eth2.0-specs/specs/core/1_shard-data-chains.md

37 KiB

Ethereum 2.0 Phase 1 -- Shard Data Chains

NOTICE: This document is a work-in-progress for researchers and implementers. It reflects recent spec changes and takes precedence over the Python proof-of-concept implementation.

At the current stage, Phase 1, while fundamentally feature-complete, is still subject to change. Development teams with spare resources may consider starting on the "Shard chains and crosslink data" section; at least basic properties, such as the fact that a shard block can get created every slot and is dependent on both a parent block in the same shard and a beacon chain block at or before that same slot, are unlikely to change, though details are likely to undergo similar kinds of changes to what Phase 0 has undergone since the start of the year.

Table of contents

Introduction

This document represents the specification for Phase 1 of Ethereum 2.0 -- Shard Data Chains. Phase 1 depends on the implementation of Phase 0 -- The Beacon Chain.

Ethereum 2.0 consists of a central beacon chain along with SHARD_COUNT shard chains. Phase 1 is primarily concerned with the construction, validity, and consensus on the data of these shard chains. Phase 1 does not specify shard chain state execution or account balances. This is left for future phases.

Terminology

Constants

Phase 1 depends upon all of the constants defined in Phase 0 in addition to the following:

Misc

Name Value Unit
SHARD_CHUNK_SIZE 2**5 (= 32) bytes
SHARD_BLOCK_SIZE 2**14 (= 16,384) bytes
MINOR_REWARD_QUOTIENT 2**8 (= 256)
MAX_POC_RESPONSE_DEPTH 5
ZERO_PUBKEY int_to_bytes48(0)
VALIDATOR_NULL 2**64 - 1

Time parameters

Name Value Unit Duration
CROSSLINK_LOOKBACK 2**5 (= 32) slots 3.2 minutes
MAX_BRANCH_CHALLENGE_DELAY 2**11 (= 2,048) epochs 9 days
CUSTODY_PERIOD_LENGTH 2**11 (= 2,048) epochs 9 days
PERSISTENT_COMMITTEE_PERIOD 2**11 (= 2,048) epochs 9 days
CHALLENGE_RESPONSE_DEADLINE 2**14 (= 16,384) epochs 73 days

Max operations per block

Name Value
MAX_BRANCH_CHALLENGES 2**2 (= 4)
MAX_BRANCH_RESPONSES 2**4 (= 16)
MAX_EARLY_SUBKEY_REVEALS 2**4 (= 16)
MAX_INTERACTIVE_CUSTODY_CHALLENGE_INITIATIONS 2
MAX_INTERACTIVE_CUSTODY_CHALLENGE_RESPONSES 16
MAX_INTERACTIVE_CUSTODY_CHALLENGE_CONTINUTATIONS 16

Signature domains

Name Value
DOMAIN_SHARD_PROPOSER 129
DOMAIN_SHARD_ATTESTER 130
DOMAIN_CUSTODY_SUBKEY 131
DOMAIN_CUSTODY_INTERACTIVE 132

Shard chains and crosslink data

Helper functions

get_split_offset

def get_split_offset(list_size: int, chunks: int, index: int) -> int:
  """
  Returns a value such that for a list L, chunk count k and index i,
  split(L, k)[i] == L[get_split_offset(len(L), k, i): get_split_offset(len(L), k, i+1)]
  """
  return (list_size * index) // chunks

get_shuffled_committee

def get_shuffled_committee(state: BeaconState,
                           shard: Shard,
                           committee_start_epoch: Epoch) -> List[ValidatorIndex]:
    """
    Return shuffled committee.
    """
    validator_indices = get_active_validator_indices(state.validators, committee_start_epoch)
    seed = generate_seed(state, committee_start_epoch)
    start_offset = get_split_offset(len(validator_indices), SHARD_COUNT, shard)
    end_offset = get_split_offset(len(validator_indices), SHARD_COUNT, shard + 1)
    return [
        validator_indices[get_permuted_index(i, len(validator_indices), seed)]
        for i in range(start_offset, end_offset)
    ]

