Why Vektor exists

We went looking for an edge and didn’t find one.

Last updated: July 2026

Most prop tools start from the assumption that they beat the market and work backwards to the evidence. This one started from a model and a question, and the answer came back no. That answer is the reason the product looks the way it does, so it belongs on a page of its own rather than in a footnote.

What happened

  1. 012025 — 2026

    Built the engine

    A Monte Carlo simulator for NFL player props: 25,000 simulations of each week's game, sampling team pace, play mix and per-player usage to produce a full probability distribution for every player and market. The goal was the ordinary one — find prop lines the market had priced wrong.

  2. 02May 2026

    Pre-registered the test, then ran it

    Decision rules were written down before the results came in: what counted as an edge, over how many seasons, and what would kill the product if it failed. Four NFL seasons — 2022, 2023, 2024, 2025 — were run as walk-forward holdouts on the same commit with the same parameters.

  3. 03May 2026

    The edge didn't survive contact with the market

    Against a naive baseline the engine looked good. Against real sportsbook closing lines it did not replicate: the strongest grade returned −1.75%, +5.78% and +1.43% across three seasons. A follow-up audit tested 30 computable subsets. Zero passed the pre-registered bar.

  4. 04May 2026

    Published the verdict instead of burying it

    The honest reading of that data is that this engine has no repeatable edge against closing lines. That finding was written up, dated, and published rather than quietly reframed — including the seasons and probability buckets where the model is measurably miscalibrated.

  5. 05July 2026

    Built the product on the honest result

    What survived the testing was real: the engine has genuine predictive resolution, it just doesn't beat the closing line. So the product became the research itself — the distributions, the historical hit rates, the market's own no-vig prices — instead of picks it can't justify.

Who builds this

One person, working alone, without a background in tech. That is worth stating plainly for two reasons. It sets expectations: there is no support desk behind this, and features arrive one at a time. And it explains the obsession with method — coming from outside the industry, the only way to know whether any of this actually worked was to test it the way it would be tested if someone were trying to prove it wrong.

That is also why the negative result got published. A solo project with no institutional reputation to protect has nothing to trade on except whether its numbers can be checked.

Check the work

Nothing above needs to be taken on trust. The method and the commitments are both published.

  • The phase-gated methodology — how the engine was built and tested, phase by phase, with the decision rules written before the results.
  • The charter — the four commitments we hold ourselves to, including the ones that cost us money.
  • The calibration record — where the model is well calibrated and where it is measurably off.

We show you what the model and the market say. We don’t claim to beat the market.