three different things.

For June 20, at least, the distinction is no longer theoretical. A quarter of active stake was almost invisible in the ordinary committee position count.

Raw-table cross-check: canonical_beacon_committee and mainnet.int_beacon_committee_head matched exactly for June 20 at 460,800 committee rows and 199,918,709 validator-index assignments. The balance join matched 199,908,043 of those assignments; 10,666 unmatched positions were dropped, less than 0.006% of the sample.