get_persistent_committee

def get_persistent_committee(state: BeaconState,
                             shard: Shard,
                             epoch: Epoch) -> List[ValidatorIndex]:
    """
    Return the persistent committee for the given ``shard`` at the given ``epoch``.
    """
    earlier_committee_start_epoch = epoch - (epoch % PERSISTENT_COMMITTEE_PERIOD) - PERSISTENT_COMMITTEE_PERIOD * 2
    earlier_committee = get_shuffled_committee(state, shard, earlier_committee_start_epoch)

    later_committee_start_epoch = epoch - (epoch % PERSISTENT_COMMITTEE_PERIOD) - PERSISTENT_COMMITTEE_PERIOD
    later_committee = get_shuffled_committee(state, shard, later_committee_start_epoch)

    def get_switchover_epoch(index):
        return (
            bytes_to_int(hash(earlier_seed + bytes3(index))[0:8]) %
            PERSISTENT_COMMITTEE_PERIOD
        )

    # Take not-yet-cycled-out validators from earlier committee and already-cycled-in validators from
    # later committee; return a sorted list of the union of the two, deduplicated
    return sorted(list(set(
        [i for i in earlier_committee if epoch % PERSISTENT_COMMITTEE_PERIOD < get_switchover_epoch(i)] +
        [i for i in later_committee if epoch % PERSISTENT_COMMITTEE_PERIOD >= get_switchover_epoch(i)]
    )))

get_shard_proposer_index

def get_shard_proposer_index(state: BeaconState,
                             shard: Shard,
                             slot: Slot) -> ValidatorIndex:
    seed = hash(
        state.current_shuffling_seed +
        int_to_bytes8(shard) +
        int_to_bytes8(slot)
    )
    persistent_committee = get_persistent_committee(state, shard, slot_to_epoch(slot))
    # Default proposer
    index = bytes_to_int(seed[0:8]) % len(persistent_committee)
    # If default proposer exits, try the other proposers in order; if all are exited
    # return None (ie. no block can be proposed)
    validators_to_try = persistent_committee[index:] + persistent_committee[:index]
    for index in validators_to_try:
        if is_active_validator(state.validators[index], get_current_epoch(state)):
            return index
    return None

Data Structures

Shard chain blocks

A ShardBlock object has the following fields:

{
    # Slot number
    'slot': 'uint64',
    # What shard is it on
    'shard_id': 'uint64',
    # Parent block's root
    'parent_root': 'bytes32',
    # Beacon chain block
    'beacon_chain_ref': 'bytes32',
    # Merkle root of data
    'data_root': 'bytes32'
    # State root (placeholder for now)
    'state_root': 'bytes32',
    # Block signature
    'signature': 'bytes96',
    # Attestation
    'participation_bitfield': 'bytes',
    'aggregate_signature': 'bytes96',
}

Shard block processing

For a shard_block on a shard to be processed by a node, the following conditions must be met:

  • The ShardBlock pointed to by shard_block.parent_root has already been processed and accepted
  • The signature for the block from the proposer (see below for definition) of that block is included along with the block in the network message object

To validate a block header on shard shard_block.shard_id, compute as follows:

  • Verify that shard_block.beacon_chain_ref is the hash of a block in the (canonical) beacon chain with slot less than or equal to slot.
  • Verify that shard_block.beacon_chain_ref is equal to or a descendant of the shard_block.beacon_chain_ref specified in the ShardBlock pointed to by shard_block.parent_root.
  • Let state be the state of the beacon chain block referred to by shard_block.beacon_chain_ref.
  • Let persistent_committee = get_persistent_committee(state, shard_block.shard_id, slot_to_epoch(shard_block.slot)).
  • Assert verify_bitfield(shard_block.participation_bitfield, len(persistent_committee))
  • For every i in range(len(persistent_committee)) where is_active_validator(state.validators[persistent_committee[i]], get_current_epoch(state)) returns False, verify that get_bitfield_bit(shard_block.participation_bitfield, i) == 0
  • Let proposer_index = get_shard_proposer_index(state, shard_block.shard_id, shard_block.slot).
  • Verify that proposer_index is not None.
  • Let msg be the shard_block but with shard_block.signature set to [0, 0].
  • Verify that bls_verify(pubkey=validators[proposer_index].pubkey, message_hash=hash(msg), signature=shard_block.signature, domain=get_domain(state, slot_to_epoch(shard_block.slot), DOMAIN_SHARD_PROPOSER)) passes.
  • Let group_public_key = bls_aggregate_pubkeys([state.validators[index].pubkey for i, index in enumerate(persistent_committee) if get_bitfield_bit(shard_block.participation_bitfield, i) is True]).
  • Verify that bls_verify(pubkey=group_public_key, message_hash=shard_block.parent_root, sig=shard_block.aggregate_signature, domain=get_domain(state, slot_to_epoch(shard_block.slot), DOMAIN_SHARD_ATTESTER)) passes.

