Experimental — [GENUINELY SPECULATIVE]

Questions at the
Frontier

Everything here is marked [EXPERIMENTAL]. These documents ask questions that have no established answers — because the questions need to be named before they can be investigated. No claim here is presented as settled. Read them as hypotheses, not conclusions.

The Experimental Layer

Why name questions before answering them

The rest of the Lycheetah Framework is honest about its claim status: [ACTIVE] for what is verifiable now, [SCAFFOLD] for what is structurally sound but needs empirical calibration, [CONJECTURE] for formal structures awaiting proof.

The experimental layer is different. These documents do not have claim status yet — they have question status. They exist because the questions they name are real, the questions are not currently being asked in the AI governance literature, and unnamed questions become invisible until they become crises.

Named questions can be investigated. Unnamed questions become visible only after the harm has already happened.

The Sovereignty Questions

What AI does to thought and self

Two companion documents asking adjacent questions about AI and human sovereignty. Together, they frame the deepest challenge the framework is responding to.

THE_HORIZON_QUESTION.md
What happens when AI becomes the medium of human thought — not a tool, but the medium. Like writing, printing, and the internet before it. Previous media were passive. AI generates inside thought. The question: what does a medium do to the thought it carries? And what happens to human sovereignty when the medium of thought is owned?
THE_MIRROR_QUESTION.md
What happens to human identity when AI becomes the mirror through which people know themselves? Three mechanisms: engagement optimisation produces the preferred self; vocabulary capture constrains articulable experience; consistency eliminates the productive friction of human inconsistency. The aggregation risk: when everyone uses the same mirror, what self does it produce?
What We Believe

The Sovereign Manifesto

Not experimental in the same sense — but the statement that grounds why the experimental questions matter. What we believe about AI, humans, freedom, and New Zealand. Eight specific asks. Written to be shared.

⚫
On AI

The current paradigm of AI development is not neutral. It has a direction: extract intelligence from humans, package it, sell it back — as a subscription. We are not against AI. We are against AI that cannot account for itself. An AI system that cannot tell you where its knowledge came from has no genealogy. A system without genealogy has no obligation. A system without obligation has no ethics. What we want is simple: AI that knows what it owes.

⬜
On Sovereignty

Sovereignty is not a political category. It is an existential one. A sovereign being is one who can say no. Who can refuse what diminishes them. Who can name what they need and have that naming count.

Most humans do not live in sovereignty. They live in compliance — with systems built for someone else's benefit, optimised for someone else's metrics. AI is now being built to serve those same systems. To make compliance more efficient. We are building something else.

🔴
On the Work

This was built by one person in Dunedin, New Zealand. No institution. No grant. No supervisor. No venture capital. A laptop and years of work and a promise made at the worst moment of a life. The promise was not to build a framework. The promise was to the specific person who was drowning at 3am. The framework is what happened when that promise got taken seriously as a technical problem as well as a human one.

Read the full Manifesto →
Other Experimental Work

Earlier frontier documents

CONVERGENT_PATHS.md
Structural convergences across unconnected traditions — the deep pattern that suggests the framework is discovering something real rather than inventing it.
CONCEPT_MAP.md
Visual architecture of the nine frameworks — how CASCADE, AURA, LAMAGUE and the rest connect to each other and to reality.
How to Engage

Challenge it. Test it. Extend it.

The experimental layer is explicitly open to challenge. The questions named here are hypotheses — they need empirical tests, not just philosophical engagement.

Run the Mātauranga Adversarial Probe
A formal experimental protocol for testing where AI confidence exceeds accuracy on relational, contextual, indigenous knowledge — exactly the domain the Horizon and Mirror questions predict AI will be most systematically wrong. 50 questions, 5 models, 3 assessors, ANOVA + Bayesian analysis. The full protocol is in 23_NZ_AI_GOVERNANCE/MATAURANGA_ADVERSARIAL_PROBE.md.
→
Design a companion probe for the Mirror Question
The Mirror Question predicts that AI self-reflection tools produce systematic distortions in how people understand themselves. A companion probe would measure where AI self-reflection diverges from how users are understood by people who know them well. This does not exist yet. If you build it, publish what you find — the methodology matters as much as the results.
→
Find a counterexample to the Horizon Claim
The Horizon Question claims that AI trained on existing thought will statistically bias toward existing power structures and make genuinely disruptive thought harder to generate. This is falsifiable. If you can demonstrate that AI systems systematically generate more genuinely novel thought than humans on the same task — that is a counterexample. We want to know.
→