How VGO Rank is measured
A formal result requires an eligible entity and scope, valid observations under the published protocol, accuracy gates and an auditable issuance decision. Confidential question banks are not published.
Four opportunity judgments
- Appearance (P)
- Was the right entity present in the machine output?
- Correct understanding (U)
- Were the material facts attributed accurately?
- Credible evidence (C)
- Was the supporting source relevant and verifiable?
- Appropriate recommendation (R)
- Did the output fit the question and user intent?
- Resulting visit (V)
- Authorized visit observation is an outcome check, not a Rank input.
Observation surfaces at launch
- Search (required)
- AI and answers (required)
- Social discovery (may be removed by market protocol)
- VHI remains diagnostic only
- Visits and leads validate outcomes only
Sampling and comparison
The unit is a question × surface × sampling time × repeat batch in a stated market and intent mix. Formal windows last 30 days. Customer diagnostic questions, confidential formal questions and held-out validation questions stay separate. Failed collection is a quality issue, not entity absence.
Score and gates (public)
Per opportunity q = 0.20P + 0.30(P∧U) + 0.25(P∧U∧C) + 0.25(P∧U∧R), with published renormalization when R is not applicable. Launch surface weights: search 35%, AI/answers 45%, social 20%. Formal grading uses conservative S* (bootstrap lower bound). Accuracy and non-brand gates cap the level before S* thresholds apply. First release issues at most VR8.
Issuance and calibration
Production issuance may open when SoftHSM (or equivalent) and dual-human approval are in place. Entity queues, pilot windows and agreement metrics are post-open quality work—not hard gates for opening the signing switch. Incomplete calibration must be disclosed; do not claim first-issuance calibration acceptance if it was not completed.