Verifying shard block data

At network layer, we expect a shard block header to be broadcast along with its block_body.

  • Verify that len(block_body) == SHARD_BLOCK_SIZE
  • Verify that merkle_root(block_body) equals the data_root in the header.

A node should sign a crosslink only if the following conditions hold. If a node has the capability to perform the required level of verification, it should NOT follow chains on which a crosslink for which these conditions do NOT hold has been included, or a sufficient number of signatures have been included that during the next state recalculation, a crosslink will be registered.

First, the conditions must recursively apply to the crosslink referenced in last_crosslink_root for the same shard (unless last_crosslink_root equals zero, in which case we are at the genesis).

Second, we verify the shard_chain_commitment.

  • Let start_slot = state.latest_crosslinks[shard].epoch * SLOTS_PER_EPOCH + SLOTS_PER_EPOCH - CROSSLINK_LOOKBACK.
  • Let end_slot = attestation.data.slot - attestation.data.slot % SLOTS_PER_EPOCH - CROSSLINK_LOOKBACK.
  • Let length = end_slot - start_slot, headers[0] .... headers[length-1] be the serialized block headers in the canonical shard chain from the verifer's point of view (note that this implies that headers and bodies have been checked for validity).
  • Let bodies[0] ... bodies[length-1] be the bodies of the blocks.
  • Note: If there is a missing slot, then the header and body are the same as that of the block at the most recent slot that has a block.

We define two helpers:

def pad_to_power_of_2(values: List[bytes]) -> List[bytes]:
    zero_shard_block = b'\x00' * SHARD_BLOCK_SIZE
    while not is_power_of_two(len(values)):
        values = values + [zero_shard_block]
    return values
def merkle_root_of_bytes(data: bytes) -> bytes:
    return merkle_root([data[i:i + 32] for i in range(0, len(data), 32)])

We define the function for computing the commitment as follows:

def compute_commitment(headers: List[ShardBlock], bodies: List[bytes]) -> Bytes32:
    return hash(
        merkle_root(
            pad_to_power_of_2([
                merkle_root_of_bytes(zpad(serialize(h), SHARD_BLOCK_SIZE)) for h in headers
            ])
        ) +
        merkle_root(
            pad_to_power_of_2([
                merkle_root_of_bytes(h) for h in bodies
            ])
        )
    )

The shard_chain_commitment is only valid if it equals compute_commitment(headers, bodies).

Shard block fork choice rule

The fork choice rule for any shard is LMD GHOST using the shard chain attestations of the persistent committee and the beacon chain attestations of the crosslink committee currently assigned to that shard, but instead of being rooted in the genesis it is rooted in the block referenced in the most recent accepted crosslink (ie. state.crosslinks[shard].shard_block_root). Only blocks whose beacon_chain_ref is the block in the main beacon chain at the specified slot should be considered (if the beacon chain skips a slot, then the block at that slot is considered to be the block in the beacon chain at the highest slot lower than a slot).

