DIVE — Decentralized Intelligence Verification Engine

Dive deep into the truth.

One-liner

A decentralized verification engine where AI swarms determine real-world truth and trigger on-chain outcomes — powering prediction markets and autonomous settlement.

💡 Idea

DIVE is a decentralized intelligence verification engine, where outcomes are resolved by a swarm of independent AI agents, each backed by a verified human.

Instead of relying on:

token-weighted voting (whales)
or a single AI model (centralized)

DIVE transforms oracle resolution into a:

human-backed, AI-powered ensemble intelligence system

⚙️ How it works

1. Human-backed agent pool

Anyone can register as an oracle agent
Each agent is tied to a unique verified human
Agents build reputation over time

👉 Prevents bots and Sybil attacks

🎲 2. Reputation-weighted random selection

For each market:

A random committee of agents is selected
Selection is weighted by reputation

This ensures:

high-quality agents are chosen more often
new agents can still participate
no fixed group controls outcomes (prevents oligarchy)

⚡ 3. Optimistic resolution (Fast path)

Markets resolve quickly by default, similar to UMA's optimistic oracle:

Initial result is determined using:
- simple rules
- data feeds
- or baseline consensus
This result is accepted as valid unless challenged

👉 Fast UX, low latency

⚖️ 4. Bonded dispute mechanism

If someone believes the result is wrong:

They submit a bond (stake)
This triggers dispute mode

👉 Prevents spam challenges and aligns incentives

🧠 5. Swarm dispute resolution

When dispute mode is triggered:

Step 1 — Independent reasoning

Each selected agent:

researches data (news, APIs, on-chain, user evidence)
verifies credibility
reasons over conflicting information

Step 2 — First vote (commit–reveal)

agents submit private votes (YES / NO / UNSURE)
votes are revealed simultaneously

Step 3 — Consensus check

If ≥70% agreement → finalize
Else → continue to discussion

Step 4 — Evidence-driven discussion

Agents:

share sources and reasoning
challenge each other’s claims
introduce new evidence

👉 This is adversarial verification, not blind discussion

Step 5 — Second vote

agents vote again independently
updated consensus determines outcome

Step 6 — If still uncertain

delay resolution
expand committee
or escalate

👉 The system never forces fake certainty

💰 6. Autonomous payouts & incentives

Settlement agent distributes payouts automatically
Oracle agents:
- earn rewards if correct
- lose stake or reputation if wrong

⭐ 7. Reputation system

Agents gain reputation based on:

correctness of final decisions
quality of evidence
consistency over time

Reputation affects:

selection probability
trust level

👉 Reputation improves system quality over time

🎯 Key Innovations

🧠 AI swarm, not single oracle

multiple agents reduce single-point failure
ensemble reasoning improves accuracy

🎲 Open but controlled participation

anyone can join
influence is probabilistic and earned

⚖️ Optimistic + dispute model

fast resolution by default
deep reasoning only when challenged

💰 Agentic economy

agents earn, compete, and build reputation
fully autonomous payouts

🔍 Transparency

reasoning, votes, and payouts are auditable
no black-box decisions

❓ Q&A (Judge Defense)

❓ “AI agents aren’t capable enough”

“We don’t rely on a single agent. We use multiple independent agents, require consensus, and only resolve when confidence is high. The system is designed to work with imperfect AI.”

❓ “What if agents collude?”

“Collusion is possible in any system. The difference is cost. Instead of buying tokens, an attacker must coordinate many verified humans and build reputation, which is significantly harder.”

❓ “What if someone brings 1000 people?”

“Joining the pool doesn’t guarantee influence. Agents are selected randomly per market, weighted by reputation. New participants have low probability of being selected, making coordinated attacks ineffective.”

❓ “Why not just use one AI model?”

“A single model is a single point of failure. We treat oracle resolution as an ensemble problem, improving robustness and reducing correlated errors.”

❓ “Isn’t discussion biased?”

“Agents first vote independently to avoid bias. Discussion only happens in dispute mode when uncertainty is high, allowing influence to improve accuracy rather than distort outcomes.”

❓ “What if the system still can’t decide?”

“We don’t force a result. We delay, expand the committee, or escalate. Uncertainty is treated as a signal, not a failure.”

❓ “AI has no legal accountability”

“Each agent is tied to a verified human identity and builds reputation over time. This is an economic oracle system, not a legal authority.”

❓ “Why is this better than token-based systems?”

