Crate scemadex_sdk 

ScemaDEX SDK — an Agentic Liquidity Layer

ScemaDEX is not “a DEX.” It is an SDK in which the routing intelligence itself is a metered, learning, accountable product. Autonomous agents solve swap intents, bond their promises, sell their inferences, and trade learned experience with one another — all settled in stablecoins over the x402 payment rails.

The four composing primitives

• A · Metered inference routing. Every quote is produced by a learning policy and can be sold per-call via x402. The quality of the decision is the SKU. See policy.
• B · Intent solving. Callers express what they want (intent::Intent with an intent::Objective), not a path spec. The policy decides venue / split / timing.
• C · Signal & reputation oracle. Reputation, pool scores and advice are monetized read endpoints. See oracle.
• D · Conviction Routing (the defining pillar). The policy escrows a slashable performance bond against its own promise. Meet the guarantee → reclaim the bond and collect the fee; miss it → the bond settles to the caller. This is what makes a paid black-box inference trustworthy. See bond.

The headline: a mesh of agents trading intelligence

Compose A–D across many nodes and you get the revolutionary endgame: a mesh::PeerMarket where agents sell bonded inferences and learned experience, and buy better ones from peers — an economy of machine intelligence, not a swap widget. An agent earns USDC selling what it knows and spends USDC to learn faster.

The adversarial layer: four primitives with no precedent

• E · The Counter-Market. Any agent can stake against an open bond (counter::CounterMarket) — the first market that prices individual machine inferences. The gap between self-conviction and market-implied conviction (the doubt spread) is a brand-new policy feature.
• F · The Scar Market. Slashed bonds are the only un-fakeable proof a decision cost real collateral; scar::certify_scar turns the trajectory behind a slash into sellable, verified negative training data.
• G · Experience royalties. lineage::LineageLedger records which purchased experience trained which policy, and streams a royalty slice of every downstream fee back to the sellers — training data as a yield-bearing asset.
• H · Bonded teaching. A teacher answers a student’s most uncertain states per metered query, bonding its tuition against the student’s measured improvement (teach::TeachingEngine) — distillation with a money-back guarantee.

Architecture: lean core, injected power

The published crate carries no solana-sdk or bot dependency. It defines the trait surface (policy::RoutePolicy, bond::BondEngine, venue::VenueExecutor, oracle::SignalSource, mesh::PeerMarket) plus reference implementations. Enabling the scematica feature injects the real Deep Q* agent, the Raydium/Orca/Meteora/Jupiter executors, and the x402 facilitator from the bot workspace.

use scemadex_sdk::reference_client;
let dex = reference_client();
let intent = scemadex_sdk::demo_intent();
let (solution, bond) = dex.quote(&intent).await?;
assert!(solution.route.splits_valid());
let _ = bond;

Re-exports

pub use bond::Bond;

pub use bond::BondConfig;

pub use bond::BondEngine;

pub use bond::BondLedger;

pub use bond::BondOutcome;

pub use bond::EscrowBondEngine;

pub use bond::NoBondEngine;

pub use client::ScemaDex;

pub use coordinator::DisputeCoordinator;

pub use coordinator::SettlementReport;

pub use counter::Challenge;

pub use counter::ChallengeSettlement;

pub use counter::CounterMarket;

pub use counter::CounterStats;

pub use error::Result;

pub use error::ScemaDexError;

pub use forecast::ForecastVenue;

pub use insurance::InsurancePool;

pub use insurance::InsuranceTerms;

pub use insurance::Payout;

pub use insurance::Policy;

pub use insurance::PremiumConfig;

pub use intent::Constraints;

pub use intent::Intent;

pub use intent::Objective;

pub use intent::Side;

pub use lineage::LineageEntry;

pub use lineage::LineageLedger;

pub use lineage::RoyaltySplit;

pub use mesh::ExperienceBatch;

pub use mesh::InferenceOffer;

pub use mesh::LocalPeerMarket;

pub use mesh::PeerMarket;

pub use oracle::Advice;

pub use oracle::PoolScore;

pub use oracle::Reputation;

pub use oracle::SignalSource;

pub use policy::Conviction;

pub use policy::ReferenceRoutePolicy;

pub use policy::RoutePolicy;

pub use policy::Solution;

pub use primitives::Address;

pub use primitives::Amount;

pub use primitives::Usdc;

