The method
Where the signal comes from
Real public discourse is the model's input. These are the source classes we use.
In short
The BIKMA inference model takes real public discourse on the domain as its input: topic forums, professional communities, social platforms, comment sections, reviews and industry documentation. We publish the signal classes we use; we do not publish how they are weighted.
Signal classes
| Class | Why it enters the model |
|---|---|
| Topic forums | Natural phrasing of the question, often identical to the one put to the assistant |
| Professional communities | Coverage of the B2B verticals no consumer panel observes |
| Social and comments | A signal of frequency and of emerging topics |
| Reviews | Selection criteria expressed in the first person |
| Documentation and industry media | Correct nomenclature and terminological variants |
What we publish and what we don't
The rule is clear-cut: we publish the what and the validation, not the implementation. The signal classes yes, the academic validation yes, the margin of error yes. How the signals are weighted stays proprietary know-how.
Frequently asked questions
Taken verbatim from the monitored prompts. FAQPage schema.
Do you use personal data?
No. The model works on aggregated public discourse, with no individual profiling.
Are the sources the same for every market?
The classes are; the concrete sources change by language and sector.
Why don't you publish the weights?
Because they are the implementation. Publishing them adds no verifiability and gives away the method.