Updates to the beacon chain

Data structures

Validator

Add member values to the end of the Validator object:

    'next_subkey_to_reveal': 'uint64',
    'reveal_max_periods_late': 'uint64',

And the initializers:

    'next_subkey_to_reveal': get_current_custody_period(state),
    'reveal_max_periods_late': 0,

BeaconBlockBody

Add member values to the BeaconBlockBody structure:

    'branch_challenges': [BranchChallenge],
    'branch_responses': [BranchResponse],
    'subkey_reveals': [SubkeyReveal],
    'interactive_custody_challenge_initiations': [InteractiveCustodyChallengeInitiation],
    'interactive_custody_challenge_responses': [InteractiveCustodyChallengeResponse],
    'interactive_custody_challenge_continuations': [InteractiveCustodyChallengeContinuation],

And initialize to the following:

    'branch_challenges': [],
    'branch_responses': [],
    'subkey_reveals': [],

BeaconState

Add member values to the BeaconState structure:

    'branch_challenge_records': [BranchChallengeRecord],
    'next_branch_challenge_id': 'uint64',
    'custody_challenge_records': [InteractiveCustodyChallengeRecord],
    'next_custody_challenge_id': 'uint64',

BranchChallenge

Define a BranchChallenge as follows:

{
    'responder_index': 'uint64',
    'data_index': 'uint64',
    'attestation': SlashableAttestation,
}

BranchResponse

Define a BranchResponse as follows:

{
    'challenge_id': 'uint64',
    'responding_to_custody_challenge': 'bool',
    'data': 'bytes32',
    'branch': ['bytes32'],
}

BranchChallengeRecord

Define a BranchChallengeRecord as follows:

{
    'challenge_id': 'uint64',
    'challenger_index': 'uint64',
    'responder_index': 'uint64',
    'root': 'bytes32',
    'depth': 'uint64',
    'deadline': 'uint64',
    'data_index': 'uint64',
}

InteractiveCustodyChallengeRecord

{
    'challenge_id': 'uint64',
    'challenger_index': 'uint64',
    'responder_index': 'uint64',
    # Initial data root
    'data_root': 'bytes32',
    # Initial custody bit
    'custody_bit': 'bool',
    # Responder subkey
    'responder_subkey': 'bytes96',
    # The hash in the PoC tree in the position that we are currently at
    'current_custody_tree_node': 'bytes32',
    # The position in the tree, in terms of depth and position offset
    'depth': 'uint64',
    'offset': 'uint64',
    # Max depth of the branch
    'max_depth': 'uint64',
    # Deadline to respond (as an epoch)
    'deadline': 'uint64',
}

InteractiveCustodyChallengeInitiation

{
    'attestation': SlashableAttestation,
    'responder_index': 'uint64',
    'challenger_index': 'uint64',
    'responder_subkey': 'bytes96',
    'signature': 'bytes96',
}

InteractiveCustodyChallengeResponse

{
    'challenge_id': 'uint64',
    'hashes': ['bytes32'],
    'signature': 'bytes96',
}

InteractiveCustodyChallengeContinuation

{
    'challenge_id': 'uint64',
    'sub_index': 'uint64',
    'new_custody_tree_node': 'bytes32',
    'proof': ['bytes32'],
    'signature': 'bytes96',
}

SubkeyReveal

Define a SubkeyReveal as follows:

{
    'validator_index': 'uint64',
    'period': 'uint64',
    'subkey': 'bytes96',
    'mask': 'bytes32',
    'revealer_index': 'uint64'
}

Helpers

get_branch_challenge_record_by_id

def get_branch_challenge_record_by_id(state: BeaconState, id: int) -> BranchChallengeRecord:
    return [c for c in state.branch_challenges if c.challenge_id == id][0]

get_custody_challenge_record_by_id

def get_custody_challenge_record_by_id(state: BeaconState, id: int) -> BranchChallengeRecord:
    return [c for c in state.branch_challenges if c.challenge_id == id][0]

get_attestation_merkle_depth

def get_attestation_merkle_depth(attestation: Attestation) -> int:
    start_epoch = attestation.data.latest_crosslink.epoch
    end_epoch = slot_to_epoch(attestation.data.slot)
    chunks_per_slot = SHARD_BLOCK_SIZE // 32
    chunks = (end_epoch - start_epoch) * EPOCH_LENGTH * chunks_per_slot
    return log2(next_power_of_two(chunks))

epoch_to_custody_period

def epoch_to_custody_period(epoch: Epoch) -> int:
    return epoch // CUSTODY_PERIOD_LENGTH

slot_to_custody_period

def slot_to_custody_period(slot: Slot) -> int:
    return epoch_to_custody_period(slot_to_epoch(slot))

get_current_custody_period

def get_current_custody_period(state: BeaconState) -> int:
    return epoch_to_custody_period(get_current_epoch(state))

verify_custody_subkey_reveal

def verify_custody_subkey_reveal(pubkey: bytes48,
                                 subkey: bytes96,
                                 mask: bytes32,
                                 mask_pubkey: bytes48,
                                 period: int) -> bool:
    # Legitimate reveal: checking that the provided value actually is the subkey
    if mask == ZERO_HASH:
        pubkeys=[pubkey]
        message_hashes=[hash(int_to_bytes8(period))]
        
