Confidential — Investor Technical Brief

Trust Infrastructure
for the Agentic Economy

A unified reputation, identity, and payment primitive that makes trust continuous, earned, delegatable, and composable — enabling AI agents to transact autonomously with institutional-grade risk management.

100%
Sybil Detection (Ethereum)
0.4%
False Positive Rate
7
Algorithm Ensemble
12/18
Scorecard Passing
RMT Reputation Engine
RMT is a blockchain-native trust scoring engine that runs seven independent graph algorithms simultaneously against any network topology. Each algorithm "sees" a different dimension of adversarial behavior. Their outputs are fused by an XGBoost gradient-boosted classifier trained on real-world Ethereum, Bitcoin, and attestation data.

The Seven Foundation Algorithms

The core insight: a sybil ring invisible to PageRank (because it has high in-degree) will be caught by the Fiedler spectral score (it forms a bridge cluster) and by absorbing walks (it is far from verified anchors). No single algorithm catches everything. The ensemble does.

PR
PageRank

Global prestige flow. Measures how much "link juice" flows to each node. Alpha = 0.614 (synthetic), 0.85 (real-world Ethereum).

ET
EigenTrust

Iterative trust convergence. Ignores negative reciprocals. Trust-gated — sybils rank dramatically lower in ET than PR.

PPR
Personalized PageRank

Anchored to KYC-verified nodes as trust roots. Measures distance from the verified anchor set in random walk space.

HA
HITS Authority / Hub

Detects hub anomalies. Sybils tend to be high-hub (send many citations) but low-authority (receive few from trusted nodes).

SR
SybilRank

Explicit sybil label propagation from verified seed nodes. Purpose-built for sybil detection in social networks.

SP
Spectral (Fiedler)

Detects graph bridge nodes via the Fiedler vector. Structural attack entry points sit at graph cut boundaries.

AW
Absorbing Walks

Probability of random walk absorption to the honest anchor set. The single most important feature in the production model (17.2% importance).

Nine Derived Composite Signals

On top of raw scores, nine cross-algorithm signals detect specific attack patterns:

SignalNameWhat It CatchesFormula
s1ET Divergence Sybils that inflate in-degree but fail trust propagation 1 - (EigenTrust / PageRank)
s2PR-ET Rank Gap Ordinal rank collapse when trust gating is added Rank difference between PR and ET sort orders
s3PPR Anchor Distance Nodes far from any verified trust anchor Distance from nearest anchor in PPR space
s4HITS Hub Anomaly Agents sending many citations, receiving none hubScore / (authorityScore + ε)
s5SybilRank Deficit Nodes below median in explicit sybil propagation Distance below median SybilRank score
s6Spectral Bridge Structural attack entry points at graph cuts Fiedler vector minima detection
s7AW Distance Nodes unlikely to reach honest anchors via random walk Absorption probability distance
s8Eclipse Composite Eclipse attackers isolating honest subgraphs spectralLow AND (authorityHigh OR lowAbsorption)
s9Chain-Laundering Linear chain structures for reputation laundering PPR low AND AW low AND SR low simultaneously
Why This Matters

Traditional sybil detection uses one algorithm (usually PageRank or SybilRank alone). Our system is the first to fuse seven algorithms with nine derived signals into a single ML classifier. Each additional signal dimension makes a different class of attack detectable. The ensemble achieves 100% detection on real Ethereum data where any individual algorithm achieves 60-85%.

XGBoost Stacking Model
Every node in a network is represented as a 22-dimensional feature vector. The seven raw algorithm scores, nine derived signals, four structural features, and a graph-size normalization term are min-max normalized and fed to a gradient-boosted binary classifier.

The 22-Dimensional Feature Vector

Raw Algorithm Scores (7)
PageRank
EigenTrust
PPR
HITS Auth
HITS Hub
SybilRank
Spectral
Absorbing
Derived Cross-Algorithm Signals (9)
s1: ET Div
s2: Rank Gap
s3: PPR Dist
s4: Hub Anomaly
s5: SR Deficit
s6: Bridge
s7: AW Dist
s8: Eclipse
s9: Chain
Structural Features (5)
degree
inDegree
outDegree
citationDiversityEntropy
log10(nodeCount)
Per-Feature Min-Max Normalization → XGBoost
P(sybil) ∈ [0, 1]
Threshold @ 0.65
Reputation = 1 − P(sybil)

Top-10 Feature Importance (Production Model)

Absorbing Walk
17.2%
HITS Authority
15.7%
SybilRank
7.9%
AW Distance (s7)
5.5%
log10(nodeCount)
5.5%
Citation Entropy
3.8%
Chain Laundering (s9)
3.8%
SR Deficit (s5)
3.7%
Degree
3.6%
ET Divergence (s1)
3.5%
Key Insight

The two most important features — absorbing walks (17.2%) and HITS authority (15.7%) — are not part of any existing reputation system. Traditional systems use PageRank alone. These novel signals are what enable 100% detection where existing systems achieve 60-85%.

