Dealer vs. Dealerless Key Generation
Dealer-based key generation uses a trusted party to distribute shares; dealerless DKG eliminates that single point of trust.
Key Takeaways
- Dealer-based key generation relies on a single trusted party to create the full secret, split it into key shares using Shamir secret sharing, and distribute them to participants. The dealer must be trusted not to retain or leak the secret.
- Dealerless (distributed) key generation removes that single point of trust: participants jointly produce shares through a DKG protocol so that no single party ever holds the complete secret.
- FROST supports both modes: RFC 9591 defines a trusted dealer approach in its appendix, while separate protocols like ChillDKG provide a dealerless alternative for Bitcoin.
What Is Dealer vs. Dealerless Key Generation?
In any threshold signature scheme, a group of participants needs a set of key shares that together represent a single signing key. The question is how those shares get created. There are two fundamental approaches: dealer-based and dealerless.
In dealer-based key generation, one party (the dealer) generates the full private key, splits it into shares using a technique like Shamir's secret sharing, and hands each share to a participant. The dealer then deletes the original key. This approach is simple and efficient, but it requires complete trust in the dealer during the generation ceremony.
In dealerless key generation, there is no single party who ever knows the full secret. Instead, every participant acts as a partial dealer: each one generates their own random polynomial, shares pieces of it with the others, and the final key shares are assembled from the combined contributions. The group ends up with a valid public key and individual shares, but the corresponding private key was never constructed in one place.
How It Works
Dealer-Based Key Generation
The dealer approach follows the classic Shamir secret sharing pattern. For a t-of-n threshold scheme:
- The dealer generates a random secret s (the group private key)
- The dealer constructs a random polynomial f(x) of degree t-1 with s as the constant term
- The dealer evaluates f(x) at n distinct points to produce n shares
- Each participant receives their share over a secure channel
- The dealer publishes the group public key (s multiplied by the generator point G) and deletes the secret and all shares
To allow participants to verify their shares, the dealer can use Feldman's verifiable secret sharing (VSS). In this variant, the dealer publishes commitments to the polynomial coefficients as group elements. Each participant checks that their share is consistent with these public commitments, catching any dealer who distributes invalid shares.
# Dealer-based FROST key generation (pseudocode)
# Dealer creates a polynomial f(x) = s + a1*x + a2*x^2 + ... + a(t-1)*x^(t-1)
secret = random_scalar()
coefficients = [secret] + [random_scalar() for _ in range(t - 1)]
# Compute shares for each participant
for i in range(1, n + 1):
share_i = evaluate_polynomial(coefficients, i)
send_securely(participant_i, share_i)
# Publish Feldman VSS commitments
for coeff in coefficients:
publish(coeff * G) # group element commitment
# Delete secret and coefficients
secure_delete(secret, coefficients)RFC 9591 (the FROST specification) documents this approach in Appendix C. It defines both a basic Shamir split (C.1) and a verifiable variant (C.2) with Feldman commitments.
Dealerless Key Generation (DKG)
A distributed key generation protocol eliminates the dealer entirely. The most common design is Pedersen's DKG, which works by having every participant run their own Feldman VSS instance in parallel:
- Each participant generates a random polynomial of degree t-1 and computes shares from it for every other participant
- Each participant broadcasts commitments to their polynomial coefficients and sends individual shares to each peer over secure channels
- Every participant verifies the shares they receive against the published commitments
- Each participant sums all the shares they received (one from each peer) to produce their final key share
- The group public key is computed by summing all participants' commitment points for the constant term
The result is identical in structure to a dealer-based setup: a group public key and t-of-n private shares. The difference is that no single participant ever held the full private key. The secret is implicitly defined as the sum of all participants' random constant terms, but this sum is never computed by any individual.
