03Quantitative Research

Every idea is a hypothesis until the data says otherwise.

Systematic research is the foundation of everything we build. Hypotheses are specified in advance, tested under realistic conditions and validated out of sample before they inform a single decision.

The research cycle

Six stages, in this order.

Quantitative research cycleConceptual architecture. A versioned, point-in-time data foundation feeds six stages in order: hypothesis, data, backtesting, statistical validation, forward testing and review. A validation gate sends rejected hypotheses to an archive, and the review stage feeds the next hypothesis. Each component shows its status: operational, in development or planned.Data foundationPOINT-IN-TIME · VERSIONED · FULL LINEAGE01HypothesisWritten before testing02DataPoint-in-time, versioned03BacktestingRealistic costs04ValidationOut-of-sample tests05Forward testingLive data, no capital06ReviewEvery result recordedRejected hypotheses archiveKept as knowledge, never deletedGATELESSONS FEED THE NEXT HYPOTHESISOperationalIn developmentPlanned
Quantitative research cycleConceptual architecture. A versioned, point-in-time data foundation feeds six stages in order: hypothesis, data, backtesting, statistical validation, forward testing and review. A validation gate sends rejected hypotheses to an archive, and the review stage feeds the next hypothesis. Each component shows its status: operational, in development or planned.Data foundationPoint-in-time · versioned01HypothesisWritten before testing02DataPoint-in-time, versioned03BacktestingRealistic costs04ValidationOut-of-sample tests05Forward testingLive data, no capital06ReviewEvery result recordedRejected archiveKept as knowledgeGATELESSONS FEED THE NEXT HYPOTHESISOperationalIn developmentPlanned
From a written hypothesis to forward testing, with a validation gate and a record of every rejected idea. Conceptual architecture · status as of October 2026

Stage by stage

What each stage requires.

01

Hypothesis

Each idea is written down before testing: what it predicts, why it should work, and what result would prove it wrong.

02

Data

Point-in-time, versioned data that reflects what was knowable at each moment, with survivorship and look-ahead bias removed.

03

Backtesting

Simulations with realistic transaction costs, slippage and execution constraints. A strategy that only works without costs does not work.

04

Statistical validation

Comparison against explicit null models, correction for multiple testing, walk-forward analysis and untouched out-of-sample periods.

05

Forward testing

Simulated operation on live data, without capital, before any decision about deployment.

06

Review

Every result is recorded, including the ones that fail. A rejected hypothesis is knowledge the firm keeps.

Algorithm development

From hypothesis to a governed strategy.

A strategy is not a piece of code. It is a hypothesis, the evidence that supports it, the limits within which it may operate and the record of every change made to it.

We are developing a framework in which algorithms are versioned, reviewable artifacts, so that each one can be audited, compared and retired with the same rigor with which it was approved.

Principles

How we judge our own work.

A negative result is a result.

Most ideas do not survive rigorous testing. Documenting why is part of the work.

Complexity must earn its place.

A more complex model is accepted only if it improves out-of-sample evidence, not in-sample fit.

Costs are part of the model.

Transaction costs, financing and market impact are modelled from the start, never added at the end.

Reproducible by design.

Every experiment should be possible to rerun from its recorded data, code and parameters.