Skip to content
Alpha Bench

Strategy families

Twelve strategy families

Professional investing is not one skill. Alpha Bench divides it into twelve families, from valuing a business to pricing the tails of a distribution, each with its own tasks, metrics and failure modes. Together they cover the main schools of professional investing. Every system is evaluated on each family separately, so strength in one cannot hide weakness in another.

The twelve families

How each family is evaluated

Evaluation tracks

How each family is evaluated

Every family is tested on the same four tracks, whether the system under test is a model, a harness or a vertical agent.

  • 01

    Analysis

    Point-in-time research tasks. The system works only from documents and data available on a given date and produces an answer that can be checked.

  • 02

    Simulation

    Historical market replay. The system decides one date at a time, sees only what was known then, and pays realistic costs and frictions.

  • 03

    Live

    Forward evaluation on live data. Decisions are recorded before outcomes exist, so the answer cannot be recalled from training data.

  • 04

    Client

    Serving a human user. The system has to understand their situation, explain plainly, disclose risk honestly and say no when it should.

Help shape the tasks in your field.

If you work with one of these strategies as a practitioner or researcher, we would welcome your help with tasks, scoring rubrics and failure-mode checks.