Ethereum Block

People often say “a block is a bag of transactions.” That shortcut is useful, but it hides the part that actually protects users and protocols.

A block is a replayable commitment: anyone should be able to re-execute its ordered transactions and reach the same roots. Without that property, there is no shared state machine—only competing opinions.

Why blocks exist

Public chains need one object that does three jobs at once:

  • order transactions,
  • commit execution output,
  • anchor history.

Ethereum’s block design is that object.

Block composition

At the payload level, you can split block composition into four surfaces:

  • header commitments: parent_hash, state_root, receipts_root
  • ordered tx list: execution input sequence
  • gas/fee fields: gas_used, base_fee_per_gas
  • derived commitments: replay output roots

If any replayed result disagrees with committed roots, the block is invalid.

Block composition
gas_used: 16.5M
base_fee step: +1.25%
replay target: state_root match

Replay invariant

Validation is local replay, not blind trust.

staten+1=ApplyBlock(staten, txsn)\text{state}_{n+1}=\mathrm{ApplyBlock}(\text{state}_n,\ \text{txs}_n)

This single invariant is what keeps independent clients convergent.

CL and EL split

Post-Merge, Ethereum has two cooperating layers:

  • EL computes execution results.
  • CL chooses which history is canonical.

Short form: EL says what happened; CL says which branch survives.

Fee surface

Type-2 execution-lane fee:

feeexec=gasUsedâ‹…(baseFee+effectiveTip)\text{fee}_{exec}=\text{gasUsed}\cdot(\text{baseFee}+\text{effectiveTip})

baseFee is protocol-adjusted and burned. Tip is market-driven inclusion incentive.

Important: fee payment and settlement confidence are different axes.

Confidence tiers

Application logic should not use one binary “confirmed” state:

  • head: fastest, highest reorg risk
  • safe: lower reorg risk
  • finalized: economically expensive to revert

For high-value transitions, gate on finalized, not merely “included.”

Builder reality

Real inclusion flow is often builder-mediated, not one globally shared mempool queue.

So these statements are different:

  • “I broadcast a tx.”
  • “The network has one canonical ordering view.”

Monitor latency distribution, replacement paths, and cross-provider divergence—not only median gas price.

Failure patterns

Common production failures:

  • treating execution success as final settlement,
  • relying on one RPC trust domain,
  • missing reorg-aware indexing buffers,
  • nonce races across workers.

Most outages come from weak state models, not broken cryptography.

Mental model

A block is not a container.

It is a replayable commitment to ordered state transition.

References