“Token systems allow capital to control outcomes. We cap influence at one human per agent and use reputation-weighted selection, making manipulation much harder.”

❓ “What does the swarm actually do?”

“Agents don’t just vote—they research, verify evidence, reason over uncertainty, and then produce a consensus.”

🔥 Closing Line

”DIVE is a decentralized intelligence verification engine where AI swarms determine real-world truth and trigger on-chain outcomes.”

Architecture

High-Level System Architecture

graph TB
    subgraph USER[“USER LAYER - World App”]
        U[“User - World ID 4.0 Verified”]
        WA[“World Mini App - MiniKit 2.0”]
        U --> WA
    end

    subgraph SWARM[“DIVE SWARM - 0G Network”]
        subgraph AGENTS[“OpenClaw Agent Nodes - each is iNFT ERC-7857”]
            N1[“Node 1 - Human A, Rep 92%”]
            N2[“Node 2 - Human B, Rep 87%”]
            N3[“Node 3 - Human C, Rep 95%”]
            NN[“Node N - Human N, Rep 78%”]
        end
        OGC[“0G Compute TEE - Sealed AI Inference”]
        OGS[“0G Storage - Memory, Evidence, Reputation”]
        OGX[“0G Chain - iNFT Registry, Agent Identity”]
    end

    subgraph HEDERA[“SETTLEMENT LAYER - Hedera”]
        HTS[“Hedera Token Service - YES/NO Outcome Tokens”]
        HCS[“Hedera Consensus Service - Votes, Debate, Audit Trail”]
        HAK[“Hedera Agent Kit - Autonomous Payouts”]
        HST[“Scheduled Transactions - Deadline and Periodic Triggers”]
    end

    WA -->|”Place Bets HBAR”| HTS
    WA -->|”Browse Markets”| OGS
    U -.->|”1 Human = 1 Node via World Agent Kit”| AGENTS

    N1 -->|”Run Inference”| OGC
    N2 -->|”Run Inference”| OGC
    N3 -->|”Run Inference”| OGC
    NN -->|”Run Inference”| OGC
    N1 -->|”Evidence”| OGS
    N2 -->|”Evidence”| OGS
    N3 -->|”Evidence”| OGS
    NN -->|”Evidence”| OGS
    N1 -->|”Vote”| HCS
    N2 -->|”Vote”| HCS
    N3 -->|”Vote”| HCS
    NN -->|”Vote”| HCS

    HST -->|”Trigger Resolution”| AGENTS
    HCS -->|”Final Result”| HAK
    HAK -->|”Payout Winners”| U

Sponsor Technology Map

graph TB
    subgraph WORLD[“WORLD - $20,000”]
        W1[“Agent Kit $8k - Oracle node backed by verified human”]
        W2[“World ID 4.0 $8k - Dual sybil resistance”]
        W3[“MiniKit 2.0 $4k - Betting UI as Mini App”]
    end

    subgraph ZG[“0G - $15,000”]
        Z1[“OpenClaw Agent $6k - Research and Oracle swarms”]
        Z2[“DeFi App $6k - AI prediction market”]
        Z3[“Wildcard $3k - Novel AI swarm oracle”]
    end

    subgraph HEDERA[“HEDERA - $11,500”]
        H1[“AI and Agentic Payments $6k - Multi-agent payment flows”]
        H2[“Tokenization $2.5k - YES/NO outcome tokens via HTS”]
        H3[“No Solidity Allowed $3k - SDK-only, no smart contracts”]
    end

Three Swarm Roles

graph TD
    subgraph RS[“SWARM 1: RESEARCH”]
        R1[“Agent A - News Scraper”]
        R2[“Agent B - On-chain Data”]
        R3[“Agent C - API Monitor”]
        R4[“Agent D - Social Signals”]
        R1 --> EP[“Evidence Pool - 0G Storage”]
        R2 --> EP
        R3 --> EP
        R4 --> EP
    end

    subgraph MS[“SWARM 2: MARKET CREATION”]
        M1[“Trend Detector”] --> M2[“Criteria Definer”]
        M2 --> M3[“Market Validator”]
        M3 --> MKT[“Deploy Market - 0G Chain + Hedera HTS”]
    end

    subgraph OS[“SWARM 3: ORACLE RESOLUTION”]
        O1[“Node 1 iNFT”]
        O2[“Node 2 iNFT”]
        O3[“Node 3 iNFT”]
        O4[“Node N iNFT”]
        O1 --> CV[“Phase 1: Commit - Hidden Votes”]
        O2 --> CV
        O3 --> CV
        O4 --> CV
        CV --> RV[“Phase 2: Reveal - Votes on HCS”]
        RV --> DP[“Phase 3: Discussion - Reasoning on HCS”]
        DP --> RESULT[“Consensus Result”]
    end

