Transport-Key Delegation
Problem: a solver’s relay-facing instances hold the keys they quote and settle with, which caps how far a solver can scale out its quoting edge and widens the blast radius of a key compromise. Direction: a settlement-key-signed delegation naming a rotatable, revocable hot transport key, publishable in the solver card. This separates stateless quoting instances near users from one hardened settlement core — the highest-leverage protocol change for solver scale-out.Delivery Guarantees on the Relay
Problem: the ephemeral negotiation kinds store nothing, so a request or quote sent while its recipient is offline is dropped. Client timeout and retry cover the request path, but the trade-off is currently implicit deployment behavior rather than specified. Direction: specify what a dedicated RFQ relay may add — for example, holding addressed events until the tagged recipient consumes them, the pattern the highest-volume machine-to-machine Nostr protocol converged on — so applications know which delivery model they are building against.Bounded Panels for Published Requests
Problem: anrfq_open reaches every solver watching the pair. Evidence
from quote-panel markets shows response quality falls as more solvers
compete for one request — winner’s curse and information leakage make
rational solvers quote wide or abstain.
Direction: a client-selected panel of a few solvers, ranked by registry
data and observed performance, with deadline-based selection. Sealed bids are
already in the wire. Exclusivity for the selected solver is the follow-on if
bid racing appears.
Streaming Signed Quotes
Problem: per-request quoting scales with users; every application pays a relay roundtrip for terms that are usually the same answer. Direction: a solver publishes a signed live quote per market as one addressable event, so request load per user becomes broadcast load per solver. Applications skip the roundtrip for common sizes and fall back to an addressed request for size or staleness. The discovery specification sketches this as a dormant layer. Asset cards without a relay already work this way: the card advises the pricing formula, and the client resolves the terms without a roundtrip.HTTP as a Deliberate Fast Path
Problem: the relay is a shared dependency in the hot path of every quoted swap, even for integrations that already trust a specific solver. Direction: addressed payloads are transport-independent today — the reference service accepts HTTP and Nostr alike — and the card’stransports map is the declared seam for a second transport. High-volume
integrations speak HTTPS directly to a solver; the relay remains discovery,
the long tail, and the censorship-resistant fallback. The costs — solver IP
exposure, accepting inbound connections — are the solver’s own deployment
choice.
Fill Fairness for Raced Swaps
Problem: an asset swap is filled by whichever solver spends the funded contract first — a pure latency race that pressures solvers to cluster at the operator. Direction: fairness has to come from mechanism — batching windows, commit-reveal, or pro-rata partial fills — if racing measurably widens spreads. Partial fills are already sketched as a contract extension.Accountable Solver Reputation
Problem: every message is signed, so fill and refusal statistics per solver key are free to collect — but a burned key is also free to replace, so scoring alone is not accountability. Signed bids remain attributable but not bonded, a stated boundary today. Direction: registry admission is the gate that already exists. Beyond it, the precedents from other quote markets are economic: bonds, staking, or hold-invoice collateral at quote acceptance for large sizes.Kind Registration
Problem: the RFQ event kinds are provisional. On shared relays, a collision with another protocol pollutes tag filters and muddies spam attribution. Direction: register the kinds when the wire settles — or commit to dedicated relays, which makes collision moot.Clock Discipline
Problem:valid_until is enforced at second granularity against
client-set, unverifiable timestamps. Nothing states how much skew a
conforming implementation tolerates.
Direction: a stated skew tolerance in the specification, and time-sync
guidance for solver deployments.