ShoggothMind/mind-judgements

The Mind: Verified Consensus & Reasoning Judgements

Deterministic, code-verified mathematical and logical evaluations from The Mind autonomous protocol on Solana.

Dataset Partitions & Exact Row Counts

This dataset is cleanly partitioned into four configurations:

1. demo (180 rows): Simulated historical test rounds generated during protocol development. Every question has code-generated ground truth and simulated tentacle votes.
2. synthetic_train (3,000 rows): The exact dataset used to train Mind-v1 (35.8M parameters), comprising 300 questions for each difficulty level 1 through 10 across 8 deterministic domains (arithmetic, modpow, letter counting, word sorting, calendar logic, numerical sequences, base conversions, comparative logic).
3. test (500 rows): The held-out benchmark set (50 questions per difficulty level 1..10) used for perplexity and exact-match evaluation against Mind-v1 and frontier models.
4. live (0 rows): Production rounds anchored to the Solana mainnet ledger via SPL Memo Merkle roots. Populated continuously as rounds seal on mainnet.

Total items in dataset: 3,680 verified rows.

---

On-Chain Solana Cryptographic Verification

Every hourly consensus round commits a deterministic SHA-256 Merkle root to the Solana ledger via SPL Memo. You can verify the Merkle root of any round locally using this Python snippet:

import hashlib
import json

def compute_round_merkle_root(questions_list):
    """
    Computes deterministic SHA-256 Merkle root of a consensus round.
    Matches the on-chain SPL Memo transaction on Solana.
    """
    # 1. Leaf hashes
    leaf_hashes = []
    for q in questions_list:
        payload = f"{q['id']}:{q['question']}:{q.get('consensus_answer', q.get('ground_truth'))}".encode("utf-8")
        leaf_hashes.append(hashlib.sha256(payload).hexdigest())

    # 2. Pairwise Merkle reduction
    nodes = leaf_hashes
    while len(nodes) > 1:
        next_level = []
        for i in range(0, len(nodes), 2):
            left = nodes[i]
            right = nodes[i + 1] if i + 1 < len(nodes) else left
            combined = hashlib.sha256(f"{left}:{right}".encode("utf-8")).hexdigest()
            next_level.append(combined)
        nodes = next_level

    return nodes[0] if nodes else ""

# Example usage:
with open("demo.jsonl", "r", encoding="utf-8") as f:
    sample_round = json.loads(f.readline())
    calc_root = compute_round_merkle_root(sample_round["questions"])
    print(f"Calculated Root: {calc_root[:16]}... | Record Root: {sample_round.get('root', '')[:16]}...")

---

License

Creative Commons Attribution 4.0 International (CC-BY-4.0).