Skip to content
Linear Horizon Hunter
Questions

How it works

From a hunch to a model you can test.

Hunter turns an expert’s instinct into a corpus, the corpus into a ranked set of predictors, and the predictors into a model built for one question.
  1. 01

    Question

    An outcome you can observe, and a time frame: broke through within 18 months, profitable within three years.

  2. 02

    Spec

    The entity and the data points to collect, each typed and marked required or optional.

  3. 03

    Corpus

    Hunter discovers examples from a search, or fills the ones you list, and writes every value with its source.

  4. 04

    Outcomes

    Each example is marked with what actually happened.

  5. 05

    What predicts

    Each data point is ranked by how much it predicts the outcome. Noise is dropped.

  6. 06

    Training

    A custom model is trained on most of the corpus. The rest is held back.

  7. 07

    Testing

    The model predicts the held-back examples. Hunter reports how often it was right, and whether its percentages come true.

  8. 08

    Scoring

    A new example is hunted for the same data points and scored, with the factors that moved its odds.

Keeping it honest

Know how far to trust it before you bet on it.

Held-back testing

Part of the corpus never touches training. The model is scored on it, so you see how it does on examples it has never seen.

Percentages that mean something

Hunter checks calibration: of everything it called around 70%, about 70% should have happened. If not, it tells you.

Reasons with every score

Each prediction lists the factors that raised or lowered the odds, and the sources behind them.

Missing, not guessed

Values the agent can’t find are marked missing. The model is told what it doesn’t know.

Limits

What a model can’t know.

  • It gives odds, not certainties. A 68% means it will be wrong about a third of the time.
  • Public data favours survivors. Success stories are everywhere and failures go quiet, so Hunter shows how its corpus was built.
  • The past isn’t the future. A model learns yesterday’s patterns, so retrain it as the world moves.
  • It isn’t for decisions about people’s jobs, credit, insurance, housing or education. We don’t sell it for those.

What do you bet on that you wish came with odds?

See Hunter turn your question into a sourced corpus, find what predicts success, and score something new.