HBIF is Polynovea's broader behavioural-intelligence framework: domain-specific reality and intelligence at Layer 1, candidate shared behavioural state and dynamics at Layer 2, and a long-horizon domain-independent human behavioural model at Layer 3.
HBIF is not one model, one API, one scoring formula, one database or one LLM. It is the framework through which Polynovea separates domain-specific behavioural intelligence from shared research that may, or may not, transfer across domains.
The architecture exists to avoid two opposite errors: treating every domain as unrelated and throwing away useful learning, or pretending a universal behavioural system already exists before the evidence supports it.
Each layer answers a different research question. Layer 3 is not a product tier or a synonym for macro behaviour; it is the domain-independent convergence target.
Layer 1 is where Polynovea builds real products for real operating environments. Every domain keeps its own ontology, state, workflows, evidence boundary, interfaces and validation programme.
Layer 2 studies whether deeper behavioural state and dynamics can be represented without carrying a domain's surface categories directly into the shared layer. The current technical reference is the Behavioral Phase Model (BPM).
Layer 3 is the long-horizon attempt to identify mathematical and behavioural structures that survive domain removal and no longer depend on labels such as employee, guest, customer, booking, sale or venue.
Products create value first. Evidence becomes research input only where it is lawful, permitted and scientifically suitable.
A compelling pattern is a research candidate, not a canonical mechanism.
A candidate shared mechanism should survive evidence from meaningfully different domains rather than only the domain where it was discovered.
The mapping from domain-specific observations into a shared construct must be stated instead of being hidden inside a convenient label.
Candidates must be tested outside the discovery setting, including failure modes, boundary conditions and null or baseline comparisons.
Probabilistic state, competing explanations and evidence maturity must not be flattened into factual certainty by a model or LLM interface.
A shared construct has to improve explanation, state estimation, prediction, intervention reasoning or decision quality - not merely look mathematically interesting.
Research moves into canonical HBIF or product use only through explicit evidence and governance gates.
Standard customers receive the contracted product without automatically becoming research partners. Polynovea does not treat raw customer databases as unrestricted HBIF property.
Where research participation exists, permissions should define the exact event or data classes, purpose, retention, geography, access, revocation, publication and whether raw, derived, aggregated or anonymised forms may be used.
Polynovea has executable experimental assets, preserved outputs, a large historical Hospitality corpus, calibrated behavioural-state tooling and explicit research-to-product mappings. What it does not yet have is proven cross-domain universality.
Mathematical, statistical and dynamical-systems experiments have been implemented against real datasets rather than existing only as conceptual notes.
BPM operationalises a subset of the research into a working state-and-dynamics pipeline and currently anchors the Layer 2 technical reference.
Candidate structures still need independent domains, replication, falsification and boundary testing before they can be treated as domain-independent mechanisms.
No. HBIF is an evolving behavioural-intelligence framework and research architecture. Polynovea has implemented experimental infrastructure and domain evidence, but cross-domain validity remains unproven and the company does not claim a solved universal model of human behaviour.
No. Layer 1 is deliberately domain-specific. A Workplace product does not prove a Hospitality or Education product can be copied from it. Each domain must earn its own ontology, workflows, interfaces, evidence and product validity.
The Behavioral Phase Model (BPM) is the current Layer 2 technical reference. It is research engineering, not a claim that one shared behavioural state model has already been universally validated across domains.
No. Domain learning can become a candidate for shared research, but cross-domain transfer is earned through evidence. Some mechanisms may transfer, some may remain domain-specific, and some may fail completely.
MellowBase is the planned managed cloud service built on Polynovea's Content Operations OSS foundation. Its product role is managed Content Operations. Where a customer separately opts into governed research participation, MellowBase can provide broad cross-context observations for Layer 3 experiments; ordinary MellowBase usage is not automatically HBIF research data.
No. Standard product use does not grant unrestricted research rights. Research participation and the exact permitted data or derived evidence classes must be separately governed, contractually defined and privacy-aware.
No. An LLM may consume bounded HBIF outputs, but it should not erase provenance, uncertainty or competing hypotheses. Interpretation, recommendation, policy decisions and actual interventions remain distinct layers of authority.