Method
What Gavora measures, and how we know.
Most workplace assessments are one method: ask people about themselves, then describe them back. Gavora is built differently, and this page says exactly how. If you work in psychometrics, this is for you.
Three layers of measurement, not one.
Performance
Timed puzzles, perception tasks, memory span. Behavior speaks for itself.
Situational judgment
Real scenarios with pre-coded response options. Choices under context.
Self-report
Balanced-keyed items, reserved for the disposition behavior cannot reach.
Layer weights renormalize when a layer is missing, and the layers cross-check each other: a person who claims calm under load and then drops four of five message streams has told us something one questionnaire never could.
Five constructs, with their lineage.
Cognitive FlexibilityGrounded / Adaptive
Grounded in the set-shifting and cognitive flexibility literature (the Miyake and Friedman executive function tradition). Items deliberately separate capacity to shift from willingness to shift, the classic conflation in self-report flexibility scales.
How it is measured: Rule-change scenarios with pre-coded response options, plus self-report items balanced across poles for acquiescence control.
Human Signal ProcessingStructural / Relational
Descends from the social-perception performance tradition (RMET-class stimulus reading, MSCEIT-style scored perception) rather than self-rated empathy, which correlates poorly with measured accuracy.
How it is measured: The Read: original photographic eye-region stimuli scored against target mental states, plus filmed scenario reads. Performance carries weight a self-description cannot.
Pattern Recognition SpeedMethodical / Instinctive
Built on remote-associates (Mednick) and visual-matrix completion paradigms. The style-not-ability trap is handled directly: latency only counts as signal on correct responses, so speed is read as cognitive style, never as intelligence.
How it is measured: Timed word-association and original matrix items capturing both accuracy and response latency, gated so only correct-response speed contributes.
Cognitive Load ToleranceFocused / Expansive
From the working-memory-capacity literature: complex span tasks (operation span) and dual-task paradigms, with the engagement checks the originals require.
How it is measured: A shortened operation-span block with math-accuracy quality gates, and a five-stream message triage simulating a loaded sales floor.
Ambiguity MonetizationSteady / Pioneering
Gavora's proprietary construct. It draws on tolerance-of-ambiguity research but measures something that literature does not: what a person does inside ambiguity, not how it makes them feel. By design it validates through criterion evidence over time, not convergent correlation with an existing scale, because no existing scale targets it.
How it is measured: A generative task rubric-scored blind, behavioral forced-choices, and one hard rule: no item may mention money or business outcomes, because contaminating a trait measure with the outcome it should predict fails discriminant validity by definition.
Design commitments.
Bipolar spectra, not deficits. Every dimension has two poles and both are strengths. There is no low score anywhere in the instrument, which is not a kindness; it is the construct model. A Methodical pattern-reader is not a slow Instinctive one.
Style is never ability. Wherever speed or accuracy is captured, it is gated so the instrument cannot quietly become an intelligence test. Latency counts only on correct answers; span tasks carry engagement checks; the generative task is rubric-scored blind.
Fairness is instrumented, not asserted. Demographic questions are optional, collected in distributed micro-moments, and used for one purpose: differential item functioning analysis across groups. Items flagged for bias, demographic assumption, or social desirability are rewritten or cut. An AI validation pass triages every item against the writing rubric before human review; humans make every final call.
Consent is the architecture. No manager sees a profile that was not deliberately shared, sharing is revocable, and the conversational layer refuses employment-decision questions in code. These are not policies; they are structural properties of the system, red-teamed and logged.
Where the evidence stands.
Honestly: Gavora is in its validation phase. Dimensions one through four inherit convergent validity targets from established benchmark instruments. Dimension five, the proprietary one, validates the hard way: against recorded outcome events over time in a live cohort of four automotive retail groups, several hundred profiles. Reliability analysis will trim the item bank; augmentation projections are labeled provisional in the product itself until the data earns otherwise. An assessment that never says "provisional" is hiding something.
And what we will not claim: Gavora does not measure intelligence, predict individual performance, or rank human worth, and it refuses use in hiring, firing, promotion, and pay. Those limits are part of the method too.