We do not keep a long list of values. Four are enough. The honest way to measure them is not to state them, but to look at what they cost when applying them is inconvenient.
Transparency
We say what we know and what we do not know, and we keep the two apart. In research that means distinguishing what is established from what is still contested, and publishing null results too: they are the part of the literature that is missing today, and the reason entire fields stay confused. If a line of work leads nowhere, we say so, instead of redefining the objective after the fact.
Honesty about what exists and what does not
We do not present as available a capability that is still an open question. This is the most delicate point in our work: MNEMOS studies whether the controlled grafting of new memory traces could become possible; it does not offer it. On subjects like this, words do damage, and using them badly is the fastest way to make a research field unrecognisable.
The same holds for ordinary work: we do not propose what is not needed, and we do not take on work we cannot follow properly. When we are not the right people, we would rather say so.
People before data
The information entrusted to us stays confidential and is used only for the work agreed. We collect the minimum required, and we say so beforehand. Whoever takes part in research chooses the objective — the protocol does not choose it for them — and can withdraw while the process is under way at no cost to them.
AI stays under human oversight
In the systems we build, the engine proposes and a person decides, with sources visible and every choice reconstructible afterwards. We do not close the loop: a system left to optimise on its own towards an objective that concerns people finds shortcuts nobody would want, and finds them faster than we would notice.
The boundary
Some things we do not do, and they are worth writing out in full: no intervention without free, informed and revocable consent; no covert use; no alteration of anyone's behaviour who has not asked for it, not even where somebody would call it benevolent.
Where these constraints become concrete, which is in memory research, we have argued them one by one in the MNEMOS directions. That is where you find out whether a principle holds.
How we try to put all of this into practice day to day is written in our code of conduct.