Not a compliance document. Not a list of prohibitions. A positional statement: what we hold non-negotiable, and why the project's purpose shapes every decision we make.
Every project needs a point it navigates toward — not a quarterly goal, but an IFR: the ideal final result that makes bad decisions legible as bad.
This is not a metaphor. It is the operational IFR of the project. Every instrument, every case, every research line is evaluated against this horizon. When a decision is difficult, we ask: does this move toward coordinated movement, or away from it? Does this amplify, or does this destroy?
Five principles that structure how StratoAtlas is built and how it is used.
StratoAtlas produces diagnostic material — structural maps, classified contradictions, research lines. It does not produce decisions. The decision belongs to the human actor. Instruments that remove human judgment from the loop are structurally incompatible with what we are building. We do not automate the positional layer.
The core claim of CDSA is that some contradictions should be held, not eliminated. This applies to the project itself. We do not simplify what is genuinely complex in order to make it more convenient, more marketable, or more legible to a wider audience. Premature resolution is a failure mode, not a feature.
Every piece of knowledge in the system carries an epistemic status: seed, hypothesis, working model, or canonical. We do not present a hypothesis as a fact, a working model as a law, or a seed as a published finding. The system is designed to make uncertainty visible — not to conceal it.
StratoAtlas works with real organisations, real cases, real people. Diagnostic work at this level carries responsibility. We do not publish material that could harm the individuals or organisations involved in cases, even when that material would be analytically useful. The case corpus is curated. Sensitivity is a structural criterion, not an afterthought.
The multilogue configuration places AI in structurally differentiated roles — some generative, some critical, some observational. In all roles, the AI participant is a contributor to the thinking, not the source of canonical knowledge. Promotion from hypothesis to canon requires human verification. The epistemological stack is designed to make this non-negotiable.
Concrete commitments — what the project will do, and what it will not.
StratoAtlas is built on top of AI systems — and uses them in a specific way: as structurally differentiated participants in a research configuration, not as oracles or automation engines.
The core structural commitment of the project aligns with the better direction of the field: human oversight is not an obstacle to good work — it is a structural requirement of it. Every instrument we build, every case we publish, every research line we hold open reflects this commitment.
StratoAtlas research produces empirical evidence about how multi-agent AI configurations behave when structurally differentiated — what divergence between models actually means, how epistemic status degrades when positions lose their distinctness, and what architectural conditions make human-AI collaboration genuinely productive rather than superficially so.
This is not advocacy. It is documented field observation — the kind that can feed back into how AI systems are built to support rather than replace human thinking. We believe this work is useful. We make it available under CC BY 4.0.
This page exists as a decision-making instrument, not as a declaration. When a decision is unclear — about what to publish, how to use a case, whether to automate something, how to position the project — this page is where to start.
The question is simple: does the proposed decision move toward the manifesto's IFR, or away from it? Does it increase clarity, or does it obscure? Does it hold the structural tension that drives the work, or does it collapse it for convenience?
If it moves away — that is the signal to stop. Not to debate, not to rationalise. To stop, return to this frame, and find a different path.