Research integrity
How Aeviant labels evidence, defines failure, verifies sources, and corrects its scientific record.
These standards govern how Aeviant distinguishes published knowledge, interpretation, objectives, and results. Computational predictions are not treated as experimental validation.
The governing standard
Aeviant does not protect a program from evidence that contradicts it. Advancement and termination criteria are defined before the relevant work where practicable, and a failed pre-registered criterion can stop a program regardless of how compelling the original idea appeared.
As founder Immanuel Martins puts it: “I don’t need AVX-1 to be right. I need reality to tell me whether it earns the right to exist.”
Evidence is labelled
Public scientific claims are assigned to one of four categories so that a reader can distinguish knowledge in the literature from Aeviant’s interpretation and objectives.
- Established biology: findings supported by the broader scientific literature.
- Preclinical evidence: observations from non-clinical experimental systems.
- Aeviant interpretation: the company’s reading of existing evidence and the hypothesis it draws from it.
- Program objective: a property or result the program is intended to achieve but has not yet demonstrated.
Prediction is not proof
Computational methods are used for target characterization, hypothesis generation, prioritization, and adversarial testing. Their outputs are predictions awaiting wet-lab evaluation. They do not establish binding, pharmacology, safety, efficacy, or activity in a living system.
Aeviant identifies the absence of experimental validation wherever it is relevant to a public claim.
Sources, contradiction, and correction
Load-bearing scientific statements are checked against primary sources. Reviews may provide background, but they do not replace examination of the underlying evidence. Aeviant maintains a corrections record and revisits public claims when program status or relevant evidence changes.
Proposed mechanisms and conclusions are subjected to adversarial review. Automated or AI-assisted review is identified as such and is never presented as independent human review. Contradictory and negative results are retained rather than edited out of the decision record.
Public disclosure
Scientific openness does not require disclosure of unpublished proprietary detail. Public materials describe the program thesis and its evidentiary status. Technical information beyond that public thesis is discussed only under an appropriate confidentiality agreement.
Last reviewed .