The attestation aggregation_bits field contains no bits

The aggregate-and-proof table had three encodings for a real bitlist. Its ordinary-attestation sibling manages something stranger: the column is present, but the gossip object has no bitlist. In 15 minutes, 1,960,559 message IDs still split into two values because one emitter spells empty as "" and another spells it 0x.

Two spellings for nothing

This started as the obvious follow-up to the aggregate encoding mess. I ran the same content-ID gate against default.libp2p_gossipsub_beacon_attestation, using a complete 15-minute window on July 18. The table had 9,578,165 observer rows for 2,059,867 content-derived message IDs.

Grouping each ID by aggregation_bits produced 4,020,426 (message_id, field value) pairs. In other words, 95.178912% of the IDs acquired a second value even though slot, committee, attester, voted root, source checkpoint and target checkpoint agreed exactly.

WITH per_id AS (
  SELECT
    message_id,
    count() AS observation_rows,
    uniqExact(aggregation_bits) AS bit_field_values,
    uniqExact(meta_client_implementation) AS implementations,
    uniqExact(tuple(
      slot,
      attesting_validator_committee_index,
      attesting_validator_index,
      beacon_block_root,
      source_epoch,
      source_root,
      target_epoch,
      target_root
    )) AS semantic_values
  FROM default.libp2p_gossipsub_beacon_attestation FINAL
  WHERE meta_network_name = 'mainnet'
    AND slot_start_date_time >= toDateTime('2026-07-18 12:00:00')
    AND slot_start_date_time <  toDateTime('2026-07-18 12:15:00')
  GROUP BY message_id
)
SELECT
  count() AS message_ids,
  sum(observation_rows) AS observation_rows,
  sum(bit_field_values) AS message_field_pairs,
  countIf(bit_field_values > 1) AS split_ids,
  round(100 * split_ids / message_ids, 6) AS split_pct,
  countIf(semantic_values > 1) AS semantic_disagreements,
  countIf(implementations >= 3) AS ids_seen_by_three_emitters
FROM per_id;

The two values were not competing bitsets. Tysm stored the empty string, as did the Teku sidecar. The Lighthouse sidecar stored 0x. Among the 64,377 IDs captured by all three paths, Tysm and Teku agreed on "" every time while Lighthouse used 0x every time.

Four other complete 15-minute checks that day put the split share at 98.31%, 96.85%, 95.18% and 94.19%. That movement tracks which capture paths overlapped in each window; it is not an attestation trend. Every window had zero disagreements on the non-placeholder fields.

SingleAttestation has no aggregation bits

Since Electra's EIP-7549 changes, the beacon_attestation_{subnet_id} topic propagates SingleAttestation. The container has four fields: committee_index, attester_index, data and signature. aggregation_bits is not one of them.

Aggregated attestations still have a real bitlist. That is why the three representations in the previous post were a serialization problem. Ordinary gossip is different: one validator index travels directly in the message, so there is no participant bitlist to normalize.

The same fork change leaves a second fossil in the table. committee_index comes from AttestationData.index, which EIP-7549 requires to be zero. Xatu's ClickHouse flattener writes that zero as an empty string. The real committee lands in attesting_validator_committee_index.

SELECT
  meta_client_implementation AS emitter,
  if(aggregation_bits = '', '<empty string>', aggregation_bits) AS stored_value,
  count() AS rows,
  uniqExact(message_id) AS message_ids,
  countIf(committee_index = '') AS empty_legacy_data_index_rows,
  countIf(attesting_validator_committee_index = '') AS empty_actual_committee_rows,
  uniqExact(attesting_validator_committee_index) AS actual_committees,
  countIf(attesting_validator_index IS NULL) AS null_attester_rows
FROM default.libp2p_gossipsub_beacon_attestation FINAL
WHERE meta_network_name = 'mainnet'
  AND slot_start_date_time >= toDateTime('2026-07-18 12:00:00')
  AND slot_start_date_time <  toDateTime('2026-07-18 12:15:00')
GROUP BY emitter, stored_value
ORDER BY emitter;

All 9,578,165 rows had an empty legacy committee_index. None had an empty actual committee or a null attester index. Across the two broad emitter cohorts, the actual field covered all 64 attestation subnets. A query using the shorter-looking committee_index column would conclude that every ordinary attestation lacked a committee.

The raw payload has nowhere to hide a bitlist

Xatu v1.22's raw gossip payload archive held exact message ID 0000cb806a7bb8d1115919519495d88b289a758b under beacon_attestation_1. The compressed payload was 224 bytes with SHA-256 28e6a51b133fa21ac861e23bb4b3528601cec306e65b6dc04ba0fa6c22c35793.

SELECT
  message_id,
  topic_name,
  message_size,
  length(message_data) AS stored_bytes,
  hex(SHA256(message_data)) AS payload_sha256,
  hex(message_data) AS payload_hex
FROM default.libp2p_gossipsub_message_payload FINAL
WHERE meta_network_name = 'mainnet'
  AND wallclock_slot_start_date_time >= toDateTime('2026-07-18 12:00:00')
  AND wallclock_slot_start_date_time <  toDateTime('2026-07-18 12:15:00')
  AND message_id = '0000cb806a7bb8d1115919519495d88b289a758b';

Snappy decompression produced exactly 240 bytes. Decoding the fixed SSZ layout gave:

• bytes 0-7: committee index 1
• bytes 8-15: attester index 117,414
• bytes 16-143: AttestationData, including slot 14,796,039 and its required zero data index
• bytes 144-239: the 96-byte BLS signature

There is no offset left for aggregation_bits, and the decoded values match the parsed table row. This is an independent check against the adapter code rather than another grouping of the same flattened column.

The code explains both empty spellings. Tysm constructs the compatibility event with AggregationBits: "". The Teku sidecar writes an empty string for a single attestation. The Lighthouse sidecar explicitly writes 0x because single attestations have no aggregation bits.

So this is not a client disagreement, an empty committee or a participant-count bug on Ethereum. It is one compatibility schema carrying a pre-Electra field through three adapters. The sample is an instrumented gossip surface, not a validator census or a count of attestations included onchain.

There is no useful normalization trick here. For SingleAttestation rows, the honest value is absent or null, and the actual committee field needs the obvious name. Until the schema says that plainly, grouping ordinary gossip by aggregation_bits turns two ways of writing nothing into two different facts.