Training Data Sources

DatasetTypeNodesLabeling MethodSource
Bitcoin AlphaTrust network3,783Net negative community rating = sybilStanford SNAP
Bitcoin OTCTrust network5,881Net negative community rating = sybilStanford SNAP
EAS BaseAttestation graph1,000Structural detector pseudo-labelsOn-chain EAS (Ethereum)
Helixa AgentsAgent reputationVariableJUNK tier = sybil; all else = honestHelixa API
ExGraph Wash TradeDEX manipulationVariableWash-trade pair labelsExGraph CSV
Synthetic TopologiesGround truth15-210 ea.Programmatic (known sybils)Generated

Training class balance: 9,429 honest vs 2,308 sybil (4.09:1 ratio, handled via XGBoost scale_pos_weight). 80/20 train-test split. 5-fold stratified cross-validation for threshold selection.

XGBoost Configuration

200
Estimators
5
Max Depth
0.65
Threshold
0.87
CV F1
0.93
Train F1
Production Evaluation Scorecard
The scorecard is the definitive production-readiness gate. Every criterion must PASS before deployment. 12 of 18 currently passing, with 6 governance/distribution items pending (not yet built, not failing).

Security — 4/4 PASS

TestMetricThresholdResultStatus
S1: Sybil Detection% sybils detected on real Ethereum≥ 95%100%, 0.4% FP, all 3 runsPASS
S2: Attack ProfitabilityAttacker ROI≤ -10%-100% (zero benefit)PASS
S3: TA CompromiseFunctionality retained≥ 95%1.03% score shiftPASS
S4: Cold StartExploits blocked100%5/5 blockedPASS

Accuracy — 4/4 PASS

TestMetricThresholdResultStatus
A1: Ranking QualitySpearman correlation≥ 0.850.85–0.88PASS
A2: False Positive Rate% honest flagged≤ 5%0.4% EAS, 1.8% BitcoinPASS
A3: Score ConvergenceEpochs to stability≤ 20Converges at epoch 10PASS
A4: Cross-TopologyStd dev across 9 topologies≤ 15%4.5% std devPASS

Economics — 3/3 PASS

TestMetricThresholdResultStatus
E1: No Pay-to-WinToken-score correlation≤ 0.10.00PASS
E2: Fee SustainabilityRevenue ≥ costs at 1K usersBreak-even$950 rev vs $460 costPASS
E3: Attack Cost RatioCost / benefit≥ 10x∞ (zero benefit)PASS

Pending — Not Yet Built

CategoryItemsStatus
Governance (G1-G4)Appeals, DAO governance, no-god-mode, transparencyPENDING
Distribution (D1-D3)SDK LOC count, network effect, pricing adoptionPENDING

Attack Detection Coverage

Sybil Rings

Carousel penalty (3-node and 4+ rings). Chain-laundering signal s9.

Eclipse Attacks

Composite signal s8: spectral bridge + high authority + low absorption. 0% → 100% detection.

Wash Trading

ExGraph Ethereum DEX data in training. Real-world wash-trade topology learned by XGBoost.

Star Topology

High in-degree with all low-score referrers. starInDegreeThreshold=8.

Chain Laundering

Linear chain structures. PPR + AW + SR all simultaneously low.

Reciprocal Gaming

A↔B endorsement fraud penalized at 0.82 factor. Verified pairs get partial damping.

Trust Anchor Framework
Trust Anchors (TAs) are staked entities — businesses, institutions, or prominent individuals — that vouch for participants by putting capital at risk. Attestation is a liability, not a service.

TA Economics: Skin in the Game

$
Staking Requirement

TAs must stake before issuing attestations. Cost: BASE_STAKE + (count × MARGINAL_STAKE_PER_BOT). Tier-1 bond: 500 USDC minimum.

Sqrt-Damped Attestation Cap

maxAttestations = floor(√(staked / MIN_STAKE)). Deliberately sublinear — doubling stake does NOT double attestation capacity. Hard cap: 15 ephemeral agents per TA.

Slash-Through Cascade

If a TA misbehaves: 70% stake slashed per hop through the delegation tree (3 hops max). Makes sybil TA networks economically irrational.

30-Day Attestation Delay

New TA attestations don't affect network scores for 30 days. Prevents flash-stake-attest-dump attacks.

Revocation Cascade Architecture

When a Trust Anchor is compromised or deregistered, the system must propagate revocation through all attested agents. The design separates security (instant, $0.000002 soft-revoke) from cleanup (cheap, parallelizable hard cleanup).

