Aletheia: The Evidence-Grounded Debate Arena

A Decentralized Fact-Adjudicated Coliseum on GenLayer

Aletheia is a decentralized dispute and debate terminal where users challenge claims (Theses) not through mere rhetoric, but through verifiable evidence. Unlike typical subjective AI forums, every argument proposed or challenged in Aletheia must be anchored to a public Evidence URL. GenLayer's validators fetch the live webpage content on-chain, and an LLM arbiter judges the claims strictly against the retrieved web facts.

---

The Dragon Coliseum flow

Below is the architectural flow of Aletheia, showing how claims and evidence are processed and verified:

flowchart TD
    subgraph Arena ["The Dragon Coliseum"]
        Topic["Topic: Claim and Evidence Link"]
        Contender["Challenger: Counter Claim and Evidence Link"]
    end

    subgraph GenLayer ["GenLayer VM Execution"]
        Crawler["Web Crawler: Fetching Evidence Text"]
        Arbiter["Dragon Arbiter LLM Adjudicator"]
    end

    subgraph Consensus ["Validator Consensus Jury"]
        Leader["Leader Node: Coordinates Evaluation"]
        Validators["Validators: Verify and Agree on Block"]
    end

    subgraph Storage ["On-Chain Registry"]
        TreeMap["Storage: Rotate Proponent or Defend Thesis"]
        Queue["Circular Buffer Log Queue"]
    end

    Topic -->|Submit Clash| Leader
    Contender -->|Submit Clash| Leader
    Leader -->|Fetch URLs| Crawler
    Crawler -->|Adjudicate facts| Arbiter
    Arbiter -->|Categorical Verdict| Leader
    Leader -->|Propose block| Validators
    Validators -->|Verify and reach consensus| Storage
    Storage -->|If Overthrown| TreeMap
    Storage -->|Prune Event Logs| Queue

Loading

---

Key Innovations & Resolutions

Aletheia is designed to address key production constraints of LLM-native smart contracts:

1. Factual Grounding (Web Crawling): We enforce factual grounding by requiring an evidence_url on proposal and clash. The contract fetches the live page content using gl.nondet.web.get(url) and truncates it to fit LLM window sizes, ensuring the judge bases its verdict strictly on verifiable public text.
2. Consensus Stability (Tight Margin Tolerance): By grounding evaluations on specific crawled text rather than general training data, we naturally reduce LLM variance. This allows us to tighten the validator consensus tolerance window to 12 points (down from 30 in previous builds), improving audit accuracy.
3. Aligned Exception Routing: Exception types are unified under ERR_USER and ERR_LLM. The leader node and validators raise the same tagged exceptions during parsing errors, preventing consensus failure when handling malformed LLM outputs.
4. O(1) Circular Event Buffer: Instead of clearing and rebuilding the event ledger on every write, we overwrite elements at a modulo index (total_debates % MAX_LOG_SIZE). This keeps log prunings at an O(1) gas fee on-chain, while the view function sorts it chronologically off-chain.
5. TreeMap Storage Partitioning: We store state fields in independent maps (arena_topics, arena_proponents, arena_claims, etc.) instead of serializing the entire arena object to a JSON string. This eliminates the gas cost of serialization/deserialization during writes.

---

Contract Public API

Mutating Transactions (Writes)

• propose_thesis(topic: str, opening_claim: str, evidence_url: str) -> str
  • Establishes a new debate arena, setting the creator as the proponent.
• clash_thesis(arena_id: str, contender_claim: str, contender_evidence_url: str)
  • Launches an AI consensus duel. The incumbent stands unless the challenger is factually superior.

Query View Methods (Views)

• get_stats() -> dict
  • Returns global counters: total arenas, total debates, total overthrows.
• get_arena(arena_id: str) -> dict
  • Returns detailed state and progression history for a specific topic.
• get_arenas(start: int) -> list
  • Paged list of debate arenas, newest first.
• get_ledger(start: int) -> list
  • Paged list of historical event logs sorted chronologically (newest first).

---

Setup & Testing

1. Compile & Lint Contract

pip install genvm-linter
genvm-lint contracts/aletheia.py

2. Run Integration Tests

Make sure the GenLayer test runner is installed, then execute:

gltest tests/integration/ -v -s --network studionet

3. Start Frontend Local Dev

cd frontend
npm install
npm run dev

---

License

MIT. Owned and maintained by 9ja_maxx.