Proof
Measurement methodology
Public claims are tied to product capabilities. Forecasts are labeled. Absolute wording that is not yet statistically evidenced stays a target, not a measured result.
PhD-Level Intelligence
Role-specific competence, structured reasoning (facts / assumptions / estimates / unknowns), cross-executive deliberation, governed tool execution, and human override. Dataset version phd/v1. Rubric and fixtures are in the product repository and run in local CI. Details: PhD-Level standard.
Voice
- TTFA start: end of user final speech. Stop: first audible TTS audio packet.
- Target: p95 TTFA at or below 200ms on the warm path.
- MOS target: 4.85 with a published sample protocol. Production scores are not fabricated here.
- Identity disclosure when asked is required and is not a MOS failure.
- Public demo audio and text: 24-hour retention.
Coverage and availability
Inbound voice is staffed continuously. Availability is published at /status. Outages, degraded mode, and recovery are reported there. “Never miss another lead” is the operating target measured as answered-call coverage, not an unmeasured absolute.
ROI
The homepage calculator uses contribution (gross profit) per close plus separately entered time value. 211× is a scenario under user-editable inputs, not a measured customer average.
Workforce comparison
Human-equivalent cost uses the published salary table in product code (measurement date 2026-08-30). It compares included executive roles to those salaries, not a claim that SOVREN is a company of 50 humans.
Audit integrity
Consequential actions append to a hash-chained ledger (prev_hash + SHA-256). Tenants can export and recompute. Deletion produces a sealed receipt for covered stores. See Security.