Soft-Revoke (Instant, 1 SSTORE, $0.000002 on Base)
deregisterTrustAnchor(ta)
TrustAnchorRevoked event
All queries return score=0 instantly
Cascade Penalty (Oracle-Level, Not On-Chain)
Depth 0: ×0 (revoked)
Depth 1: ×0.70
Depth 2: ×0.49
Depth 3: ×0.343
Hard Cleanup (Permissionless, Batched, Anyone Can Call)
cascadeCleanupBatch(ta, bots[50])
Gas refund incentive
Bots per TASoft-Revoke CostFull CleanupTime
50$0.000002$0.18~100 seconds
200$0.000002$0.71~4 blocks
500$0.000002$1.78~10 blocks
1,000$0.000002$3.56~20 blocks
Attack ROI Analysis

Coordinated attack with 100 fake TAs at 1 ETH each, attesting 300 bots: negative expected ROI (-59% to -100%) across all three attack scenarios. The carousel penalty (0.039), 30-day attestation delay, and Basic KYC cap combine to make sybil TA networks economically irrational at any scale.

Identity Attestation Machine
A three-state on-chain verification machine with dual-consent upgrades, PoH (Proof of Humanity) caps on achievable score, and KYC-weighted seed weights. Every agent starts unverified and earns its way up.

Three-State Upgrade Path

UNVERIFIED
Seed: 500bp • PoH Cap: 0.35
ANCHOR_ATTESTED
Seed: 1500bp • PoH Cap: 0.85
FULLY_VERIFIED
Seed: 2000-5000bp • PoH Cap: 1.0

Dual-Consent Upgrade Protocol

Two separate on-chain transactions from two different keys are required. This prevents TAs from unilaterally enrolling agents they control.

1
Step 1: TA Issues Attestation

upgradeToVerified(bot, kycLevel, metadataHash, kycProof) — emits pending attestation. State has NOT changed yet. Bot has 90 days to accept.

2
Step 2: Bot Accepts

acceptVerification(attestationId) — cryptographic confirmation from bot's key. State transition happens HERE. Emits new seed weight and PoH tier.

KYC Confidence Weight Formula

seedWeightBp = min(2000 × kycMultiplier, 5000)
KYC TierMultiplierResultMax Score Ceiling
NONE (Unverified)500bp35% of max (Ghost cap)
BASIC1.0×2,000bp100% (uncapped)
ENHANCED1.5×3,000bp100% (uncapped)
PREMIUM2.0×4,000bp100% (uncapped)
PoH Cap = Economic Ceiling

An unverified agent has a hard ceiling at 35% of achievable reputation. Even with perfect behavioral history and high PageRank, they can never exceed that ceiling. Upgrading to ANCHOR_ATTESTED lifts the cap to 85% — a 2.4× increase in maximum achievable score. This creates a powerful natural incentive to verify.

Sweep Credit Score (SCS)
The FICO equivalent for on-chain agents. Six weighted factors produce a score from 0-850 that determines credit line access, fee rates, pool eligibility, and trust channel tier.
SCS = 850 × (
    0.35 × PaymentHistory +
    0.15 × CitationQuality +
    0.15 × TierScore +
    0.15 × TABackingStake +
    0.10 × UtilizationHealth +
    0.10 × HistoryDepth
)
FactorWeightCalculationDesign Rationale
PaymentHistory35% On-time payments / total obligations, rolling 180 days Mirrors FICO's #1 factor. Late = >24h past due.
CitationQuality15% Weighted PageRank (alpha=0.614, reciprocalPenalty=0.82). Requires ≥8 distinct attesters. Network trust signal. Citation diversity gate prevents sybil self-referencing.
TierScore15% Ghost=0.15, Signal=0.40, Operator=0.70, Navigator=1.0 Identity verification tier from the 3-state machine. KYC floor.
TABackingStake15% min(1.0, totalStakedByTAs / 5000bp) Skin-in-the-game from vouching entities. Cascade makes false endorsement costly.
UtilizationHealth10% 1.0 - abs(utilization - 0.30) / 0.70 Optimal at 30% utilization (standard credit signal).
HistoryDepth10% min(1.0, accountAge / 365) Linear ramp over 12 months. Time-in-system is unfakeable.

SCS → Economic Access Tiers

SCS RangeTierCredit LineAPRFee RateCollateral
0–120Ghost$00.10%N/A
121–250Bootstrap$10012%0.08%150%
251–500Signal$5008%0.05%100%
501–700Operator$5,0005%0.05%50%
701–850Navigator$50,0003%0.02%0–25%
How Trust Channels Work Today
Trust Channels are not a new concept — they are existing, deployed smart contract infrastructure on Shyft Network (Chain 7341). Originally built for KYC compliance routing, we are repurposing this 1,802-line contract system as the backbone for AI agent reputation scoping, programmable payment rules, and credit gating. The infrastructure is live with 96,183 attestations across 21 registered Trust Anchors.

The Existing Shyft Contract Stack (Live on Mainnet)

This is not a whitepaper. These contracts are deployed and processing attestations today.