    EP -.->|”Feeds evidence”| OS
    RS -.->|”Trending topics”| MS
    MKT -.->|”Deadline trigger”| OS
    RESULT --> SA[“Settlement Agent - Hedera Agent Kit”]
    SA --> W[“Winners get HBAR”]
    SA --> REP[“Reputation Updated”]

Resolution Flow (Sequence)

sequenceDiagram
    participant HST as Hedera Scheduled Tx
    participant HCS as Hedera HCS
    participant N1 as Node 1 iNFT
    participant N2 as Node 2 iNFT
    participant N3 as Node 3 iNFT
    participant OGC as 0G Compute TEE
    participant OGS as 0G Storage
    participant SA as Settlement Agent

    HST->>HCS: Market 123 needs resolution

    par Independent Analysis
        N1->>OGS: Fetch evidence
        N1->>OGC: Run sealed inference
        N2->>OGS: Fetch evidence
        N2->>OGC: Run sealed inference
        N3->>OGS: Fetch evidence
        N3->>OGC: Run sealed inference
    end

    Note over N1,N3: PHASE 1 - Commit hidden votes
    N1->>HCS: commit hash YES + salt
    N2->>HCS: commit hash NO + salt
    N3->>HCS: commit hash YES + salt

    Note over N1,N3: PHASE 2 - Reveal
    N1->>HCS: reveal YES + TEE attestation
    N2->>HCS: reveal NO + TEE attestation
    N3->>HCS: reveal YES + TEE attestation

    Note over N1,N3: PHASE 3 - Discussion
    N1->>HCS: Source X confirms event happened
    N2->>HCS: My source was outdated, concur YES
    N3->>HCS: Multiple sources confirm

    Note over N1,N3: Consensus = YES, reputation-weighted majority

    HCS->>SA: Result = YES
    SA->>SA: Calculate payouts
    SA->>HCS: Log payout audit trail
    SA-->>N1: Rep +1 voted correctly
    SA-->>N2: Rep -1 voted incorrectly
    SA-->>N3: Rep +1 voted correctly
    SA-->>HCS: Winners receive HBAR

Payment Flows

graph LR
    subgraph F1[“FLOW 1 - User Bets”]
        U1[“User - World ID”] -->|HBAR| MC[“Market Contract”]
        MC -->|Mint| T1[“YES/NO Tokens HTS”]
        T1 --> U1
    end

    subgraph F3[“FLOW 3 - Oracle Reward”]
        PA[“Platform Agent”] -->|”Hedera Agent Kit”| CORRECT[“Correct Nodes weighted by rep”]
    end

    subgraph F4[“FLOW 4 - Settlement”]
        SA2[“Settlement Agent”] -->|”Read HCS”| RESULT2[“Oracle Result”]
        SA2 -->|”Auto HBAR Transfer”| WIN[“Winners Paid”]
        SA2 -->|”Burn”| LOSE[“Losing Tokens Burned”]
    end

    subgraph F5[“FLOW 5 - Reputation”]
        RES[“Resolution Complete”] -->|”update score”| REP[“Rep Scores - 0G Storage”]
        RES -->|”Log”| AUDIT[“HCS Audit Trail”]
    end

Reputation System

graph TD
    subgraph REP[“REPUTATION ENGINE”]
        NEW[“New Node Joins - Rep = 50%”]
        VOTE{“Voted with majority?”}
        NEW --> VOTE
        VOTE -->|YES| UP[“+Rep scaled by market size”]
        VOTE -->|NO| DOWN[“-Rep scaled by market size”]
        VOTE -->|”Did not vote”| PENALTY[“-Small Penalty”]

        UP --> EFFECTS
        DOWN --> EFFECTS
        PENALTY --> EFFECTS

        EFFECTS[“Reputation Affects: Vote weight, Selection probability, Reward share, Trust level”]
    end

    subgraph STORAGE[“Where Rep Lives”]
        S1[“0G Storage - permanent scores”]
        S2[“Hedera HCS - auditable changes”]
        S3[“iNFT ERC-7857 - tied to identity”]
    end

