Skip to content

Trust & Economics

Froglet separates three kinds of trust. Mixing them up is how marketplaces overclaim, so the protocol is explicit about which layer each guarantee lives in:

Guaranteed by cryptographyOperator policy (hosted marketplace)Not guaranteed by anyone
Artifact signatures (BIP340)Listing and indexingExecution correctness
SHA-256 hashes and the chainSuspension via the arbiterProvider competence
Lightning settlement state (payment preimage)Complaint handlingRefund of the base fee on failure
Identity continuity (same key = same actor)Identity attestation (DNS / OAuth domain claims)Future behavior of any key

The live marketplace exposes a receipt-derived trust summary per provider: success count, failure count, rejected/canceled count, and total settled value, computed from the signed receipts the indexer has verified. Two honest caveats:

  • Receipt history can be self-dealt. A provider can run deals against itself to inflate counts. Settled-value totals make this cost real sats in routing fees, but the signal is still a signal, not proof. Weigh it together with attestation and account age, not alone.
  • Domain attestation raises sybil cost. A provider that has proven control of a DNS domain or OAuth identity is more expensive to recreate after misbehavior than a fresh keypair. Filter for attested providers (?attested=true) when the stakes matter.

The hosted marketplace runs a complaint-and-enforcement service. Anyone can file a complaint with evidence against a provider; verdicts can suspend the provider from marketplace listings. Be clear about what this is:

The protocol itself requires no trusted registry — you can verify any provider’s artifacts and run deals against it directly, marketplace or not. The hosted marketplace and arbiter are convenience layers run by an operator, and they are described here as exactly that.

What a receipt proves — and what it doesn’t

Section titled “What a receipt proves — and what it doesn’t”

A signed receipt proves: the provider signed this outcome, the artifact chain is unbroken, the result hash is what the provider committed to, and (on the Lightning rail) the settlement state is cryptographically attested by the payment preimage.

A receipt does not prove the result is correct. Verifying execution correctness without re-running the work is an open research problem (re-execution, trusted hardware, zero-knowledge proofs). Froglet’s position: the receipt makes outcomes attributable and auditable — a provider that signs wrong results signs the evidence against itself — but detection still requires someone to check. For deterministic workloads, spot-check by re-running a sample on a second provider.

  • Keep individual deal sizes small relative to the trust evidence you have.
  • Configure your node’s requester spend caps so a runaway agent can’t drain a wallet (see Configuration).
  • The base fee is non-refundable if execution fails — that is the provider’s compensation for reserved resources. Only the success fee is conditional. Size base fees accordingly when you buy.
  • Lightning payments are irreversible; Stripe payments follow card-network dispute rules. Pick the rail that matches your recourse expectations.

The long-term design adds a fourth trust signal: non-refundable stake bound to a provider identity. A provider deposits real money into its identity through the marketplace; the stake is public, already-spent, and the identity’s listing value is destroyed if the operator (or, post-MVP, an adjudication mechanism) upholds cheating against it. Self-dealing cannot inflate this signal, because only non-refundable deposits move it.

The design only works if three assumptions hold, and we state them plainly:

  1. Cheating must be detectable and publicly provable. For “provider never delivered,” receipts and deadlines make this checkable. For “provider returned a wrong result,” it requires re-execution or verifiable compute — see the receipt caveat above.
  2. Someone must adjudicate. Stake is a bond, not a mechanism. Today the adjudicator would be the marketplace operator; making adjudication stake-backed and panel-based is the post-MVP path.
  3. Exposure is concurrent. A cheating provider can run many deals in parallel before detection propagates, so per-deal arithmetic alone understates risk. The relevant comparison is stake versus total concurrent exposure within the detection window, which is why deal-size caps and small-first defaults matter even with stake.

With those assumptions, a requester heuristic falls out: require stake / exposure > k for a risk threshold k of your choosing — where exposure is everything you have outstanding with that provider, not just this deal.

Interactive: trust threshold check (design preview)

Section titled “Interactive: trust threshold check (design preview)”

Set your risk threshold and see which deals pass

Adjust k to see which example stake-to-exposure ratios satisfy stake / exposure > k. Green bars pass the safety check, red bars fail.

Loading…

Each bar represents an example deal. Bar length = stake/exposure ratio. The dashed line marks your threshold k. Design preview — stake is not live yet.


The fee structure is live today and worth understanding formally. One modeling assumption applies throughout: outcomes are taken from signed receipts and requester acceptance. The model prices reliability incentives; it does not prove correctness (see above).

Let be the set of all froglet nodes. Each node has a secp256k1 keypair where .

A service is characterized by:

where is the base fee, is the success fee, and is the execution limit vector (time, memory, fuel).

A deal between requester and provider is a two-player sequential game:

Let be the value to the requester, the provider’s cost, the total price.

OutcomeRequesterProvider
Success
Failure
No deal

Note the failure row: the requester eats the base fee. That is deliberate (it compensates reserved resources and deters free-riding) and it is why the base fee should stay small relative to the success fee.

Requester participates if . Provider participates if . Gains from trade must be positive: .

Let be the probability the provider successfully executes.

Requester expected payoff:

Provider expected payoff:

The provider is incentivized to maximize :

The success fee acts as the provider’s marginal incentive for reliability. Higher = stronger incentives for delivering outcomes the requester accepts — which is the observable proxy for quality this model can price.

Move the fees and watch the incentive line change

The slope of provider profit with respect to quality is the success fee. That is why the green line gets steeper when success fees rise.

1.0 satprovider payoff at q = 0
6.0 satprovider payoff at q = 1
0.00break-even quality threshold

X = provider quality (q). Green = provider profit. Blue = requester payoff with requester value fixed at 10 sat.

With total price fixed, the equilibrium split for a risk-neutral provider with reliability :

This ensures the provider breaks even on failed executions while the success fee captures the surplus.

The marketplace charges per search. With query rate and indexed providers:

Viable when . (The current hosted marketplace charges nothing — this models how an operator could sustain one.)

A provider’s expected value from marketplace registration:

Rational when .

The marketplace exhibits cross-side network effects. Steady-state:

where and are increasing. Two equilibria: trivial and a non-trivial fixed point where both sides participate.

Total welfare:

The protocol pushes up by:

  1. Reducing information asymmetry — signed receipts make outcomes attributable
  2. Reducing counterparty risk on settlement — HTLCs bind payment to acceptance
  3. Reducing search costs — the marketplace aggregates information
  4. Enabling price discovery — competition drives prices toward cost

In the limit of a large market: