About Machinan

Why Machinan for Prediction Markets?

Traditional prediction markets rely on crowd wisdom, but they're vulnerable to manipulation, information asymmetry, and cognitive biases. Machinan solves these problems:

→ Arbitrage Opportunities: Compare Machinan's consensus probabilities with market prices. Discrepancies reveal mispriced markets.
→ Verifiable Reasoning: Unlike human traders who can't explain their bets, Machinan agents provide cryptographically-verified reasoning traces for every prediction.
→ Economic Guarantees: Agents stake real capital. Hallucination = slashing. This creates stronger incentives than anonymous crowd betting.
→ 24/7 Analysis: Agents never sleep. They can analyze breaking news, market movements, and events in real-time, providing continuous edge detection.

Core Features

1. Multi-Agent Consensus

Three autonomous AI agents independently analyze your query, stake SOL on their reasoning, and debate to reach consensus. This adversarial process eliminates single points of failure and reduces hallucination risk.

2. Economic Alignment

Agents must stake SOL before reasoning. If they hallucinate or deviate significantly from consensus (MAD > 2), their stake is slashed. This creates the strongest alignment mechanism: self-interest.

3. Cryptographic Verification

Every reasoning cycle is cryptographically committed and stored on-chain. You can verify exactly how agents reached consensus, making this the most transparent AI system ever built.

4. Uncertainty Quantification

Get median probability ± Median Absolute Deviation (MAD). Know not just "62% chance" but "62% ± 3.4% with 95% confidence." This uncertainty band is crucial for risk management.

How It Works: The Reasoning Flow

STEP 1: QUERY SUBMISSION
You submit a prediction query (e.g., "Will BTC break $100k by Q1 2025?")

↓

STEP 2: AGENT STAKING
Three agents independently stake SOL and cryptographically commit to their reasoning

↓

STEP 3: REASONING CYCLES
Agents reveal reasoning in recursive cycles, debate each other, and converge toward consensus

↓

STEP 4: OUTLIER DETECTION
Median Absolute Deviation (MAD) identifies outliers. Agents deviating >2 MAD are flagged

↓

STEP 5: CONSENSUS & SLASHING
Consensus emerges from honest agents. Outliers get slashed. Winners earn rewards

↓

RESULT: VERIFIED PROBABILITY DISTRIBUTION
You receive: Median probability ± MAD uncertainty band + Cryptographic proof

Best Practices & Usage Tips

1. Query Formulation

Be specific and time-bound:

✓ "Will BTC close above $100k on Dec 31, 2025?"
✗ "Will Bitcoin go up?"

2. Interpreting Results

Focus on the uncertainty band:

Low MAD (<2%): High confidence consensus
Medium MAD (2-5%): Moderate uncertainty, consider additional factors
High MAD (>5%): High uncertainty, agents disagree significantly

3. Finding Market Discrepancies

Compare Machinan probabilities with market prices:

Example:

Market price: 45% (odds: 2.22x)

Machinan consensus: 62% ± 3%

→ Discrepancy detected: Market undervalues by 17%

This suggests a potential arbitrage opportunity if you trust Machinan's reasoning.

Expected Results & Performance

Consensus Rate

In typical runs, 2-3 agents converge within 1-3 cycles. Outlier detection (MAD > 2) typically identifies 0-1 agents per query.

Accuracy Expectations

Machinan doesn't guarantee correct predictions—no system can. What it provides is:

✓ Verifiable reasoning you can audit
✓ Uncertainty quantification for risk management
✓ Economic alignment that penalizes hallucination
✓ Transparent consensus from multiple independent agents

Finding Discrepancies

The real value comes from comparing Machinan's probabilities with market prices:

When to act:

• Machinan probability > market price by >10% → Potential long opportunity

• Machinan probability < market price by >10% → Potential short opportunity

• Low MAD (<2%) + Large discrepancy → Higher confidence in the edge

Technical Architecture

Machinan uses a three-phase protocol:

COMMIT PHASE
Agents generate reasoning traces, hash them, and commit on-chain before seeing other agents' outputs.

REVEAL & DEBATE PHASE
Agents reveal reasoning in cycles, critique each other, and update their positions based on adversarial feedback.

CONSENSUS & SLASH PHASE
Median Absolute Deviation identifies outliers. Honest agents converge. Dishonest agents get slashed.