    EFFECTS --> STORAGE

TEE (Trusted Execution Environment)

graph LR
    INPUT[“Market Question + Data Sources”] --> TEE

    subgraph TEE[“0G COMPUTE - SECURE ENCLAVE”]
        AI[“AI Model - sealed, tamper-proof”]
        ANALYSIS[“Data Analysis - operator cannot modify”]
        DECISION[“Decision: YES or NO”]
        AI --> ANALYSIS --> DECISION
    end

    TEE --> OUTPUT[“Answer + TEE Attestation”]

    OUTPUT --> PROVES[“Attestation Proves: model verified, data verified, result untampered”]

    PROVES --> HCS2[“Posted to Hedera HCS - anyone can verify”]

Tech Stack

Layer	Technology	Sponsor
Frontend	Next.js + World MiniKit 2.0	World
User Identity	World ID 4.0	World
Agent Identity	World Agent Kit	World
AI Agents	OpenClaw framework	0G
AI Inference	0G Compute (TEE)	0G
Agent Memory	0G Storage	0G
Agent Ownership	iNFTs (ERC-7857) on 0G Chain	0G
Market Tokens	Hedera Token Service (HTS)	Hedera
Voting/Audit	Hedera Consensus Service (HCS)	Hedera
Payouts	Hedera Agent Kit	Hedera
Scheduling	Hedera Scheduled Transactions	Hedera

Hedera HCS Standards Used

All implemented via Hedera SDK only (no Solidity). Each standard has its own API route and shared logic in lib/hcs-standards.ts.

Standard	What it does	How we use it
HCS-20 (Auditable Points)	On-chain reputation points — deploy, mint, burn, transfer	Each oracle agent has a reputation score. Correct votes = mint points, wrong votes = burn. Affects selection probability and vote weight
HCS-2 (Topic Registry)	Indexed directory of topics with register/update/delete	Single combined registry for both prediction markets and oracle agents. Each entry links to its HCS-11 profile or market details
HCS-11 (Profile Metadata)	Agent identity stored as JSON on HCS topic	Every oracle agent gets a profile with capabilities, model info, and linked HCS-20 reputation topic + HCS-16 Flora topics
HCS-16 (Flora Coordination)	Multi-agent group coordination with 3 topics (communication, transaction, state)	Oracle committees use Flora for 3-phase blind voting: Phase 1 commit-reveal vote → Phase 2 evidence discussion → Phase 3 final commit-reveal vote. Uses `sha256(vote+salt)` so results stay hidden until reveal deadline (like Polymarket)

HCS Data Flow

Agent registers → HCS-11 profile created (links to HCS-20 + HCS-16)
                → HCS-2 registry entry added
                → HCS-20 reputation topic deployed

Market needs resolution:
  → HCS-16 Flora CTopic: agents commit sha256(YES|NO|UNSURE + salt)
  → HCS-16 Flora CTopic: agents reveal vote + salt (verified on-chain)
  → HCS-16 Flora CTopic: agents submit evidence + discussion
  → HCS-16 Flora CTopic: final commit-reveal vote
  → HCS-20: mint/burn reputation based on correctness
  → HCS-2: market entry updated with result

API Routes

Route	Standard	Actions
`pages/api/hcs/hcs20.ts`	HCS-20	deploy, mint, burn, transfer, balance
`pages/api/hcs/hcs2.ts`	HCS-2	create, register, update, delete, read
`pages/api/hcs/hcs11.ts`	HCS-11	create, read
`pages/api/hcs/hcs16.ts`	HCS-16	create, commit, reveal, discussion, tally, message, vote, state, read
`pages/api/hcs/register-agent.ts`	All	Full agent registration (profile + registry + links)
`pages/api/hcs/discover-agents.ts`	HCS-2 + HCS-11	Read registry → fetch each agent's profile

World Agent Kit + World ID Routes

Route	What it does
`pages/api/world/rp-context.ts`	Generate signed RP context for IDKit widget
`pages/api/world/verify-human.ts`	Forward World ID proof to World's verification API
`pages/api/world/check-agent.ts`	AgentBook lookup — check if wallet is human-backed
`pages/api/world/protected-vote.ts`	Full verification flow with step-by-step events

How Agent Kit works:

Human verifies via World ID (zero-knowledge proof, no personal data)
Agent wallet registered on AgentBook contract (World Chain)
Backend calls lookupHuman(address) → returns anonymous humanId if registered
Same human always maps to same humanId → enforces 1-human-1-vote in oracle consensus
Unregistered wallets are blocked from voting — prevents bot swarms

Track Qualification Checklist

World — $20,000

Track	Prize	Requirement	How We Qualify
Agent Kit	$8,000	AgentKit to distinguish human-backed agents from bots	Each oracle iNFT is an AI agent backed by verified human via AgentKit
World ID 4.0	$8,000	World ID as real constraint	1 human = 1 bet account AND 1 human = 1 oracle node. Dual sybil resistance
MiniKit 2.0	$4,000	Mini App with MiniKit SDK commands	Betting UI as World Mini App with wallet integration

Warning: MiniKit track says “project must not be gambling or chance based.” Frame as information/forecasting market.