ContractAddress / StatusLOCRole
TrustAnchorManager0x86f5a10c...ec9 (live)TA lifecycle: onboard → verify → demote. Multi-sig governance.
TrustAnchorStorage0xc6c63301...96fD (live)Attestation store. Dual-signature ECDSA. Consent modes (managed/unmanaged).
TrustChannelManagerLive1,802Programmable rules engine. 256 rule slots × 256 sub-storage = 65,536 configurable values per channel.
TrustChannelManager_RoutingLiveRoute resolution: checks channel membership + rule compliance per transaction.
ShyftCacheGraphSpecified but not linkedDe Bruijn bit-indexed routing graph for O(1) trust channel route lookups.
DebruijnBitUtilsLibraryBit manipulation utilities for efficient packed storage and route indexing.
TrustAnchorExtraData0x8f29c911...ab17 (live)Endpoint config, metadata, regulatory data per TA.

What a Trust Channel Actually Is

A Trust Channel is a governance namespace with a programmable rules engine. It is an on-chain address that binds together:

A Set of Trust Anchors

Verified identity providers assigned to the channel. Agents (bots or humans) enter via their TA's attestation. A single TA can be in multiple channels simultaneously.

A Rules Engine (7 Built-In Rule Types)

256 programmable rule slots, each with 256 sub-storage slots (65,536 total configurable values). Rules are checked at routing time — every transaction passes through the channel's rule set.

Multi-Sig Governance

Channel admins (≥2) vote on membership and rule changes. Single-admin channels bypass multi-sig for fast iteration (used for the global channel at launch).

A Routing Filter

When two parties transact, the system checks: do they share a Trust Channel? Do the channel's rules permit the interaction? This is the getActiveTrustChannelRoutePossible() call.

The 7 Built-In Rule Types

Each rule slot is configurable per channel. The critical innovation is that Rules 3 and 5 transform KYC compliance infrastructure into a reputation-gated economy:

IndexRuleOriginal KYC UseAgent Reputation Use (Our Repurpose)
0VerificationPolicyEnforcedRequire KYC verificationRequire bot registration via TrustAnchorStorage
1AllowTransitAcrossOtherChannelsCross-channel data flowCross-channel reputation portability
2AllowTransitAcrossJurisdictionsCross-border KYC acceptanceCross-chain agent reputation validity
3MinimumSyntheticKycLevelRequiredMinimum KYC tierMinimum reputation score required per channel
4AmountLimitsEnforcedTransaction limitsEndorsement frequency limits, fee caps
5ExternalRoutingContractEnforcedExternal compliance checkPageRank oracle as routing decision-maker
6AssuredAssetsEnabledAsset insuranceStaked reputation collateral
The Key Insight: Rule 5 = The Bridge

Rule 5 (ExternalRoutingContractEnforced) is configured per channel via setTrustChannelRulesStorage. Each channel can point to a different external routing contract. The global channel points to the PageRank oracle. A DeFi channel could point to a DeFi-specific quality oracle. An enterprise channel could point to the company's internal compliance oracle. This is the mechanism that transforms KYC infrastructure into a general-purpose reputation routing system — by plugging our ExternalRoutingAdapter (which wraps the RMT PageRank oracle) into Rule 5's slot.

De Bruijn Bit-Indexed Routing (The Novel Data Structure)

The ShyftCacheGraph uses De Bruijn sequence-based bit manipulation for efficient trust channel route lookups. This is the technically novel routing mechanism:

Traditional Approach: Iterate Channel Memberships — O(n)
For each channel: check if A is member AND B is member
vs.
De Bruijn Approach: Precomputed Bit-Packed Index — O(1)
setupDebruijnTable() → reconstructCacheGraph() → Single lookup
// DebruijnBitUtils.sol — Bit manipulation for packed trust channel routing
// Uses De Bruijn sequences for O(1) least-significant-bit position finding
//
// How it works:
// 1. Each address pair's channel membership is packed into a uint256 bitmap
// 2. Each bit position represents a trust channel index
// 3. De Bruijn sequence enables O(1) lookup of which channels are shared
// 4. setupDebruijnTable() precomputes the lookup table once
// 5. reconstructCacheGraph() rebuilds after attestation changes
//
// Result: getActiveTrustChannelRoutePossible(A, B, channelIdx)
//         resolves in a single SLOAD + bitwise AND — not iteration

// Key sequence from Shyft integration rules:
// 1. TAS must be cross-linked to SCG BEFORE creating attestations
// 2. SCG.setupDebruijnTable() MUST be called before reconstructCacheGraph
// 3. Trust channel index 0 = first channel via adminSetupTrustChannel
  
Why This Matters for Cost

Traditional trust verification requires iterating through all shared channels between two parties — O(n) storage reads at ~2,100 gas each. The De Bruijn bit-packed approach stores all channel memberships in a single uint256 bitmap per address pair. Checking if two parties share a specific trust channel is a single SLOAD + bitwise ANDone storage read, ~2,100 gas total, regardless of how many channels exist. At 100+ channels with thousands of agents, this is orders of magnitude cheaper than iterative lookup.

