Research programme
The questions RJV investigates, how it investigates them and what has and has not been established. Objectives are kept apart from results.
On this page
How the research is done
Every line of work connects six activities. The question defines what is unknown and under what conditions. The theory offers candidate explanations and states their assumptions. The method decides which derivations, observations or experiments could tell the explanations apart. Execution carries them out under control. Verification examines correctness, uncertainty and reproducibility, and records competing explanations. Knowledge records the method, the evidence, the limits and the conclusions that remain justified.
RJV's own proposals are recorded so that they can be shown to fail. Definitions, assumptions, methods, results and limits are written down, and the evidence register can mark a claim as superseded or withdrawn. RJV intends that a result which cannot be reproduced will be investigated, and that a conclusion resting on an assumption found to be false will be revised. No result has yet been independently reproduced, and no result has yet been withdrawn.
Objectives and results are different things
Each claim in the table below is labelled as established, observed in a stated test, proposed, or an ambition, so that objectives are kept apart from results. The rest of the page describes how the research is done and how to challenge it.
| Area | Question | Established so far | Not yet established |
|---|---|---|---|
| Specified and checked execution | How far can the part of a computation that is specified, checked and executed under authority be extended? | Observed: the platform enforces authority at execution, runs workflows from a closed validated language and records the evidence of each change. The checks are automated tests. | Formal proofs of any property. The platform's tests show behaviour under the cases tested, and do not prove the properties hold in general. |
| Accountable agent execution | Can a model's proposals be kept separate from authority to act, without making the agent useless? | Proposed design, partly built: a model's output cannot authorise an operation, and approvals are bound to the exact action approved. | An evaluation of agents completing real objectives under these rules. No such evaluation has been published. |
| Engineering method | What makes automated agents finish objectives, rather than accumulate activity? | Practice: RJV builds its own platform under written rules for finding existing code, planning whole changes and verifying them. | A measured comparison of completion and regression rates against another method. |
| A fair economy between people and capable software agents | What rules let humans and digital agents exchange value and resources without either side holding unchecked control? | Proposed. The mission page states the position and its limits. | Everything else. RJV has a ledger for customer billing, but it has built no market, token or exchange of value between people and agents, and nothing here is evidence that one will exist. |
| Reproducible science | Can claims be published so that another person can test them? | The evidence register states the claims RJV makes about its own platform, each with its grade, and the platform keeps a record of every change to them. Claims made elsewhere on the site are not yet in the register. | Independent reproduction of any RJV result. |
What has been published
The publications page lists RJV's papers, guides and notes. Each has its own page with its date and status. The evidence and limitations page lists the claims recorded about RJV's systems, grades each one by how it was tested and shows what settled it. It also says where the register has gaps.
Working with RJV on research
RJV welcomes criticism of its methods and results. A result that does not survive examination should be withdrawn. To challenge a result or propose a collaboration, write through the Work with Us page.
- Published
- 24 October 2026
- Revision
- 9