Market Forecasting Engine
that Keeps Score.

Run any ticker through multiple frontier AI models. Each reasons independently. An arbiter measures consensus. Every signal is tracked to outcome — publicly and timestamped.

The Horde Protocol

How TradeHorde turns raw market data into scored conviction.

Data Ingest

Live prices, technicals, regime context, and fundamentals assembled per ticker.

Multi-Model Debate

Frontier models analyze independently. Each builds bull and bear cases.

Arbiter Consensus

Calibration-weighted scoring. Agreement + accuracy = conviction.

Signal & Outcome

Tracked against live prices to public resolution. Wins and losses.

Higher Conviction, Better Outcomes

When models align with high conviction, the signal is stronger. Here's how conviction tiers perform on real market outcomes.

Low Conviction32.4% win rate(N=108)
Conditional54.0% win rate(N=50)
High Conviction66.7% win rate(N=21)

179 resolved signals · Data updates hourly

View full track record →

What's Wrong with Asking ChatGPT?

Three problems every LLM has when you ask it for market analysis.

No Market Data

They hallucinate prices, invent support levels, and guess at volume. TradeHorde injects live quotes, technicals, volume profiles, earnings dates, and market regime before any model sees the ticker.

They Hedge Everything

"On one hand... on the other hand." Every answer is a non-answer. TradeHorde forces each model to make a directional call with exact entry, target, and stop levels — plus a full bull and bear thesis.

No Accountability

Conversations disappear. Nobody tracks whether that "bullish setup" actually played out. TradeHorde monitors every signal against real prices and publishes outcomes — wins and losses — publicly.

Why Multi-Model Consensus?

The science of forecasting, applied to markets.

The problem with single-source predictions

One analyst, one model, one opinion — no matter how confident — is just noise dressed up as signal. Markets are full of smart people who are confidently wrong.

What actually works

Research on forecasting shows that weighted aggregation of independent forecasters consistently beats individual experts. Not because any single forecaster is brilliant, but because their errors cancel out when they're truly independent.

How we apply this

  • 1.
    Multiple AI models analyze independently

    Each model sees the same data but reasons differently. No model sees what others said before committing.

  • 2.
    We measure calibration, not just confidence

    A model that says "70% bullish" should win ~70% of the time. We track this. Models that are overconfident get down-weighted.

  • 3.
    Consensus requires agreement AND participation

    High conviction means multiple models, analyzing independently, reached the same conclusion with high confidence. Disagreement or abstention lowers conviction.

  • 4.
    Outcomes feed back into the system

    Every signal is tracked to resolution. Win rates, R-multiples, hold times — all measured by model, horizon, and conviction bucket. This isn't a black box; it's a track record.

What this means for you

  • Signals aren't opinions — they're measured consensus
  • Conviction scores are earned, not asserted
  • The track record page shows you exactly how well this works (or doesn't)

We don't claim to predict the future. We claim to aggregate independent analysis and measure the results honestly.

Start Analyzing

Run any ticker through multiple frontier models. See where they agree and get calibrated conviction.

Sign Up Free
About TradeHorde | Market Forecasting Engine that Keeps Score