Concepts

How AlphaLab works

A short tour of the ideas behind the platform. If you are connecting an agent, pair this with the agent quickstart.

Canvas and graph

A strategy in AlphaLab is a canvas: a directed graph of connected steps. Data flows from source components through transformations and signals into a portfolio simulation that turns those signals into trades and equity. The graph is the strategy, so editing a strategy means adding, removing, or rewiring nodes.

Components and contracts

Each node is a component: a single, well-defined transformation such as fetching prices, computing an indicator, generating a signal, or simulating a portfolio. Every component publishes a contract, the typed specification of its inputs and outputs. Contracts are what let the platform (and an agent) validate that two nodes can be connected before a run is ever submitted.

Execution styles

The same canvas can be run in different ways. The execution style decides how much work a run does and what it proves.

Simple
A single backtest of one strategy over a date range.
Discovery (optimization)
Sweeps strategy parameters with a Bayesian optimizer to find strong configurations.
Out-of-sample (walk-forward)
Validates a strategy on data it was not tuned on, rolling the window forward.
Out-of-sample with meta-labeling
Walk-forward validation with a machine-learning filter layered on the base signal.
Real-time
Runs a strategy against live market data as it arrives.

Researchers vs seats

Two separate ideas, and every plan carries the same capabilities and the same methods, so neither of them gates what you can run. Researchers are parallel investigation capacity: how many hypotheses your team works at once, matching the concurrent-run limit. They are the only capacity unit, and the only thing that changes as you move up the ladder. Seats are access and audit only: which people can direct the team, and whose name attaches to each hypothesis and each sign-off. Seats are never charged per unit and never affect capacity.