builderwatch

builderwatch

who built the last blocks of ethereum. every block header carries 32 bytes of extraData, and a builder writes its name there; a validator that builds its own block leaves whatever its client puts there. read the last hour of headers from a public rpc, count, and the market for blocks is one table. no keys, no relay api, no explorer.

$ builderwatch
who built the last 300 blocks: 25,982,343 to 25,982,642, 10:43 to 11:43 utc, 2026-09-15

builder          blocks  share   full  base fee  fee recipient
titan               168  56.0%  55.5%     0.065  0x4838…5f97    titanbuilder.xyz
quasar               49  16.3%  63.0%     0.063  0x3963…aa49    quasar.win
buildernet           28   9.3%  23.8%     0.062  0xdadb…3711    buildernet.org
no builder           20   6.7%  17.0%     0.065  15 addresses   validators' own nodes: (empty) x5, geth 1.17.3 x4, geth 1.17.5 x3
eureka               17   5.7%  65.1%     0.062  0xfb74…571b    eurekabuilder.xyz
bombora              12   4.0%  45.6%     0.067  0x85c2…cb33    bombora.build
bob the builder       3   1.0%  20.2%     0.067  0xd87f…4d96    bobthebuilder.xyz
bitget                1   0.3%   8.1%     0.066  0x1d5b…00dc    bitget.com
builder+              1   0.3%  34.0%     0.073  0x400b…8485    btcs.com/builder
.                     1   0.3%  11.8%     0.066  0x7d0a…0474    unknown, the extra data as it is

slots without a block: 3 of 303 (1.0%)
gas limit 60,000,000; the mean block was 50.6% full, 30.4m gas; median base fee 0.064 gwei; 81,038 transactions and 1,627 blobs in the window
share is of blocks built, not of slots. a block with a builder name in its extra data was bought through mev-boost;
"no builder" is a validator that built its own block with its execution client, named by that client.

two builders made 72% of the blocks in that hour, and the top five made 91%. the blocks validators built themselves were 17% full against 55% for titan's: a builder fills a block with everything that pays, a local node takes what its own mempool saw. three slots in the hour got no block at all.

install

pipx install git+https://github.com/alinaschanz/builderwatch

or clone it and run python -m builderwatch from the folder. python 3.10 or newer, no dependencies.

use

builderwatch                       # the last 300 blocks, about an hour
builderwatch --since 6h            # the blocks of the last six hours, found by bisecting timestamps
builderwatch --blocks 50
builderwatch --json                # the same, every number, the fee recipients and the client strings
builderwatch --csv blocks.csv      # one row per block: builder, extra data as text, recipient, gas, base fee, blobs
builderwatch --since 1h --summary-append data/daily.csv --quiet   # one row per builder, the dataset
builderwatch --rpc https://your.node

how it works

• one eth_getBlockByNumber per block with transaction hashes only, twenty-five to a batch request, four batches in flight. an hour of headers is a few megabytes; two thousand blocks took 68 seconds from publicnode. endpoints publicnode, mevblocker, drpc, tenderly, blastapi in that order, or --rpc yours. the newest block is left out: it can still be reorganised, and a lagging endpoint may not have it.
• names. extraData is decoded as utf-8 and matched against a small table in builderwatch/builders.py: a needle, a short name, the builder's site, and where the line comes from (the first block of a 2,000-block sample on 2026-09-15, or the note that a 2023-2024 name did not show up in that sample). anything else is printed as the text it is, so a new builder appears as itself instead of as a wrong guess. the fee recipient is reported from the headers, not from a table: a builder pays itself, so the address next to the name is the builder's.
• local blocks. geth writes an rlp list [version, "geth", "go1.26.4", "linux"] by default; besu, nethermind, reth and erigon write a version string; some validators clear the field. those blocks are grouped under "no builder" with the clients that made them.
• missed slots. every block since the merge sits on a 12-second slot, so a 24-second gap between two blocks is one slot whose proposer produced nothing (or whose block was reorganised away). the count is gaps inside the window, from the header timestamps alone.
• full is gas used over gas limit; the base fee is the block's own.

reading the table

• share is of the blocks in the window, not of all slots and not of value: a builder that wins half the blocks does not earn half the fees.
• a name in extraData is what the builder chose to write, and nothing stops anyone from writing someone else's. the fee recipient column is the harder fact: the address the block's fees went to.
• a validator whose relay returned no bid, or who runs without mev-boost, builds locally; a local block is not a worse block, it is a block that did not go through the auction.
• an hour is a sample. shares move by a few points from hour to hour; the dataset keeps one hour a night for the longer view.

the dataset

data/daily.csv gets one row per builder every night at 00:43 utc, from the last hour of blocks: blocks, share, how full, the median base fee, the fee recipient, and the slots without a block. columns are in data/README.md; a rerun for the same day replaces the day.

exit codes and scripting

0 after a table, 2 when the window is empty, the duration does not parse or no rpc answered. --json carries the builders with every fee recipient and every client string seen, so builderwatch --json | jq '.builders[] | select(.kind == "local")' lists the self-built blocks.

see also

• blobwatch: who bought the blob space in the same blocks
• gasweek: the base fee those blocks charged, by hour of day
• the notes: alinaschanz.life, the short version on x

verify a release

every release carries the sdist and the wheel, a SHA256SUMS file, an opentimestamps proof of that file, and a build provenance attestation made in github's own signing flow. with the files downloaded into one folder:

sha256sum -c SHA256SUMS
gh attestation verify ./*.whl --owner alinaschanz
ots verify SHA256SUMS.ots

the commit itself is signed.

license

mit. numbers, not calls. not financial advice.