What H.A.R.I. is not.
A platform in regulated, consequential domains earns credibility by being precise about what it is — and equally precise about what it is not. Every entry below is a non-claim, and the list is load-bearing.
H.A.R.I. is not a chatbot.
It does not maintain a conversation, generate text in response to prompts, or have a persona. The kernel produces no free-form output — only three canonical states with structured metadata. Conversational AI solves a different problem; H.A.R.I. governs the problems that follow.
H.A.R.I. is not a single AI model.
It is an architecture. It sits alongside any model from any provider and governs the actions the model’s outputs would otherwise trigger. Replacing the model changes the input, not the platform. The model proposes; the deterministic layer disposes.
H.A.R.I. is not an autonomous decision-maker.
It does not act on its own initiative. Every irreversible action requires a human authorization step, mediated by the Time Sovereignty Layer. Reversible actions proceed under ALLOW only when preconditions are satisfied. Human authority is preserved by design, not by convention.
H.A.R.I. is not a surveillance system.
No facial recognition. No identity accumulated over time. No profiles. No matching a person across contexts. Where it reads context — stadium safety, for example — it reads situational signals (density, flow, structural risk), never persons.
H.A.R.I. is not a scoring or social-credit engine.
No probabilities, rankings or scores attached to individuals. Probabilistic evidence — model outputs, sensor confidences — is input to be evaluated by deterministic policy, never the system’s own assessment of a person.
H.A.R.I. is not an agent factory.
It does not generate agents, deploy models or write code. It governs what already-deployed AI systems and authorized data sources participate in. Agent factories optimise for generation; governance layers optimise for refusing to act without a deterministic basis.
H.A.R.I. is not a consumer product.
No app, no dashboard, no consumer-facing surface. Institutional infrastructure, engaged by federations, leagues, clubs and regulated operators under controlled disclosure, NDA and scoped pilots.
H.A.R.I. is not a black-box automation tool.
Every decision is reproducible: inputs, policy version, output and provenance are written to a hash-chained log signed with ECDSA P-256. Any past decision can be replayed. There is no vendor magic to trust — only a chain to verify.
Now the affirmative side.
The kernel, the five layers, the three canonical outputs and the Time Sovereignty Layer.