What this is

Performix's outputs drive people decisions, so trust rests on three things: the diagnostic is measured, not generated; predictions are confidence-gated; and outputs are interpretable.

Measured, not generated

The diagnostic engine is real psychometrics (IRT, MaxDiff, Monte Carlo / EVPI–EVSI, Wilson intervals), precomputed from the measurement substrate — not produced on demand by a language model. AI handles plumbing around the engine (schema mapping, evidence extraction, ETL), never the scoring. A CAMS score traces to items and a model, not to a prompt.

Confidence gating (decide vs. abstain)

Conformal-Interval Confidence Gating wraps a base prediction and decides whether to act or abstain based on prediction uncertainty — so the system declines to push an action it isn't confident enough to support, rather than guessing.

Interpretability

Interpretability Rules answer "why this nudge / why this flag" over a precomputed prediction, so an executive sees the basis of a recommendation, not a black box.

Protected feedback

CAMS items surface inline at strategic moments as protected feedback — not an exposed survey — keeping the signal honest and the respondent protected.

Honest scope

Performix supports better people decisions; it does not make them, and its accuracy depends on the upstream measurement services it consumes (toolbox reincarnation, Principia methods) and on coverage for the team in question. A binding constraint is a diagnosis to act on, not a verdict.

See also

Concepts · Architecture · APIs