Methodology DGN-OBS-1.0
This page fixes the vocabulary the Observatory uses. It is published so that any observation Dgenius records can be checked, disputed and reproduced against a stated standard.
Answer states
An ambiguous output is recorded as UNCERTAIN. It is never promoted to a stronger state to improve a number.
Metric definitions
The share of measured observations in which the company appeared in any state.
Presence is not preference. An answer may include a company while steering the buyer elsewhere.
The share of relevant observations in which the company entered a recommended or shortlist state.
Recommendation share is not market share and does not measure revenue. It describes measured answers only.
Broader presence, including appearances that are not recommendations.
A mention can be neutral, comparative or unfavourable. State is not sentiment.
The share of applicable observations in which the company did not appear at all.
Absence in a measured sample is not proof of absence everywhere. It is evidence within the stated sample.
How differently the company is represented across the measured systems, expressed as the spread between the system with the highest and lowest presence rate.
Systems differ in retrieval, recency and configuration. Variance describes representation, not model quality.
How frequently named competitors occupy the relevant answer universe rather than the company.
Bounded by the defined competitor set. Undefined competitors are not counted.
Which sources repeatedly appear across the measured answers, and how often.
Recurrence describes what was cited alongside answers. It is not proof that a source caused the answer.
Standing caveats
Visibility scopes
Collection
A category is expressed as the buying questions a real purchaser asks — shortlist, comparison, requirement, risk and switching intents — not as keywords.
Each intent is expressed as a family of phrasings so a single lucky or unlucky wording cannot be mistaken for a finding.
Answers vary across sessions. Observations are repeated across systems and sessions so variance is measured rather than ignored.
Every observation is classified into one documented state by an analyst. Ambiguous outputs are recorded as UNCERTAIN and never promoted.
Company, category, intent, prompt family, system, timestamp, state, competitors, citations, sources, methodology version and analyst are stored with each observation.
Remeasurement adds rows. Original observations are preserved so change over time remains measurable.
What an observation is
One classified answer from one system, to one prompt, at one moment, about one company. Nothing more is claimed for it.
What an observation is not
It is not a ranking, a score, a market position, a traffic estimate or a prediction. Aggregates are only as strong as the sample and the protocol behind them.
Where the data comes from
Public research observations are collected by Dgenius analysts under a defined public protocol, or supplied by a client who has explicitly approved publication. Confidential engagement data is never published, in any form, including aggregate, unless approval is explicit.
Automation status
Where an authorised model or API integration exists, it is named in the series methodology. Where it does not, observations are recorded manually by an analyst. Dgenius never simulates a model answer.