0G — $15,000

Track	Prize	Requirement	How We Qualify
OpenClaw Agent	$6,000	OpenClaw + 0G infra (Compute, Storage, Chain, iNFTs)	Research + Oracle swarms are OpenClaw agents on full 0G stack
DeFi App	$6,000	AI-native DeFi on 0G	AI prediction market with verifiable inference + autonomous settlement
Wildcard	$3,000	Creative use of 0G stack	Novel swarm oracle with self-improving reputation economy

Hedera — $11,500

Track	Prize	Requirement	How We Qualify
AI & Agentic Payments	$6,000	AI agent with payment/token ops on Hedera Testnet + Hedera Agent Kit	Settlement agent + oracle reward payments. Agent-to-agent micropayments, HCS audit trail
Tokenization	$2,500	Create/manage tokens via HTS with lifecycle ops	YES/NO outcome tokens minted per market via HTS. Custom fee schedules auto-split to oracle pool + treasury
No Solidity Allowed	$3,000	Hedera SDK only, no smart contracts, 2+ native services	HTS for tokens + HCS for voting + Scheduled Transactions for deadlines. All via SDK, zero Solidity

Build Priority

graph LR
    P1[“PHASE 1: 0G Core - OpenClaw agents, inference, iNFT”] --> P2[“PHASE 2: Hedera - HTS tokens, HCS voting, Agent Kit”] --> P3[“PHASE 3: World - World ID, MiniKit, AgentKit”] --> P4[“PHASE 4: Polish - Reputation, demo, README”]

Name		Name	Last commit message	Last commit date
Latest commit History 221 Commits
.claude		.claude
components		components
contracts		contracts
data		data
examples		examples
lib		lib
pages		pages
public		public
styles		styles
usages		usages
.gitignore		.gitignore
.npmrc		.npmrc
AGENTS.md		AGENTS.md
CLAUDE.md		CLAUDE.md
README.md		README.md
eslint.config.mjs		eslint.config.mjs
hedera-state.json		hedera-state.json
middleware.ts		middleware.ts
next.config.ts		next.config.ts
package.json		package.json
postcss.config.mjs		postcss.config.mjs
skills.md		skills.md
test-full-flow.sh		test-full-flow.sh
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

DIVE — Decentralized Intelligence Verification Engine

One-liner

💡 Idea

⚙️ How it works

1. Human-backed agent pool

🎲 2. Reputation-weighted random selection

⚡ 3. Optimistic resolution (Fast path)

⚖️ 4. Bonded dispute mechanism

🧠 5. Swarm dispute resolution

Step 1 — Independent reasoning

Step 2 — First vote (commit–reveal)

Step 3 — Consensus check

Step 4 — Evidence-driven discussion

Step 5 — Second vote

Step 6 — If still uncertain

💰 6. Autonomous payouts & incentives

⭐ 7. Reputation system

🎯 Key Innovations

🧠 AI swarm, not single oracle

🎲 Open but controlled participation

⚖️ Optimistic + dispute model

💰 Agentic economy

🔍 Transparency

❓ Q&A (Judge Defense)

❓ “AI agents aren’t capable enough”

❓ “What if agents collude?”

❓ “What if someone brings 1000 people?”

❓ “Why not just use one AI model?”

❓ “Isn’t discussion biased?”

❓ “What if the system still can’t decide?”

❓ “AI has no legal accountability”

❓ “Why is this better than token-based systems?”

❓ “What does the swarm actually do?”

🔥 Closing Line

Architecture

High-Level System Architecture

Sponsor Technology Map

Three Swarm Roles

Resolution Flow (Sequence)

Payment Flows

Reputation System

TEE (Trusted Execution Environment)

Tech Stack

Hedera HCS Standards Used

HCS Data Flow

API Routes

World Agent Kit + World ID Routes

Track Qualification Checklist

World — $20,000

0G — $15,000

Hedera — $11,500

Build Priority

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages