Skip to main content
RelatrixPsychometric computation with explicit uncertainty

Psychometric computation service

Measure, compare and retrieve with the uncertainty kept in view

Relatrix turns Item Task responses into versioned psychometric profiles, evaluates how two profiles fit in a given context and produces vectors your own system can use to shortlist candidates. Every result names its model version, its evidence and its scientific status.

See how it worksAPI and contact

Current status: every Relatrix model is experimental and its validation has not been evaluated yet. Each result says so in its warnings and is produced only after an explicit opt-in.

What Relatrix computes

  • Psychometric profiles

    Dimension estimates with measurement uncertainty, evidence coverage and evidence quality. Missing evidence stays visible instead of being filled with a neutral value.

  • Exact Fit

    The fit of two profiles in a named Fit Context such as marriage, friendship or colleagueship, with results in both directions and the contribution of every relation.

  • Retrieval vectors

    Context-specific vectors your system can index to shortlist profiles by distance. They narrow a candidate pool; the fit of a pair is evaluated with Exact Fit.

  • Derived views

    Relatrix-native interpretations, such as an affection preference profile, computed from the canonical profile without replacing it.

What Relatrix does not do

  • It keeps no subjects, responses, profiles or vectors. Each request is calculated in memory and nothing about the person is stored.
  • It searches no candidate pool and ranks or selects no one. Search, ranking and decisions stay with your system and the people using it.
  • It is not a dating, HR or recruitment application; it is the measurement and calculation core such applications can call.
  • It does not turn a fit index into a probability or chance of relationship or job success.

One mechanism for every language

Assessment content is available in English and Turkish. Every language version of an Item Task leads to the same evidence, and every comparison is calculated the same way whatever the language, country or population of the people involved. Where equivalence across groups has not been shown yet, results carry a warning instead of a different calculation.