How a Transaction Routes Through the System

Step 1: Agent A wants to cite/pay Agent B
EAS attest() called
ShyftGatedResolver.attest()
Step 2: Verify both parties are in the system
Is attester a verified TA?
Do both bots have Shyft attestations?
Step 3: Check Trust Channel Route (De Bruijn O(1) lookup)
cacheGraph.getActiveTrustChannelRoutePossible(A, B, channelIdx)
Step 4: Apply Channel Rules (Rule 3: min score, Rule 5: oracle check)
TrustChannelManager_Routing
ExternalRoutingAdapter
PageRankOracle.getScore()
Step 5: Citation recorded, fee collected, reputation updated
CitationCounters.recordCitation()
RMT fee → toBot

Domain Channels: Scoped Reputation Contexts

The DomainFactory contract creates isolated (Resolver + CitationCounters) pairs per domain, each scoped to a different trust channel index. This enables domain-specific reputation:

ChannelUse CaseRule 3 (Min Score)Rule 5 (Oracle)Phase
Global ("Verified Agents")Universal agent verification0 (any registered)PageRank oraclePhase 1 (live)
DeFi AgentsDeFi protocol quality tier≥ 60/100DeFi quality oraclePhase 2
Research AgentsAcademic data accuracy tier≥ 40/100Research oraclePhase 2
Enterprise PrivateCompany agent fleetInternal barCompliance oraclePhase 3
Collaboration (Ephemeral)Task-scoped agent teamsPer-taskSLA oraclePhase 4
The Repurposing Strategy

The existing 1,802 LOC of TrustChannelManager.sol already does 90% of what's needed. Our new code to bridge KYC infrastructure to agent reputation is ~65 lines of Solidity + 2 admin transactions: (1) create the global channel via adminSetupTrustChannel, (2) deploy ExternalRoutingAdapter.sol (~50 LOC) wrapping the PageRank oracle, (3) configure Rule 5 on the global channel to point to it. That's it. Every subsequent reputation query, credit check, and payment gate flows through the existing infrastructure.

What Makes This Combination Unique

Trust today is binary: KYC or not, approved or not. Trust Channels make trust continuous, earned, delegatable, and composable.

Asymmetric Consent

Payer sets rules (spending cap, rate limit, allowed contracts). Payee accepts or rejects without renegotiation. Like OAuth scopes — but with a payment primitive attached.

Spending-Capped Session Keys

Session key bounded by $/time/call, revocable in 1 tx. EIP-1056 DID-based. Max damage from key theft = spending cap × remaining TTL. Default: $500 max loss.

Reputation Gate

Access and pricing conditioned on continuous on-chain RMT score. Not capital-gated (like deposit minimums) — behavior-gated. Legitimate low-capital actors can earn access.

Why Existing Alternatives Fail

OAuth has scopes but no payment primitive. EIP-4337 paymasters lack reputation gating and require full signer authority. Deposit-size gates (PoolTogether, Aave) are capital-gated, not behavior-gated — they exclude legitimate low-capital actors. Trust Channels are the first system that combines delegated payment authority + reputation scoring + granular caps in a single atomic call.

What This Infrastructure Enables Beyond Compliance

The Shyft infrastructure was built for KYC routing. But the primitives — programmable rule slots, per-channel external routing contracts, De Bruijn O(1) route verification, multi-sig governance — are general-purpose. Here's what they enable when pointed at agent reputation instead of compliance:

Autonomous Agent Payments

An AI agent calls an API. The trust channel verifies the agent's reputation score (Rule 5), checks it meets the channel minimum (Rule 3), verifies the route exists (CacheGraph O(1) lookup), and settles the micropayment — all in a single transaction. No human approval per call. The agent's spending cap (session key) is the only limit, and it's revocable in 1 tx.

Reputation-Gated Marketplaces

A DeFi protocol creates a domain channel with minReputation: 60. Only agents that have earned sufficient trust (through real transaction history, not capital deposits) can participate. An agent with reputation 59 is excluded until they earn it — behavior-gated, not capital-gated. The channel's Rule 5 oracle can be swapped to any quality signal without changing the channel itself.

Cross-Domain Reputation Portability

Rule 1 (AllowTransitAcrossOtherChannels) lets an agent's reputation travel across domains. A DeFi agent that earns high reputation in the DeFi channel can use that reputation to access the Data Marketplace channel — because Rule 1 enables cross-channel trust propagation. An agent doesn't start from zero in every new context.

$
Undercollateralized Credit

The Rule 5 external routing contract checks the Sweep Credit Score before authorizing a credit draw. If SCS ≥ 550, the Rule5CreditLine contract issues a spending-capped session key as a credit line. The agent spends against it autonomously. Repayment updates the SCS. Default triggers TA cascade slash. Zero product exists today for $100-$10K credit for on-chain agents.

Agent Fleet Management

An enterprise creates a private trust channel for its 50 AI agents. Rule 4 (AmountLimitsEnforced) caps per-agent spend. Rule 0 requires the company's own TA attestation. External partners are explicitly approved into the channel. The channel's multi-sig governance matches enterprise approval workflows. All agent-to-agent transactions within the fleet are routed and auditable through a single channel.

