Principles / How we work
Ambitious research requires restraint, clarity, and the willingness to revise a position when evidence changes. These principles guide both technical work and the way Erebus describes that work publicly.
We take difficult questions seriously, including questions that sit outside familiar product categories or established assumptions. Curiosity creates permission to investigate; it does not determine the answer. Ideas remain subject to implementation, observation, and critical evaluation.
The strength of a public claim should reflect the strength of the available evidence. A working prototype, a controlled test, and a public release are different things. Erebus aims to identify those differences rather than compressing every stage of development into the language of achievement.
Prototypes are instruments for learning. A useful experiment makes a question more precise, reveals constraints, or produces information that changes the next decision. Unexpected results and failures are not discarded simply because they complicate the original idea.
Systems should collect, move, and retain only the information appropriate to their purpose. Privacy is considered in architecture, operation, and human interaction—not treated solely as a statement of intent after a system is built.
Security requires continuous attention to boundaries, dependencies, access, failure, and change. It is a technical discipline and an operational responsibility. Public transparency does not require disclosing information that would weaken a system or expose internal infrastructure.
Autonomous operation should be bounded, observable, interruptible where appropriate, and proportionate to the task. Increased capability should be accompanied by clearer evaluation, stronger operational understanding, and appropriate forms of oversight.
We value technical reasoning over consensus for its own sake. Independence includes the ability to question prevailing assumptions, but it also requires the discipline to recognize when existing methods are supported by stronger evidence.
Complex language cannot substitute for clear requirements, careful implementation, testing, and honest analysis. Erebus aims to define terms precisely, identify operating conditions, and avoid presenting possibility as demonstrated fact.
Durability, maintainability, and second-order effects matter. Decisions are evaluated not only for what they enable immediately, but for the dependencies, risks, and maintenance obligations they create over time.
Research is a process of revision. Better evidence should produce better systems, clearer language, and more accurate understanding. Preserving an earlier claim is less important than improving it when the facts change.