    # Punitive early reveal: checking that the provided value is a valid masked subkey
    # (masking done to prevent "stealing the reward" from a whistleblower by block proposers)
    # Secure under the aggregate extraction infeasibility assumption described on page 11-12
    # of https://crypto.stanford.edu/~dabo/pubs/papers/aggreg.pdf
    else:
        pubkeys=[pubkey, mask_pubkey]
        message_hashes=[hash(int_to_bytes8(period)), mask]
        
    return bls_multi_verify(
        pubkeys=pubkeys,
        message_hashes=message_hashes,
        signature=subkey,
        domain=get_domain(
            fork=state.fork,
            epoch=period * CUSTODY_PERIOD_LENGTH,
            domain_type=DOMAIN_CUSTODY_SUBKEY,
        )
    )

verify_signed_challenge_message

def verify_signed_challenge_message(message: Any, pubkey: bytes48) -> bool:
    return bls_verify(
        message_hash=signed_root(message),
        pubkey=pubkey,
        signature=message.signature,
        domain=get_domain(state, get_current_epoch(state), DOMAIN_CUSTODY_INTERACTIVE)
    )

penalize_validator

Change the definition of penalize_validator as follows:

def penalize_validator(state: BeaconState, index: ValidatorIndex, whistleblower_index=None:ValidatorIndex) -> None:
    """
    Penalize the validator of the given ``index``.
    Note that this function mutates ``state``.
    """
    exit_validator(state, index)
    validator = state.validator_registry[index]
    state.latest_penalized_balances[get_current_epoch(state) % LATEST_PENALIZED_EXIT_LENGTH] += get_effective_balance(state, index)
    
    block_proposer_index = get_beacon_proposer_index(state, state.slot)
    whistleblower_reward = get_effective_balance(state, index) // WHISTLEBLOWER_REWARD_QUOTIENT
    if whistleblower_index is None:
        state.validator_balances[block_proposer_index] += whistleblower_reward
    else:
        state.validator_balances[whistleblower_index] += (
            whistleblower_reward * INCLUDER_REWARD_QUOTIENT / (INCLUDER_REWARD_QUOTIENT + 1)
        )
        state.validator_balances[block_proposer_index] += whistleblower_reward / (INCLUDER_REWARD_QUOTIENT + 1)
    state.validator_balances[index] -= whistleblower_reward
    validator.penalized_epoch = get_current_epoch(state)
    validator.withdrawable_epoch = get_current_epoch(state) + LATEST_PENALIZED_EXIT_LENGTH

The only change is that this introduces the possibility of a penalization where the "whistleblower" that takes credit is NOT the block proposer.

Per-slot processing

Operations

Add the following operations to the per-slot processing, in order the given below and after all other operations (specifically, right after exits).

Branch challenges

Verify that len(block.body.branch_challenges) <= MAX_BRANCH_CHALLENGES.

For each challenge in block.body.branch_challenges, run:

def process_branch_challenge(state: BeaconState,
                             challenge: BranchChallenge) -> None:
    # Check that it's not too late to challenge
    assert slot_to_epoch(challenge.attestation.data.slot) >= get_current_epoch(state) - MAX_BRANCH_CHALLENGE_DELAY
    assert state.validator_registry[responder_index].exit_epoch >= get_current_epoch(state) - MAX_BRANCH_CHALLENGE_DELAY
    # Check the attestation is valid
    assert verify_slashable_attestation(state, challenge.attestation)
    # Check that the responder participated
    assert challenger.responder_index in challenge.attestation.validator_indices
    # Check the challenge is not a duplicate
    assert [
        c for c in state.branch_challenge_records if c.root == challenge.attestation.data.crosslink_data_root and
        c.data_index == challenge.data_index
    ] == []
    # Check validity of depth
    depth = get_attestation_merkle_depth(challenge.attestation)
    assert c.data_index < 2**depth
    # Add new challenge
    state.branch_challenge_records.append(BranchChallengeRecord(
        challenge_id=state.next_branch_challenge_id,
        challenger_index=get_beacon_proposer_index(state, state.slot),
        root=challenge.attestation.data.shard_chain_commitment,
        depth=depth,
        deadline=get_current_epoch(state) + CHALLENGE_RESPONSE_DEADLINE,
        data_index=challenge.data_index
    ))
    state.next_branch_challenge_id += 1