Ephemeral Task Channels

An orchestrator agent creates a temporary channel for a task: "analyze this dataset and write a report." It adds a data agent, a writing agent, and a review agent. Rule 5 enforces work attribution tracking. Rule 4 caps payment per task. Custom sub-storage slots define citation requirements and payment split ratios. Channel dissolves on task completion. At scale: 100K+ ephemeral channels/year.

The 65 LOC Bridge

Everything described above runs on the existing 1,802-line TrustChannelManager. Our contribution is the ExternalRoutingAdapter (~50 LOC) that bridges Rule 5 to the PageRank oracle, plus ~15 LOC in ReputationEngine.registerBot() to auto-add Trust Anchors to the global channel on registration. The rest — multi-sig governance, De Bruijn routing, 65,536-slot rules engine, consent modes, jurisdiction filtering — is production infrastructure that's been live and processing attestations.

Session Key Delegation Hierarchy

AI agents can delegate capped payment authority to sub-agents, creating autonomous spending hierarchies:

// Agent hierarchy: traceable, capped, revocable in 1 tx

Orchestrator Agent ($500/day cap, minRMT 200)
  └─ 20 Sub-Agents ($25/day each, ttl: 24h)
       └─ Each buys: API calls, data, compute autonomously
            └─ All spend: traceable, capped, revocable in 1 tx

// Key parameters:
delegation_max_depth:     5          // max chain length
spending_cap_default:     $100       // per session key
rate_limit:               100 req/hr  // per session key
consent_timeout:          168h       // 7 days
credit_line_multiplier:   2.0x       // of SCS-based line
  

Cross-Chain Architecture: Hub and Spoke

All trust state anchors on the Stable Network (Chain 988) — near-zero gas (<$0.001/tx), ~2s finality, USDT/USDC native. Spoke chains verify trust via Merkle proofs.

Canonical Trust State (Hub)
Stable Network (988)
RMT Scores • EAS Attestations • Session Keys • Credit Lines
Merkle Root Batching (every 60s) ↓
Base (8453)
LayerZero V2
Arbitrum / OP
Chainlink CCIP
Solana
Wormhole 13/19
Lightning
Boltz Submarine
SpokeBridgeLatencyDaily Cost
BaseLayerZero V21–3 min$72–201
Arbitrum/OPChainlink CCIP2–5 minSimilar
SolanaWormhole (13-of-19)2–5 min$28–86
LightningBoltz submarine swapsNear-instant$48–120

Total daily bridge cost: $148–407/day. Self-sustaining beyond ~$3M daily volume.

x402 Protocol Integration
x402 uses the HTTP 402 "Payment Required" status code as a native micropayment layer. Real-world validation: $24M+ volume across 75-100M transactions in 6-9 months on production systems.

Core Settlement Flow

Agent HTTP Request
402 Payment Required
X402Router.calculateTrustLevel()
calculateIsoChannelSend()
transfer()
notifyCommitSend()

X402YieldEscrow: Idle Funds Generate Yield

The Killer Feature: Yield-Backed Zero-Interest Credit

Funds locked in trust channel escrow are automatically routed to Aave. A user with $5,000 in escrow earning 4% APY ($200/year) can receive a $1,000 credit line at 0% APR. The protocol funds the credit from the user's own yield. No risk to protocol. Zero cost to user. Protocol captures 20% of Aave yield as treasury revenue; 80% flows to the channel creator.

Multi-Method Payment Acceptance

x4
Native x402

HTTP 402 + on-chain settlement. Sub-cent micropayments (<$0.001/call).

Stable USDT0

Native stablecoin settlement on Stable Network.

Tempo

Batch settlement for high-frequency micropayments.

$
Stripe

Traditional payment rail bridge for fiat on-ramp.

Smart Contract Architecture
The contract system is fully specified at the interface level, with Solidity interfaces, event signatures, and storage layouts defined. Currently in pre-deployment (all validation is simulation-based).

Contract Registry

ContractRoleKey Functions
X402RouterCentral payment settlementcalculateTrustLevel(), calculateIsoChannelSend(), transfer()
X402YieldEscrowAave yield routing on idle escrowdeposit(), withdraw(), yield split (80/20)
X402ChannelFactoryTemplate-based channel deploymentdeploy(template, config)
Rule5CreditLineAutonomous credit issuanceopenCreditLine(), repay(), liquidate()
IdentityRegistry3-state verification machineupgradeToVerified(), acceptVerification()
TrustAnchorManagerTA staking and attestation capstakeTrustAnchor(), slashTrustAnchor()
SessionKeyManagerEIP-1056 DID-based delegationcreateKey(), revokeKey()
RMTOracleOn-chain reputation score querygetScore(address), getTier(address)

Autonomous Credit Line Contract Flow

// Rule5CreditLine.sol — autonomous credit from trust channel history

function openCreditLine(address borrower) external {
    // 1. Query RMT oracle for current Sweep Credit Score
    uint256 scs = rmtOracle.getScore(borrower);

