AYVID

Research

Notes from the desk.

We publish selected research externally — methodology, philosophy, and counterintuitive findings — and keep proprietary signal work inside the firm. Below is what we’re comfortable saying out loud.

  • 2026 · Q1
    Method

    Why naive walk-forward backtests are an indictment of the trader, not the market.

    Trading every ticker every day on technicals lost 99.98% over an 8-year window in our shop. The same engine, restricted to event-conditioned windows surfaced by the graph, returned +43.65% compounded. The full report is in the operations console.

    Note 01 · Internal release
  • 2026 · Q2
    Architecture

    A four-agent swarm produces calibrated conviction. A single model does not.

    Strategist, Mapper, Auditor, Supervisor. We measure their disagreement as a first-class signal — high-disagreement trades are systematically smaller, regardless of average conviction.

    Note 02 · Internal release
  • 2026 · Q2
    Data

    What an N-th order supply-chain graph actually buys you.

    On the Red Sea closure of late 2024, the graph proposed bunker-fuel beneficiaries (BP) two hops from the shock — names a keyword search would miss. The relevant traversal is exhibit 1 in the AYVID 10-K corpus.

    Note 03 · Internal release
  • 2026 · Q2
    Risk

    Independent kill switches are not paranoia. They are uptime.

    An out-of-process C++ watchdog flattened the book three times in shadow-mode last quarter. Twice it was wrong. Once it wasn't. We will keep it.

    Note 04 · Internal release

Verifying any of the above against the live system requires console access. LP and YC partner enquiries get a passphrase the same day.

Contact the desk →