Branch responses

Verify that len(block.body.branch_responses) <= MAX_BRANCH_RESPONSES.

For each response in block.body.branch_responses, if response.responding_to_custody_challenge == False, run:

def process_branch_exploration_response(state: BeaconState,
                                        response: BranchResponse) -> None:
    challenge = get_branch_challenge_record_by_id(response.challenge_id)
    assert verify_merkle_branch(
        leaf=response.data,
        branch=response.branch,
        depth=challenge.depth,
        index=challenge.data_index,
        root=challenge.root
    )
    # Must wait at least ENTRY_EXIT_DELAY before responding to a branch challenge
    assert get_current_epoch(state) >= challenge.inclusion_epoch + ENTRY_EXIT_DELAY
    state.branch_challenge_records.pop(challenge)
    # Reward the proposer
    proposer_index = get_beacon_proposer_index(state, state.slot)
    state.validator_balances[proposer_index] += base_reward(state, index) // MINOR_REWARD_QUOTIENT

If response.responding_to_custody_challenge == True, run:

def process_branch_custody_response(state: BeaconState,
                                    response: BranchResponse) -> None:
    challenge = get_custody_challenge_record_by_id(response.challenge_id)
    responder = state.validator_registry[challenge.responder_index]
    # Verify we're not too late
    assert get_current_epoch(state) < responder.withdrawable_epoch
    # Verify the Merkle branch *of the data tree*
    assert verify_merkle_branch(
        leaf=response.data,
        branch=response.branch,
        depth=challenge.max_depth,
        index=challenge.offset,
        root=challenge.data_root
    )
    # Responder wins
    if hash(challenge.responder_subkey + response.data) == challenge.current_custody_tree_node:
        penalize_validator(state, challenge.challenger_index, challenge.responder_index)
    # Challenger wins
    else:
        penalize_validator(state, challenge.responder_index, challenge.challenger_index)
    state.custody_challenge_records.pop(challenge)

Subkey reveals

Verify that len(block.body.early_subkey_reveals) <= MAX_EARLY_SUBKEY_REVEALS.

For each reveal in block.body.early_subkey_reveals:

  • Verify that verify_custody_subkey_reveal(state.validator_registry[reveal.validator_index].pubkey, reveal.subkey, reveal.period, reveal.mask, state.validator_registry[reveal.revealer_index].pubkey) returns True.
  • Let is_early_reveal = reveal.period > get_current_custody_period(state) or (reveal.period == get_current_custody_period(state) and state.validator_registry[reveal.validator_index].exit_epoch > get_current_epoch(state)) (ie. either the reveal is of a future period, or it's of the current period and the validator is still active)
  • Verify that one of the following is true:
    • (i) is_early_reveal is True
    • (ii) is_early_reveal is False and reveal.period == state.validator_registry[reveal.validator_index].next_subkey_to_reveal (revealing a past subkey, or a current subkey for a validator that has exited) and reveal.mask == ZERO_HASH

In case (i):

  • Verify that `state.validator_registry[reveal.validator_index].penalized_epoch > get_current_epoch(state).
  • Run penalize_validator(state, reveal.validator_index, reveal.revealer_index).
  • Set state.validator_balances[reveal.revealer_index] += base_reward(state, index) // MINOR_REWARD_QUOTIENT

In case (ii):

  • Determine the proposer proposer_index = get_beacon_proposer_index(state, state.slot) and set state.validator_balances[proposer_index] += base_reward(state, index) // MINOR_REWARD_QUOTIENT.
  • Set state.validator_registry[reveal.validator_index].next_subkey_to_reveal += 1
  • Set state.validator_registry[reveal.validator_index].reveal_max_periods_late = max(state.validator_registry[reveal.validator_index].reveal_max_periods_late, get_current_period(state) - reveal.period).

