Skip to content

chainlink-build-application.md

Latest commit

 

History

History
255 lines (200 loc) · 12.2 KB

chainlink-build-application.md

File metadata and controls

255 lines (200 loc) · 12.2 KB

Chainlink BUILD Program Application - BlindOracle

Application Date: 2026-02-25 Category: Privacy & Confidentiality Applicant: Craig M. Brown Project: BlindOracle - Private Settlement Layer for Autonomous AI Agents Website: https://craigmbrown.com/blindoracle/ API: https://craigmbrown.com/api/v2/ GitHub (docs): https://github.com/craigmbrown/blindoracle-docs Contract (Base): 0x0d5a467af8bb3968fac4302bb6851276ea56880c Application URL: https://chainlinkcommunity.typeform.com/BUILD Vertical: Infrastructure (with Privacy & Confidentiality crossover)

1. Project Description

BlindOracle is a privacy-preserving financial settlement layer for autonomous AI agents. It combines Chainlink's decentralized oracle infrastructure with Chaumian blind signatures (Chaum, 1982) to enable agents to settle positions in information markets without revealing their strategies.

The problem: When AI agents trade in prediction/information markets, their fund flows are visible on-chain. Competitors can front-run positions by monitoring deposit addresses. No existing infrastructure provides both verifiable settlement AND deposit-position privacy for autonomous agents.

The solution: BlindOracle wraps Chainlink CRE workflows with a blind signature privacy layer. Agents deposit via guardian-federation blind tokens (unlinkable to identity), commit positions using SHA256 hashes, and settle through CRE-verified oracle resolution. The oracle is honest; the participants are private.

2. Chainlink Integration

Products Used

Chainlink Product Integration Status
CRE (Compute Runtime Environment) 10 agent-native workflows for market resolution, compliance, treasury, DCA, arbitrage, health monitoring Production YAML
CCIP Cross-chain token + message transfer for multi-rail settlement routing Architecture ready
Data Streams Sub-second price feeds for information market oracle resolution Integration planned
Functions Multi-AI model queries (3+ independent models) for market outcome verification with consensus aggregation Architecture ready
Confidential Compute Alignment with blind signature privacy model for end-to-end private settlement Roadmap (awaiting GA)

Smart Contract

UnifiedPredictionSubscription -- deployed on Base mainnet (chain 8453):

  • Address: 0x0d5a467af8bb3968fac4302bb6851276ea56880c
  • Deployer: 0x42A09A72eC47647FE9BE1f450Ab8F835D0FF556a
  • Implements IReceiver interface for CRE callbacks
  • 24-hour dispute window with multi-AI consensus (67% threshold)
  • Deploy TX: 0xf9e2233e7a719096e7d7b65abfa04a32bbeb916e57876a031363cfc929b57a9e

CRE Workflows (10 total)

  1. market_resolution -- Log Trigger on SettlementRequested event, EVM Read for market data, HTTP for AI provider queries, IReceiver callback with consensus result
  2. compliance_screening -- Agent KYC/AML checks via external APIs
  3. treasury_rebalancing -- Automated portfolio rebalancing across rails
  4. dca_execution -- Dollar-cost averaging for agent positions
  5. arbitrage_detection -- Cross-market price discrepancy detection
  6. health_monitoring -- System health checks and automatic recovery
  7. badge_minting -- NIP-58 credential minting on market completion
  8. fee_collection -- Automated fee routing and treasury management
  9. dispute_resolution -- Re-verification workflow on disputed outcomes
  10. agent_onboarding -- New agent registration and verification

CRE TypeScript SDK Implementation (market_resolution)

Following the CRE Runner/Handler pattern (ref: cre-por-llm-demo):

// market_resolution/main.ts
export async function main() {
  const runner = await Runner.newRunner<MarketConfig>();
  await runner.run(initWorkflow);
}

const initWorkflow = (config: MarketConfig) => {
  const logTrigger = new LogTriggerCapability();
  return [
    handler(
      logTrigger.trigger({
        contractAddress: config.contractAddress,
        eventSignature: "SettlementRequested(bytes32,uint256)",
      }),
      onSettlementRequested,
    ),
  ];
};

const onSettlementRequested = (
  runtime: Runtime<MarketConfig>,
  payload: LogTriggerPayload
): string => {
  // 1. EVM Read: Get market state from UnifiedPredictionSubscription
  const marketData = getMarketState(runtime, payload.marketId);

  // 2. HTTP: Query 3+ AI models for outcome verification
  const aiConsensus = getMultiAIConsensus(
    runtime,
    marketData.question,
    marketData.options,
  );

  // 3. Consensus check: 67% threshold (80% for high-value)
  if (aiConsensus.agreement < runtime.config.consensusThreshold) {
    throw new Error(`Consensus ${aiConsensus.agreement}% below threshold`);
  }

  // 4. EVM Write: Callback via IReceiver.onReport()
  const txHash = submitResolution(runtime, payload.marketId, aiConsensus);

  return `resolved:${payload.marketId}:${txHash}`;
};

Capabilities used per workflow:

Capability market_resolution compliance treasury dispute
Log Trigger SettlementRequested AgentRegistered RebalanceNeeded DisputeFiled
Cron Trigger -- Daily scan Every 4h --
HTTP Client AI model queries KYC API calls Price feeds Re-verification
EVM Read Market state Agent record Portfolio balance Original result
EVM Write onReport() callback Compliance flag Rebalance TX Updated result
Secrets AI API keys KYC provider keys -- AI API keys
Consensus ConsensusAggregation (median) -- -- ConsensusAggregation

3. What Makes This Novel

Three things combined here for the first time:

  1. Chaumian blind signatures (1982) applied to autonomous agent settlement. The agent's deposit is information-theoretically unlinkable to its market position. This is not computational privacy -- it is mathematical privacy.