# Dealerless DKG (Pedersen-style, pseudocode)
# Each participant i runs this independently
my_secret = random_scalar()
my_coefficients = [my_secret] + [random_scalar() for _ in range(t - 1)]
# Send shares to each peer
for j in range(1, n + 1):
if j != i:
share_for_j = evaluate_polynomial(my_coefficients, j)
send_securely(participant_j, share_for_j)
# Broadcast commitments
for coeff in my_coefficients:
broadcast(coeff * G)
# After receiving and verifying shares from all peers:
my_final_share = my_own_share + sum(received_shares)
group_pubkey = sum(all_constant_term_commitments)Rogue-Key Protection
A subtle attack on Pedersen DKG is the rogue-key attack: a malicious participant waits to see everyone else's commitments, then crafts their own contribution to bias the group key. The standard mitigation is to require each participant to prove knowledge of their secret via a Schnorr signature (a proof of possession). The FROST paper's DKG variant includes these proofs, and the ChillDKG draft BIP (by Ruffing and Nick at Blockstream) formalizes this approach for Bitcoin's secp256k1 curve.
Comparing the Two Approaches
| Property | Dealer-Based | Dealerless (DKG) |
|---|---|---|
| Trust model | Requires trusting the dealer | No single trusted party |
| Secret exposure | Full secret exists briefly on one machine | Full secret never exists anywhere |
| Communication rounds | One round (dealer to participants) | Multiple interactive rounds |
| Implementation complexity | Low | High |
| Coordination | Offline participants can receive shares later | All participants must be online simultaneously |
| Auditability | Must trust dealer deleted the secret | Cryptographically verifiable |
Use Cases
When Dealer Mode Makes Sense
- Testing and development: a dealer setup is fast and non-interactive, making it ideal for prototyping threshold signature systems
- Controlled environments: when the dealer is an HSM or a trusted institution performing a one-time key ceremony, the trust assumption may be acceptable
- Migrating existing keys: if a single key already exists (such as a legacy hot wallet), a dealer split is the only option since the secret is already known
When Dealerless DKG Is Required
- High-value custody: for institutional Bitcoin custody or protocol treasuries, the risk of a compromised dealer is too great. A DKG ceremony ensures no insider can forge signatures
- Decentralized protocols: systems like Fedimint federations and threshold bridges require DKG because there is no natural trusted party
- Regulatory compliance: some custody frameworks require proving that no single person could have accessed the full key, which only a dealerless ceremony provides
FROST and Key Generation
FROST (Flexible Round-Optimized Schnorr Threshold signatures) supports both key generation modes. The IETF specification (RFC 9591, published June 2024) focuses on the signing protocol and places the trusted dealer approach in an appendix. It deliberately does not mandate a specific DKG protocol, leaving that choice to implementations.
For Bitcoin specifically, the ChillDKG draft BIP proposes a standardized dealerless ceremony built on Pedersen DKG with proofs of possession. It adds Taproot-specific details such as key tweaking for BIP-341 compatibility, making it suitable for multisig wallets and threshold signing setups on Bitcoin.
Spark uses FROST threshold signatures as a core part of its architecture. For a deeper look at how FROST signing works in practice, see the research article on FROST threshold signatures.
Risks and Considerations
Dealer Mode Risks
- Single point of failure: if the dealer is compromised during generation, an attacker obtains the full private key. Even if the dealer honestly deletes the secret, there is no cryptographic proof of deletion
- Memory residue: secrets may persist in RAM, swap files, or backups after deletion. Side-channel attacks during generation could also extract the secret
- No forward security: if the dealer's machine is later forensically examined, the secret could be recovered unless the hardware was physically destroyed
Dealerless DKG Risks
- Liveness requirements: all participants must be online and responsive during the ceremony. A single unresponsive participant can stall the entire process
- Authenticated channels required: participants must communicate over encrypted, authenticated connections. Without this, a network attacker could intercept shares or impersonate participants
- Complexity breeds bugs: DKG protocols involve multiple rounds of communication, commitment verification, and complaint handling. Implementation errors can lead to invalid key shares or exploitable vulnerabilities
- Agreement on success: participants need a mechanism to confirm that everyone completed the ceremony with consistent results. Without explicit agreement, some participants may end up with incompatible shares
Hybrid Approaches
Some systems use a hybrid model: a dealer performs the initial key generation, then the shares are reshared using a DKG protocol to rotate out the original trust assumption. This can be practical when bootstrapping a system quickly while planning a migration to fully dealerless operation.
This glossary entry is for informational purposes only and does not constitute financial or investment advice. Always do your own research before using any protocol or technology.