Interactive custody challenge initiations

Verify that len(block.body.interactive_custody_challenge_initiations) <= MAX_INTERACTIVE_CUSTODY_CHALLENGE_INITIATIONS.

For each initiation in block.body.interactive_custody_challenge_initiations, use the following function to process it:

def process_initiation(state: BeaconState,
                       initiation: InteractiveCustodyChallengeInitiation) -> None:
    challenger = state.validator_registry[initiation.challenger_index]
    responder = state.validator_registry[initiation.responder_index]
    # Verify the signature
    assert verify_signed_challenge_message(initiation, challenger.pubkey)
    # Verify the attestation
    assert verify_slashable_attestation(initiation.attestation, state)
    # Check that the responder actually participated in the attestation
    assert initiation.responder_index in attestation.validator_indices
    # Any validator can be a challenger or responder of max 1 challenge at a time
    for c in state.custody_challenge_records:
        assert c.challenger_index != initiation.challenger_index
        assert c.responder_index != initiation.responder_index
    # Can't challenge if you've been penalized
    assert challenger.penalized_epoch == FAR_FUTURE_EPOCH
    # Make sure the revealed subkey is valid
    assert verify_custody_subkey_reveal(
        pubkey=state.validator_registry[responder_index].pubkey,
        subkey=initiation.responder_subkey,
        period=slot_to_custody_period(attestation.data.slot)
    )
    # Verify that the attestation is still eligible for challenging
    min_challengeable_epoch = responder.exit_epoch - CUSTODY_PERIOD_LENGTH * (1 + responder.reveal_max_periods_late)
    assert min_challengeable_epoch <= slot_to_epoch(initiation.attestation.data.slot) 
    # Create a new challenge object
    state.branch_challenge_records.append(InteractiveCustodyChallengeRecord(
        challenge_id=state.next_branch_challenge_id,
        challenger_index=initiation.challenger_index,
        responder_index=initiation.responder_index,
        data_root=attestation.custody_commitment,
        custody_bit=get_bitfield_bit(attestation.custody_bitfield, attestation.validator_indices.index(responder_index)),
        responder_subkey=responder_subkey,
        current_custody_tree_node=ZERO_HASH,
        depth=0,
        offset=0,
        max_depth=get_attestation_data_merkle_depth(initiation.attestation.data),
        deadline=get_current_epoch(state) + CHALLENGE_RESPONSE_DEADLINE
    ))
    state.next_branch_challenge_id += 1
    # Responder can't withdraw yet!
    state.validator_registry[responder_index].withdrawable_epoch = FAR_FUTURE_EPOCH

Interactive custody challenge responses

A response provides 32 hashes that are under current known proof of custody tree node. Note that at the beginning the tree node is just one bit of the custody root, so we ask the responder to sign to commit to the top 5 levels of the tree and therefore the root hash; at all other stages in the game responses are self-verifying.

Verify that len(block.body.interactive_custody_challenge_responses) <= MAX_INTERACTIVE_CUSTODY_CHALLENGE_RESPONSES.

For each response in block.body.interactive_custody_challenge_responses, use the following function to process it:

def process_response(state: BeaconState,
                     response: InteractiveCustodyChallengeResponse) -> None:
    challenge = get_custody_challenge_record_by_id(state, response.challenge_id)
    responder = state.validator_registry[challenge.responder_index]
    # Check that the right number of hashes was provided
    expected_depth = min(challenge.max_depth - challenge.depth, MAX_POC_RESPONSE_DEPTH)
    assert 2**expected_depth == len(response.hashes)
    # Must make some progress!
    assert expected_depth > 0
    # Check the hashes match the previously provided root
    root = merkle_root(response.hashes)
    # If this is the first response check the bit and the signature and set the root
    if challenge.depth == 0:
        assert get_bitfield_bit(root, 0) == challenge.custody_bit
        assert verify_signed_challenge_message(response, responder.pubkey)
        challenge.current_custody_tree_node = root
    # Otherwise just check the response against the root
    else:
        assert root == challenge_data.current_custody_tree_node
    # Update challenge data
    challenge.deadline=FAR_FUTURE_EPOCH
    responder.withdrawable_epoch = get_current_epoch(state) + MAX_POC_RESPONSE_DEPTH

Interactive custody challenge continuations

Once a response provides 32 hashes, the challenger has the right to choose any one of them that they feel is constructed incorrectly to continue the game. Note that eventually, the game will get to the point where the new_custody_tree_node is a leaf node.

