Live: https://ethpandaops.github.io/blocksize/

Interactive calculator for everything an Ethereum slot puts on the wire, across every consensus fork: the beacon block, the ePBS payload envelope (with its EIP-7928 block access list), and blob/DAS sidecars — each as raw SSZ and measured gossip size. All of it is derived from the spec repositories; nothing is transcribed by hand.

The design goal: a new spec release requires zero code changes here.

1. extractor/extract.py runs inside an ethereum/consensus-specs checkout (after its own pyspec build) and dumps every fork's SSZ container schemas, preset constants, config (fork epochs, blob schedule), and referenced EIPs to spec-data/consensus.json. Each container's min/max byte length as computed by the spec's own SSZ implementation (remerkleable) is included as ground truth.
2. extractor/extract_el.py dumps per-fork gas constants (intrinsic costs, EIP-7623 calldata floor, EIP-7825 tx gas cap, blob schedule) from ethereum/execution-specs (EELS, ethereum-execution on PyPI).
3. The frontend's size engine (src/lib/ssz.ts) interprets those schemas generically: SSZ min/max sizing, exact instance sizing, and byte-level construction. The test suite asserts the engine reproduces the spec's own min/max for every container in every fork — if a release introduces an SSZ construct the engine can't handle, tests fail loudly.
4. UI knobs (sliders) are discovered from the schema: every list in the block body becomes a control, capped by its SSZ limit or, for progressive (EIP-7495-style) lists, the fork's processing-limit constants. A new fork's new fields appear automatically — as gloas's payload_attestations and builder requests do today.
5. The update-specs workflow runs nightly (and on demand): it re-runs both extractors against the latest consensus-specs release and EELS HEAD, verifies the engine against the new data, and — when the specs substantively changed — commits the refresh to master and redeploys the site, no human in the loop. If verification fails on real changes, it opens a PR instead.
6. A constants-coverage ratchet (src/lib/coverage.test.ts) enumerates every size-relevant constant in the extracted data and fails when a spec update introduces one without an explicit disposition — a new size cap can't be silently ignored (this class of miss happened once, with EIP-7934's block size cap).

Validation against reality: at mainnet's average 30.3 Mgas, the model's typical block is within ~5% of the measured average raw block size (204KB, xatu, July 2026). Measured Snappy compression on real blocks (2.3×) is stronger than the model's "mixed" calldata scenario, so wire sizes err conservative.

Behavior is keyed to extracted constants, never to fork names or EIP tables: EIP-7623 floor pricing applies because FLOOR_CALLDATA_COST exists in the fork's constants, the tx cap because TX_MAX_GAS_LIMIT exists, and so on.

The tool constructs the actual serialized bytes of the configured block — correct SSZ layout, seeded-random bytes for cryptographic fields (signatures, roots, pubkeys, KZG objects), realistic small integers for counters, and calldata matching the selected scenario (all-zeros / mixed / random) — then runs real Snappy over them:

• gossip: raw (block-format) Snappy, as libp2p gossip uses
• req/resp: framed Snappy with per-64KiB-chunk overhead
• plus the analytic worst-case ceiling (32 + n + n/6)

So "compressed size" is the measured wire size of the modeled content, and the content is an explicit input rather than a hardcoded ratio.

The one modeling layer that is not spec-derived is a small set of name-keyed physical constraints (src/lib/constraints.ts): how full an aggregation bitfield gets for a validator count, how many bytes of calldata a gas limit buys, deposit-contract gas per deposit request. Unmatched fields fall back to their spec worst case.

npm install
npm run dev        # local dev server
npm test           # engine vs spec ground truth + integration tests
npm run build      # production build (dist/)

To refresh spec data locally:

# consensus: needs a consensus-specs checkout with a built pyspec
git clone --depth 1 --branch <tag> https://github.com/ethereum/consensus-specs /tmp/consensus-specs
cd /tmp/consensus-specs && uv sync --all-extras && uv run python -m pysetup.generate_specs --all-forks
uv run python <this-repo>/extractor/extract.py --tag <tag> --out <this-repo>/spec-data/consensus.json

# execution: plain PyPI package
uv venv /tmp/elenv && uv pip install --python /tmp/elenv/bin/python ethereum-execution==<version>
/tmp/elenv/bin/python extractor/extract_el.py --version <version> --out spec-data/el.json

Vite + React + TypeScript + Tailwind CSS v4, snappyjs for compression, deployed to GitHub Pages. No backend; everything runs in the browser.