mainnet.fct_data_column_availability_by_slot_blob has both blob_index and probe_count, which makes summing the latter feel safe. It is not. Across 14 complete UTC days, the source slot/column table held 63,597,380 observation units; the blob-level table summed to 335,651,250 probe_count units because it copied each column result once for every blob in the slot.
At 09:15 UTC on July 11, one Tysm observer-peer pair logged 80,749 outbound Gossipsub DropRPC events in 60 seconds. PRUNE controls appeared in 80,164 of those parent RPCs, producing 146,218 child rows across five (fork digest, topic name, encoding) tuples and no peer-exchange entries. Those child rows were not 146,218 peers leaving Ethereum.
Optimism's July 7 status update was oddly specific: after a brief unsafe-head stall, the sequencer was producing blocks again, but some nodes were not syncing. Ethereum saw another symptom at almost exactly the same time. OP Mainnet's L1 blob batcher stopped posting for 23 minutes and 24 seconds, its longest gap in a scan of 14 complete days.
engine_newPayload just gained another three and a half months of history. That is useful, but the new line has a seam in it: on March 27, the regular-node sample jumps from six observers and two execution clients to as many as fifteen observers and four clients.
If you plot "fastest client" straight through that seam, Nethermind appears to improve from 47.5% of observed slots to 94.9% in two days. It did not suddenly get twice as fast. Nine more runners walked onto the track.
A six-hour moving average that only sees one hour is not smoothing much. At 23:00 UTC on July 10, Xatu's fct_proposer_reward_hourly stored 0.303736 ETH in its moving_avg_reward_eth field. The actual six-hour mean was 0.017821 ETH.
The 17x gap came from one 12.506337 ETH relay-delivered block, included eleven seconds after the hour began. Because it landed first, the stored calculation kept re-counting it through almost every cumulative average in that hour. The previous five hours never entered the window.
A percentile does not survive being percentiled again. Xatu's hourly receipt-size table has a column labelled p95_receipt_bytes_per_transaction, but on July 10 it never matched the p95 across transactions. The table value was 1.50x to 3.90x higher, with a median ratio of 2.67x across the 24 hourly buckets.
The data was not stale and the transaction denominator was not drifting. Both paths covered the same 2,823,931 canonical transactions. They were simply calculating different statistics under the same name.
Yesterday's heartbeat post left an uncomfortable question. The live connections were old, but were the peers on the other end actually keeping up? Mostly, yes.
At 12:00 UTC on July 10, 93.76% of matched observer-peer pairs reported the same head slot. Another 205 live pairs, covering 145 remote peer keys, answered from more than 7,200 slots behind. The median lag inside that tail was 116,439 slots, or 16.2 days.
Ethereum's consensus spec just made QUIC the mandatory, preferred libp2p transport. Xatu's connection rows show that QUIC was already doing plenty of work, but almost entirely in one direction.
Across June 27 through July 10, the Tysm-instrumented mainnet sample recorded 6,884,989 outbound UDP/QUIC session opens, or 43.56% of outbound opens. The inbound side had 88. That is 0.0036% of 2.43 million inbound opens.
earliest_available_slot sounds like one clean boundary. It is actually two storage policies squeezed into one integer.
At noon UTC on July 10, I found 14,598 near-head observer-peer pairs in Xatu Status v2 telemetry. 83.00% advertised at least the old five-month block-history window, while 3.96% advertised less than PeerDAS's 18.2-day sidecar window. The split was aggressively client-shaped.
Two days ago I wrote that most observed libp2p disconnect sessions lasted one second. The number was right. The mental picture it invites is not.
Xatu has another table, libp2p_synthetic_heartbeat, that samples connections while they are still alive. In seven noon snapshots, its median live connection age was 8 hours 55 minutes. The disconnect table's median was still one second.