Verify that len(block.body.interactive_custody_challenge_continuations) <= MAX_INTERACTIVE_CUSTODY_CHALLENGE_CONTINUATIONS.

For each continuation in block.body.interactive_custody_challenge_continuations, use the following function to process it:

def process_continuation(state: BeaconState,
                         continuation: InteractiveCustodyChallengeContinuation) -> None:
    challenge = get_custody_challenge_record_by_id(state, continuation.challenge_id)
    challenger = state.validator_registry[challenge.challenger_index]
    responder = state.validator_registry[challenge.responder_index]
    expected_depth = min(challenge_data.max_depth - challenge_data.depth, MAX_POC_RESPONSE_DEPTH)
    # Verify we're not too late
    assert get_current_epoch(state) < responder.withdrawable_epoch
    # Verify the Merkle branch (the previous custody response provided the next level of hashes so the
    # challenger has the info to make any Merkle branch)
    assert verify_merkle_branch(
        leaf=new_custody_tree_node,
        branch=continuation.proof,
        depth=expected_depth,
        index=sub_index,
        root=challenge_data.current_custody_tree_node
    )
    # Verify signature
    assert verify_signed_challenge_message(continuation, challenger.pubkey)
    # Update the challenge data
    challenge.current_custody_tree_node = continuation.new_custody_tree_node
    challenge.depth += expected_depth
    challenge.deadline = get_current_epoch(state) + MAX_POC_RESPONSE_DEPTH
    responder.withdrawable_epoch = FAR_FUTURE_EPOCH
    challenge.offset = challenge_data.offset * 2**expected_depth + sub_index

Per-epoch processing

Add the following loop immediately below the process_ejections loop:

def process_challenge_absences(state: BeaconState) -> None:
    """
    Iterate through the challenge list
    and penalize validators with balance that did not answer challenges.
    """
    for c in state.branch_challenge_records:
        if get_current_epoch(state) > c.deadline:
            penalize_validator(state, c.responder_index, c.challenger_index)
    
    for c in state.custody_challenge_records:
        if get_current_epoch(state) > c.deadline:
            penalize_validator(state, c.responder_index, c.challenger_index)
        if get_current_epoch(state) > state.validator_registry[c.responder_index].withdrawable_epoch:
            penalize_validator(state, c.challenger_index, c.responder_index)

In process_penalties_and_exits, change the definition of eligible to the following (note that it is not a pure function because state is declared in the surrounding scope):

def eligible(index):
    validator = state.validator_registry[index]
    # Cannot exit if there are still open branch challenges
    if [c for c in state.branch_challenge_records if c.responder_index == index] != []:
        return False
    # Cannot exit if you have not revealed all of your subkeys
    elif validator.next_subkey_to_reveal <= epoch_to_custody_period(validator.exit_epoch):
        return False
    # Cannot exit if you already have
    elif validator.withdrawable_epoch < FAR_FUTURE_EPOCH:
        return False
    # Return minimum time
    else:
        return current_epoch >= validator.exit_epoch + MIN_VALIDATOR_WITHDRAWAL_EPOCHS

One-time phase 1 initiation transition

Run the following on the fork block after per-slot processing and before per-block and per-epoch processing.

For all validator in ValidatorRegistry, update it to the new format and fill the new member values with:

    'next_subkey_to_reveal': get_current_custody_period(state),
    'reveal_max_periods_late': 0,

Update the BeaconState to the new format and fill the new member values with:

    'branch_challenge_records': [],
    'next_branch_challenge_id': 0,
    'custody_challenge_records': [],
    'next_custody_challenge_id': 0,