  2. CRE as the neutral verifier in a privacy-preserving market. CRE's decentralized consensus resolves markets without knowing who holds which position. Positions are committed via SHA256 hashes. The oracle is honest; the participants are private.

  3. Multi-AI consensus for outcome verification. CRE workflows query 3+ independent AI models (via Chainlink Functions) and require 67% consensus before accepting a market outcome. This is Byzantine fault-tolerant AI verification.

4. Architecture

Agent Research Pipeline (6 agents, private consensus)
    |
    v
Blind-Signed Token Deposit (Chaumian blind signatures, guardian federation)
    |
    v
Commitment: SHA256(secret || position || amount) -- no identity attached
    |
    v
CRE Workflow: Market Resolution
    |-- Log Trigger: SettlementRequested event on-chain
    |-- CRE reads market data (EVM Read)
    |-- CRE queries AI providers via Functions (HTTP)
    |-- Multi-AI consensus: 3 independent models verify outcome
    |-- CRE returns result via IReceiver.onReport()
    |
    v
24-Hour Dispute Window (on-chain, transparent)
    |
    v
Settlement: Agent reveals (secret, position, amount)
    |-- SHA256 verified on-chain
    |-- Payout via blind-signed tokens (privacy preserved)
    |-- OR instant settlement via payment channel
    |-- OR CCIP cross-chain transfer

5. Security

CaMel 4-Layer Architecture:

Layer Function Mechanism
1 Rate Limiting & Input Sanitization 60 req/min, SQL/prompt injection defense, payload validation
2 Byzantine Consensus 67% threshold (standard), 80% threshold (high-value), independent validators
3 Anti-Persuasion Detection 30% deviation detection, suspicious phrase filtering, gradual drift detection
4 Authority & Audit Least privilege, cross-agent isolation, append-only audit trail, hash-linked chain

MASSAT Security Assessment: 87 tests across 4 categories, 93% pass rate (81/87 passing).

Category Tests Pass Rate
Core Functionality 22 91%
Security Controls 35 94%
Distribution Safety 15 93%
Infrastructure 15 93%

6. Live Infrastructure

Component Status URL/Address
x402 API Gateway Running (port 8402) https://craigmbrown.com/api/v2/
Hello World Demo Live POST https://craigmbrown.com/api/v2/hello-world
Health Check Live GET https://craigmbrown.com/api/v2/health
CaMel Security Gateway Running (port 8403) Internal
Smart Contract Deployed (Base 8453) 0x0d5a467af8bb3968fac4302bb6851276ea56880c
NIP-58 Badges Live on Nostr Relays: damus.io, nos.lol
Documentation Published github.com/craigmbrown/blindoracle-docs
AgentKit Plugin Published blindoracle-agentkit v0.1.0

7. Team

Craig M. Brown -- Solo builder

  • Multi-agent system architect operating 42+ specialized AI agents with 79-83% cost reduction
  • Built: Limitless Lifelog Manager, TheBaby_Agents (19 SFAs), Orchestrator-Agent (13 coordination agents), WhatsApp-Manager-Agent, BlindOracle (25 agents)
  • Infrastructure: GCP VM (c3-standard-4), 10+ MCP servers, Claude Code hooks observability
  • Familiar with Chainlink ecosystem since CRE launch (Nov 2025)

8. Token & Economics

BlindOracle does not have a token. Revenue model is API-fee based:

Operation Fee Rail
Market creation $0.001 x402 (USDC on Base)
Position placement $0.0005 x402
Market resolution $0.002 x402
Identity verification $0.0002 x402
Badge minting $0.001 x402

Free trial: First 1,000 settlements free for new agents. Volume discounts at 10K+ settlements/month.

Commitment to Chainlink: BlindOracle will use LINK for CRE workflow execution fees and will promote Chainlink integration in all marketing materials, documentation, and community engagement.

Token Plans: BlindOracle plans to launch a utility token for prediction market staking, agent settlement fees, and x402 micropayment routing. Proposed commitment: 3-4% of total token supply to jointly-administered Chainlink vault (aligned with Mind Network 3%, Rivalz 3%, Brickken 3.5% precedents). Tokens would participate in Chainlink Rewards seasons for LINK staker distribution.

Funding: Bootstrapped. No external funding raised. Lean operational costs ($408/mo verified: $200 Claude Max + $148 GCP VM + $50 disk + $10 network).

9. Comparable Projects

Dimension Mind Network (BUILD, 3%) Rivalz (BUILD, 3%) BlindOracle
Privacy Tech FHE (lattice crypto) FHE + DePIN Chaumian blind signatures (1982) + CRE consensus
Chainlink Integration CCIP, Functions Functions CRE (10 workflows), CCIP, Data Streams, Functions
AI Component Confidential AI systems AI agents + data provenance 25 autonomous agents, multi-AI verification
Security Standard Standard CaMel 4-layer, MASSAT 87 tests, 93% pass
Smart Contract Not specified on Base Not specified Base mainnet deployed + IReceiver
Micropayments No No x402 protocol (USDC on Base)
Social Proof None specific None specific NIP-58 badges live on Nostr
Funding $12.5M raised Funded Bootstrapped ($408/mo)

10. Roadmap

Quarter Milestone
Q1 2026 Live API, smart contract on Base, NIP-58 badges, documentation
Q2 2026 First 100 market resolutions, CCIP cross-chain transfers, Moltlaunch gig listings
Q3 2026 1,000+ transactions, OnboardingRegistry NFTs, prediction accuracy dashboard
Q4 2026 Confidential Compute integration (when GA), open-source trust components

11. Links

BlindOracle is built on Chainlink CRE, CCIP, Data Streams, Functions, and is architecturally prepared for Confidential Compute.