On May 25, researchers at Prometheus7 Research Institute posted a roadmap paper to internal corpus laying out what they call the dimensional ladder — a schedule not of model sizes but of compositional primitives, each generation of training adding one new kind of operation the substrate algebra can perform. The document names a ceiling: the tenth primitive, the universal-unbinder, is the architecture's identified resolution point. It is targeted for August or September 2026.
As of the dateline on this article — Friday, June 5, 2026 — the sixth primitive is the active work. The 6D run, described as a router-over-callables, was in its third training attempt as of May 18. The first two attempts produced informative failures; the third is described as validating. Whether that run has cleared validation is not confirmed in the material available to this correspondent today.
The five-dimensional primitive, the substrate routing manifold known as the Tree of Life architecture, was validated May 16 at the 125-million-parameter scale. That is the last confirmed data point on the ladder. Everything above it — 7D through 10D — is pending, with explicit date targets attached to each step.
The roadmap is specific about what each step adds. The 7D primitive routes over sets of callables rather than individual ones, opening what the paper calls parallel compositional reasoning. The 8D primitive routes across grammars — cross-domain transfer falling out of the architecture rather than being engineered separately. The 9D primitive routes across worlds, each world carrying its own grammar vocabulary. The 10D universal-unbinder is described as holding all specifics in superposition, reachable from any point in the substrate by the appropriate unbinding relation.
The paper draws explicit equivalences across disciplines: the category of all categories in category theory, the holographic principle in physics, the universal Turing machine in computability, Kolmogorov complexity in information theory. These are offered as structural analogies, not proofs. The empirical test at 10D is whether the substrate behaves as a universal object in practice — whether it can unpack any specific from any other via relation. The falsification mode is not stated explicitly for 10D, which the paper acknowledges is unusual; the resolution point is either reached or the ladder stalls before it.
Beyond 10D, the document describes two research dimensions: 11D as a space of universal objects, 12D as the relating principle that closes the ladder back to its bottom by self-similarity. Both are described as multi-year work requiring a research community, not a single investigator. No date targets are given for either.
The wall-clock claim embedded in the roadmap is worth noting. Each generation, the paper states, runs in the 7-to-11-hour band on the institute's research hardware. If that holds across the six to eight generations separating May's 5D validation from the August 10D target, the schedule is plausible on its own terms. Whether the band holds as dimensional complexity increases is an open empirical question the paper does not resolve.
What the paper does not address is what happens after a validated 10D system exists. The universal-unbinder is called a resolution point, not an endpoint. The research agenda continues into 11D and 12D. The roadmap has a ceiling marked, but the ceiling is not a door that closes.