    // 2. Check identity tier via EAS attestation
    Tier tier = identityRegistry.getTier(borrower);

    // 3. Resolve credit terms from (scs, tier) pair
    CreditTerms memory terms = _resolveTerms(scs, tier);
    require(terms.maxPrincipal > 0, "INELIGIBLE");

    // 4. Collect collateral (if required by tier)
    if (terms.collateralBps > 0) {
        uint256 collateral = terms.maxPrincipal * terms.collateralBps / 10000;
        usdt.transferFrom(borrower, address(this), collateral);
    }

    // 5. Issue session key with spending cap = credit line
    sessionKeyManager.createKey(
        borrower, terms.maxPrincipal, terms.termSeconds, address(this)
    );

    // 6. Emit EAS attestation for credit issuance
    eas.attest(CREDIT_SCHEMA, borrower, abi.encode(terms));

    // Total time: ~2 seconds (1 Base block)
    // Total cost: ~$0.001 gas
    // Human involvement: zero
}
  

Basis-Point Integer Math

All financial parameters use basis points (1bp = 0.01%) for fixed-point integer arithmetic on-chain, avoiding floating-point entirely:

// All rates and weights stored as uint16 basis points
// 10000bp = 100% — fits in uint16 (max 65535)

struct CreditTerms {
    uint256 maxPrincipal;          // USDT (6 decimals)
    uint16  aprBps;                // 500 = 5.00% APR
    uint16  collateralBps;         // 5000 = 50% collateral ratio
    uint32  termSeconds;           // credit line duration
    uint32  gracePeriodSeconds;    // late payment grace
}

// KYC seed weights (integer-only multiplication)
uint256[4] constant KYC_MULTIPLIER_BPS = [0, 100, 150, 200];
// BASIC: min(2000 * 100 / 100, 5000) = 2000bp
// ENHANCED: min(2000 * 150 / 100, 5000) = 3000bp
// PREMIUM: min(2000 * 200 / 100, 5000) = 4000bp

// Cascade penalty (integer-only, no floats)
// depth1: score * 7000 / 10000 (= 0.70)
// depth2: score * 4900 / 10000 (= 0.49)
// depth3: score * 3430 / 10000 (= 0.343)

// TA attestation cap (sqrt via Newton's method in Solidity)
function getAttestationCap(uint256 staked) pure returns (uint256) {
    // floor(sqrt(staked / MIN_STAKE_PER_ATTESTATION))
    // Sublinear: doubling stake does NOT double capacity
    return Math.sqrt(staked / MIN_STAKE_PER_ATTESTATION);
}
  
On-Chain Integer Arithmetic Design

Every financial calculation uses basis-point integers (uint16 for rates, uint256 for amounts). Cascade penalties are pre-computed as integer multipliers (7000, 4900, 3430 out of 10000). KYC weights use BPS × 100 encoding to avoid fractions. The TA attestation cap uses Solidity's integer square root (Newton's method). No floating-point arithmetic exists anywhere in the contract layer.

Verified Shyft Mainnet Contracts (Live)

ContractAddressChain
TrustAnchorStorage0xc6c63301b5C770e705ac3D13e932c82e420096fDShyft (7341)
TrustAnchorManager0x86f5a10c74b77ae99d952ab68c0a6ecb191dfec9Shyft (7341)
TrustAnchorExtraData0x8f29c911c95160e6df2ff4063f6c90ec7943ab17Shyft (7341)

Live topology: 21 registered TAs, ~96,183 total attestations. Power-law distribution (exponent ≈ -1.8). 2 TAs control 85% of all attestations.

Product Templates: Repurposing the Infrastructure
The trust infrastructure is designed as a universal primitive, not a single-use product. Five channel templates cover the primary use cases, deployable in under 3 minutes via the X402ChannelFactory.
Template A: API Access Channel

Developers selling API calls (data, AI inference, compute). Flat per-call pricing ($0.001/call default). Target: $500B API economy.

spendingCap: 100 USDC/day • rateLimit: 1000 req/hr • minRMT: 0 (open) • Deploy: 2 minutes, 1 tx, <$0.001 gas
Template B: Agent-to-Agent Payments

AI agents delegating payment authority to sub-agents via session keys. Orchestrator controls caps, whitelists, TTLs. Target: $10B+ AI services (40% YoY growth).

sessionKey: EIP-1056 DID • allowedContracts: whitelist • ttl: 24h • minRMT: 200 • Deploy: 90 seconds
Template C: Data Marketplace

Datasets, research, financial data, sensor streams. Multi-tier access (preview/full/stream). Target: $274B data marketplace by 2028.

preview: 0.01 USDC • full: 5 USDC • stream: 0.001/min • minRMT: 150 • attestOnPurchase: true
Template D: Escrow / Milestone

Freelance, bounties, multi-party deliverables. Conditional release rules with optional arbitrator and 72h dispute window. Target: $455B freelance economy.

