Research

Research sits behind every part of a plan: how portfolios are built, how they are checked, how the future is modelled, and how ideas are tested. We publish as we go.

What we work on

  1. 01Portfolio construction

    How to weight a portfolio around one person's goals, horizon and comfort with risk. Optimization, not a model portfolio.

  2. 02Diagnostics

    What a portfolio really holds: overlap between funds, concentration, and drift away from the plan.

  3. 03Scenario analysis

    Where a plan could end up, shown as calibrated ranges and likelihoods instead of a single number.

  4. 04Validation

    How to tell a real investment idea from noise before it reaches anyone's money.First paper published

Latest paper · Validation

Working paper · October 2026 · SSRN Deflate by Bits, Not Trials Why Query-Counted Deflated Sharpe Ratios Fail Under Adaptive Search, and a Sealed-Holdout Remedy Shlok Sobti · Invsify Technologies t > Ḡ⁻¹(α / K) Read the paper on SSRN →

The finding, in plain words

A backtest can make pure noise look like skill. The standard safeguard counts how many strategies were tried. We show that count is the wrong measure once a researcher keeps building on what worked, and we propose a test that holds anyway.

423
real industry strategies tested
1970 to 2026
of market history
6
formal results, with proofs
  • Finding 1Counting trials is not enough.

    The standard safeguard raises the bar with the number of strategies tried. When each new idea is chosen after seeing earlier results, that count understates how hard the data has been searched.

  • Finding 2Measure what the test reveals.

    What matters is how much a researcher can learn from the held-back data, not how many questions were asked. A bar set on that basis is valid for any researcher, however adaptive.

  • Finding 3A sealed test.

    Held-back data that answers only pass or fail, a limited number of times. It cannot be tuned against, because it reveals almost nothing.

  • Finding 4It holds for automated research too.

    We tested code-writing research agents and learning systems against it. The standard test certified their searches on pure noise. The sealed test did not.

This is methodology research. It describes how investment ideas are tested and is not investment advice or a statement about returns.

Publications

Deflate by Bits, Not TrialsWhy Query-Counted Deflated Sharpe Ratios Fail Under Adaptive Search, and a Sealed-Holdout Remedy. Shlok Sobti.SSRN →

We will keep publishing. New work is listed here first.