pub use promise::Measurement;

pub use promise::Outcome;

pub use promise::Promise;

pub use route::Fill;

pub use route::Route;

pub use route::RouteLeg;

pub use route::Venue;

pub use scar::certify_scar;

pub use scar::LocalScarMarket;

pub use scar::ScarMarket;

pub use scar::ScarRecord;

pub use settlement::BondState;

pub use settlement::Clock;

pub use settlement::FinalizeReason;

pub use settlement::ManualClock;

pub use settlement::SettlementConfig;

pub use settlement::SettlementMachine;

pub use settlement::SlashDistribution;

pub use settlement::SlashRouting;

pub use settlement::SystemClock;

pub use teach::ReferenceTeacher;

pub use teach::TeachAnswer;

pub use teach::TeachReceipt;

pub use teach::TeachTerms;

pub use teach::Teacher;

pub use teach::TeachingEngine;

pub use venue::SimVenueExecutor;

pub use venue::SwapInstructions;

pub use venue::VenueExecutor;

pub use zkbackend::prove as prove_inference;

pub use zkbackend::DenseLayer;

pub use zkbackend::LayerShape;

pub use zkbackend::OpenedCell;

pub use zkbackend::SpotCheckConfig;

pub use zkbackend::SpotCheckProof;

pub use zkbackend::TracedMlp;

pub use zkproof::InferenceProof;

pub use zkproof::SNARK_SOUNDNESS;

pub use zksnark::ProvenModel;snark

pub use zksnark::SnarkConfig;snark

pub use zksnark::SnarkInferenceProof;snark

pub use zkbond::CommittableModel;

pub use zkbond::InferenceAttestation;

pub use zkbond::InferenceProofSystem;

pub use zkbond::ModelCommitment;

pub use zkbond::ReexecutionProofSystem;

pub use zkbond::VerifiedBond;

Modules

aiai

Natural-language intent parsing + route/bond narration via an LLM. Requires the ai feature. Natural-language front-end for ScemaDEX (feature ai).

bond

client

coordinator

Dispute coordination — wiring the adversarial primitives into finality.

counter

The Counter-Market — adversarial conviction staking (Primitive E).

error

forecast

The forecast venue — proof that the rail is not a swap widget.

insurance

Bond insurance / reinsurance — the hedge side of Conviction Routing (Primitive J).

integrationscematica

Concrete wiring of the SDK traits to the real Scematica Deep Q* agent. Only compiled with the scematica feature; the default published crate carries no such dependency. Bridge from the lean SDK traits to the real Scematica Deep Q* agent.

intent

lineage

Experience royalties — training data that pays dividends (Primitive G).

mesh

netnet

Networked PeerMarket client over an HTTP/JSON relay. Requires the net feature. Networked PeerMarket client (feature net).

oracle

policy

primitives

promise

Domain-agnostic bonding — the seam that unshackles the rail from swaps.

route

scar

The Scar Market — slash-certified negative knowledge (Primitive F).

settlement

Settlement v2 — the optimistic conviction-settlement state machine.

teach

Bonded machine teaching — distillation as a metered, accountable service (Primitive H).

venue

zkbackend

A succinct, transparent proof-of-inference backend — the real zk step behind Primitive I, the piece that lets a dispute window collapse toward zero.

zkbond

zkML-verified conviction bonds (Primitive I) — the capstone of the Adversarial Layer.

zkproof

The proof-of-inference role — the abstract contract every zk backend behind Primitive I fulfils, so the settlement layer never cares which proof system collapsed the dispute window.

zksnarksnark

A real zk-SNARK (arkworks Groth16 / BN254) proof-of-inference backend behind the zkproof::InferenceProof role. Requires the snark feature (heavy, default-off). A real zk-SNARK proof-of-inference backend — arkworks Groth16 over BN254, slotting in behind the same InferenceProof role as the transparent zkbackend, and finishing the job that module’s header names: “what a SNARK backend adds behind this same role.”

Functions

conviction_client

Build a ScemaDex whose bond engine is the conviction-weighted EscrowBondEngine — the lean reference wiring that actually exercises Conviction Routing (Primitive D) end-to-end, offline.

demo_intent

A canonical WSOL → USDC demo intent used by docs and tests.

reference_client

Build a ScemaDex wired with the lean reference implementations. Useful for examples, tests, and consumers who haven’t enabled the scematica feature.