An estimator you can audit.
AVYR is explicit about the difference between what WFM normally means and what it actually computes. Every branch carries a plain-language claim and the formula AVYR states for it — and where the running engine is a proxy rather than that formula's direct computation, the classification badge says so.
Analytical architecture · three layers
The engine does not tag branches with a layer; the three layers are a reconstruction of each branch's dependencies.
Fourteen inputs · the constellation
Spec ⇄ engine
The spec is what a WFM practitioner expects the term to mean. The engine is what AVYR computes from the signals it has. Where they differ, the site says so rather than hiding it. A metric that needs a survey (CSAT, NPS, customer effort) is labeled a proxy; a composite with no industry equivalent is labeled AVYR-derived; a simplified formula is disclosed as an estimator, not the full Erlang method.
Limitations, stated
Deterministic replay on a synthetic corpus. No real customer data, no personally identifiable information. The estimator is a workload model, not a queueing simulation: it omits the traffic-variability term that Erlang C adds. Predictions are projections with assumptions, not facts.