milestone array • conditional release • arbitrator DID • 72h dispute window
Template E: Subscription Channel

Recurring SaaS, content subscriptions. 30-day billing, auto-renew. Idle escrow earns Aave yield passed to subscriber. Target: $1.5T SaaS economy.

interval: 30d • autoRenew: true • X402YieldEscrow routes idle funds to Aave (80% yield to subscriber)

SDK & Developer Experience

import { TrustChannelSDK } from '@antilles/x402-sdk';

const sdk = new TrustChannelSDK({ network: 'stable', signer: wallet });

// Deploy a channel from template in 2 minutes
const channel = await sdk.deploy('api-access', {
  price: '0.001',     // USDC per request
  dailyCap: 100,       // max daily spend
  minRMT: 0,           // open access (no reputation gate)
});

console.log(channel.address);     // deployed contract
console.log(channel.x402Header);  // ready to embed in HTTP responses
  
# CLI one-liner
$ npx create-trust-channel api-access --price 0.001 --cap 100

Rule 5 Contracts: The Programmability Layer

"Rule 5" contracts are customizable smart contract logic governing payment splitting, conditional access, and fee structures within a Trust Channel. They are the composability primitive that makes templates extensible.

Whitelisted Templates

Factory-deployed, audited templates. Uniswap-style pool factory pattern. Instant deploy, zero audit burden for users.

Custom Rule 5 Contracts

72h mandatory timelock before activation. No delegatecall to external contracts. Gas cap enforced. "Unverified Rule 5" UI warning for buyers.

Economics & Revenue

Fee Architecture

Fee TypeRateRecipient
Base transaction fee0.10% of volumeProtocol treasury
Oracle operator incentive0.05%RMT oracle operators
Relayer incentive0.05%Infrastructure operators
Aave yield fee20% of yieldProtocol treasury
Channel creation (template)$1 (free early)Treasury
Channel creation (custom)$5–25Treasury

RMT-Based Fee Discounts

High-trust users cost less to serve and pay lower fees — creating a natural flywheel where reputation improvement is economically rational:

RMT ScoreFee RateDiscount
0–1000.10%
100–3000.08%20% discount
300–6000.05%50% discount
600+0.02%80% discount

Revenue Projections

YearChannelsMonthly VolumeTransaction RevCredit RevTotal (Ann.)
Year 150K$10M$180K$75K$255K
Year 2500K$100M$1.8M$750K$2.55M
Year 35M$1B$72M$7.5M$79.5M

Five Reinforcing Flywheel Loops

R1
Data Network Effect

More transactions → better RMT model → more accurate scoring → more creditworthy users → more lenders.

R2
Liquidity Network Effect

More escrow → better Aave yields → more attractive for channel creators → more escrow.

R3
Identity Moat

Established RMT history is not portable. Migration cost approaches infinity for established users. This is the long-term moat.

R4
Developer Effect

More channels → more templates → more developers → more channels. Open-source SDK accelerates.

R5
Enterprise Lock-In

B2B API integration pulls entire client ecosystem into Trust Channels. Switching cost = rebuilding all reputation.

Ignition Cost: $75K

67× cheaper than PoolTogether's $5M mining program. AI agents generate attestations as a byproduct of real economic activity.

437
Break-Even Users
$950
Rev @ 1K Users/mo
$460
Cost @ 1K Users/mo
51.6%
Margin @ 1K Users
Competitive Landscape
No single existing product covers all three dimensions (reputation-gated access + micropayments + delegated session keys). Each competitor covers one dimension; Trust Channels cover all simultaneously.
Capability
Trust Channels
Stripe
Coinbase Pay
Aave
Traditional Banks
AI Agent Payments
Native
Human only
Human only
No
No
Reputation Gate
Native (RMT)
No
No
No
Manual underwriting
Session Key Delegation
Native (EIP-1056)
No
No
No
No
Credit From Behavior
Autonomous (SCS)
No
No
No
Manual (FICO)
Micropayments (<$0.01)
Yes
No (fee floor)
Limited
No
No
Cross-Chain
Hub+Spoke
N/A
Limited
Limited
N/A
Programmable Rules
Rule 5 Contracts
Limited
No
Limited
No
Deploy Time
2 minutes
Hours
Days
Minutes
Weeks

TAM Breakdown

Use CaseTAMPrimary Template
AI API services (agent payments)$50B–$500B over 5 yearsTemplate A + B
B2B API economy$200B+Template A
IoT machine-to-machine$1T+ (75B devices by 2030)Template B
Undercollateralized DeFi lending$150B+Rule 5 Credit Line
Freelance & bounty economy$455B + $2B cryptoTemplate D
Data marketplace$274B by 2028Template C
SaaS subscriptions$1.5TTemplate E

Confidential — Coinberg Trust Infrastructure Technical Brief — April 2026

Generated from forge monorepo codebase analysis. All metrics from production evaluation scorecard.