ORIAS / Research
What we are trying to find out.
Chimera is not a set of claims. It is a set of questions, and none of them is settled. These are those questions, written in the form that would let a result disprove them.
- Can useful engineering experience be made trustworthy?
- An experience that cannot be checked is a story. The rest of the programme depends on this one.
- Can software retain what was learned across people, models and sessions?
- Retention is not enough by itself. Retained knowledge has to carry its scope, its evidence and its expiry, or it becomes misleading with age.
- Does accumulated repository intelligence measurably improve the next actor?
- Hold the task and the repository constant, measure a baseline, then measure the next actor with the accumulated world. If the world does not measurably help, we have built a diary, not intelligence.
- Does that benefit survive when the actor changes?
- If the benefit exists only for one model, the world is a cache for that model. We test by substituting actors and watching for negative transfer. If nothing does, the idea is wrong.
- Can knowledge remain valid as software changes?
- Dependencies move and assumptions expire. We measure staleness and record scope, then test whether knowledge can be kept current — or correctly forgotten.
- Can software compare possible futures before committing to one?
- We test whether alternative branches can be executed or simulated, compared against each other and against no action, with the evidence settled independently.
- Can repeated reasoning become cheaper capability?
- If a stable pattern still requires a frontier model every time, the economics never improve.
- Can software safely propose its own next state?
- The interesting version of this question is about evidence and governance, not autonomy.
On models
The objective is not smaller models for everything. It is movement toward the cheapest reliable form of intelligence appropriate to each stable problem: sometimes a procedure, sometimes a verifier, sometimes a specialist, sometimes a frontier model, sometimes a person.