Each one runs in the browser, takes a few minutes, and answers a different question about the same problem. None of them collects an email address, and none of them sends anything anywhere.
Five questions on one decision family. Returns the earliest unmet stage in the decision path — visibility, permission, decision geometry, capacity, or the learning loop — and the mechanism behind it.
Six questions plus revenue. Returns an annual leakage band across three streams and nine value pools, with the assumptions visible. An order of magnitude, not a number to take to a CFO.
Five questions plus your typical signal-to-response time. Returns the closest of six operating archetypes, and whether the architecture carries the work or people do, benchmarked against your industry median.
All three run deterministic rule sets over what you report. There is no model call, so none of them can invent a diagnosis — but self-reported inputs are generous, and organizations reliably under-report their own latency. Treat a result as a place to start looking, not a finding.
The same reasoning is published as a machine-readable endpoint at /api/assess for agents, and as tools on the MCP server. Those return JSON, not a page — they are for software, not for reading.
A constraint you can name is a training problem: the masterclass rebuilds one decision family end to end. A decision already moving and already costing you is an advisory problem.