You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
each layer inherits most of the previous layer's functionality and adds additional restrictions:
sneakernet/implicit transport honest threat model
bip draft dependencies:
concurrent & extensions - basic data format
scrubbing - enforcing boundary between privacy domains e.g. own signing devices and collaborative txn construction
hpke - optional - for embedding of out-of-band messages into psbt files
define net settlement / payment confirmation
base definition assumes out-of-band
can be made in-band with hpke
all "messages" are just PSBTs with a partial order (can't guarantee a join semilattice in general)
should have discussion of privacy
interactive transaction construction w/ pluggable transports in honest threat model
use cases
multisig
internally segregated wallets (e.g. operational vs. customer funds)
collaborative batching
net-settlement
define message space to avoid in-band
framing format for duplex stream transports
broadcast
leader
clique
gossip
optional compatibility with lightning interactive-tx by bridging
semi-honest
considerations for on chain structure
restrictions on transport
anonymous broadcast by relay w/ anonymous p2p transport
implications for liveness
bft
threat model, incentives
coalition formation
ownership proofs & online key enrollment
listen advertisements
co-spend proposals
coalition formation proposal
role of aggregator in bootstrapping agreement
causal log structure, bft crdts = liveness in async/eventual consistency communication model (this may make sense to split, and move the non-bft component to interactive honest)
validity proofs
bft agreement
setchain/redebelly + weak coordinator (bootstrapped by aggregator)
please add these distinctions to #35 instead of here
the more accurate way to convey this is the topology of the network graph.
don't use "leader" it implies consensus/agreement, what i meant to imply is more of a hub and spoke model where certain peers (in other protocols sometimes called "supernodes") that are better connected, and to whom many of the other peers connect.
the clique model implies everyone is connected to everyone else, or at least no differentiation. so this might be a complete graph over tor, or some other transport, but also something like nym where you can send a message to any peer at any time, without a connection.
finally gossip, depending on the model, is usually the random telephone model, i.e. you are potentially connected to all peers and at every round pick one neighbor to communicate with, and share new information, or the local model, where a peer is connected to only some of the network. we mainly care about this second model, where the number of connections per peer is limited.
Activity
bc1cindy commentedon Aug 10, 2026
EDIT: clear
nothingmuch commentedon Aug 10, 2026
sorry i already called dibs on it in #11 gonna self assign to make that clear
yuval-block commentedon Oct 13, 2026
@0xZaddyy @bc1cindy @zealsham:
please add these distinctions to #35 instead of here
the more accurate way to convey this is the topology of the network graph.
don't use "leader" it implies consensus/agreement, what i meant to imply is more of a hub and spoke model where certain peers (in other protocols sometimes called "supernodes") that are better connected, and to whom many of the other peers connect.
the clique model implies everyone is connected to everyone else, or at least no differentiation. so this might be a complete graph over tor, or some other transport, but also something like nym where you can send a message to any peer at any time, without a connection.
finally gossip, depending on the model, is usually the random telephone model, i.e. you are potentially connected to all peers and at every round pick one neighbor to communicate with, and share new information, or the local model, where a peer is connected to only some of the network. we mainly care about this second model, where the number of connections per peer is limited.
[WIP] Overview
[WIP] Overview
[WIP] Overview
[WIP] Overview
[WIP] Overview
[WIP] Overview
[WIP] Overview
[WIP] Overview
[WIP] Overview
[WIP] Overview
Add an overview
Add an overview
Add an overview
Add an overview
Add an overview
Add an overview
Add an overview
Add an overview
Add an overview