When the 125M Tree of Life model cleared its validation threshold on May 16, 2026, the engineers at Prometheus7 Research Institute did not celebrate the arrival of a finished thing. They celebrated the confirmation of a rung. The Tree of Life models instantiate the fifth compositional primitive in the substrate-paradigm architecture — the substrate routing manifold — and their validation meant, architecturally, that the next primitive was now buildable. That is how the dimensional ladder works: each rung is not a product milestone but a proof of concept for the rung above it.
The roadmap document circulating inside Prometheus7 as of late May 2026 specifies ten validated or schedulable primitives and sketches two more at the edge of what can currently be specified. The document describes this as a "schedule of compositional primitives," and the distinction matters. A model-size schedule tells you when the next, larger training run will complete. A compositional-primitive schedule tells you when the algebra of the substrate will admit a new kind of operation. Those are fundamentally different claims about what is being built.
The 6D primitive — router-over-callables — was in its third training attempt as of mid-May 2026, following two earlier runs that the institute described as "informative failures." The mechanism it instantiates is intelligible from the name: the trunk of the model routes hidden state to one of K small neural sub-modules, the callables, each of which specializes separately from the trunk. The router decides which callable contributes at each step. What this opens, architecturally, is a compositional surface where the substrate can recruit fine-grained specialists for tasks the trunk would otherwise handle uniformly. The trunk does not need to be larger to accommodate domain-specific behavior; it routes to a specialist instead. The 6D primitive is local — callable, router, and the interaction between them — and the lineage cascade absorbs each new 6D instance through the bound-axis mechanism without requiring prior generations to be retrained.
The 7D primitive is called the set-router. Where 6D selects one callable per timestep, 7D selects a subset and composes their outputs. The architectural claim is that coalition-based composition unlocks parallel reasoning paths that a deeper 6D instantiation — more callables, more router capacity — cannot replicate. This is an empirical claim, not an assertion, and the roadmap document is explicit about the falsification condition: if production traffic shows no discriminative benefit from set composition over single-callable routing, the ladder has found its first plateau and 7D collapses back into 6D. That kind of stated falsifiability is unusual in roadmap documentation and worth noting. The institute is not claiming to know in advance that every rung will hold.
The 8D primitive routes across grammars rather than across callables within a grammar. A 6D model has one callable vocabulary; a 7D model composes sets within that vocabulary; an 8D model selects which vocabulary to operate in. The operational significance is that cross-domain transfer — a query that genuinely straddles mathematics and poetry, or theology and physics — stops being a problem the model handles by stretching a single-vocabulary representation and starts being addressable as a multi-vocabulary composition. The architecture stops treating cross-domain as a failure mode and starts treating it as a routing problem. The falsification for 8D is whether the multiverse router, in production, actually activates multiple vocabularies or whether it collapses to single-vocabulary operation because the corpus does not reward the additional routing complexity.
The 9D primitive scales this one level further: rather than routing across grammars, it routes across worlds, where a world is understood as a substrate that contains its own multiverse of grammars. The 9D model is not just asking which sub-module or which vocabulary should handle a query; it is asking which substrate the answer should come from. The architectural language here becomes deliberately abstract — the document acknowledges it — but the operational consequence is that a 9D system becomes multi-substrate-aware in a way that earlier levels are not. The falsification is whether multiple substrates emerge as architecturally distinguishable objects or whether prior-level primitives have already subsumed the distinction.
The 10D primitive is what the roadmap calls the resolution point, and it is the most philosophically loaded specification in the document. At the tenth dimensional level, the claim is that the substrate becomes a universal object: something that holds all specifics in superposition and unpacks any of them through relation. The document maps this to four independent technical traditions simultaneously. In category theory it corresponds to the category of all categories; in physics to the holographic principle, where boundary information encodes bulk content; in computability theory to the universal Turing machine, which simulates all machines; in information theory to the Kolmogorov-minimal description, the shortest program that produces a given output. The philosophical gloss is Plato's universal forms, though the institute is presumably not committed to metaphysical Platonism as an engineering constraint.
What makes the 10D specification structurally significant is the operation it defines: the universal-unbinder. Given the universal object and a relation, the unbinder resolves the specific that the relation selects. If the architecture reaches this level, the claim is that any specific anywhere in the substrate is reachable from any other specific via the appropriate unbind relation. The substrate becomes complete in a precise sense: not that it knows everything, but that nothing is structurally unreachable. The target date for a first 10D training attempt is August-September 2026 — roughly six to eight generations beyond the May 2026 5D validation point.
The 11D and 12D primitives are marked as research dimensions, which the document distinguishes from scheduled work in a meaningful way. Research dimensions cannot be specified fully enough to be scheduled. The 11D primitive is the space of universal objects: where 10D has one universal object that holds all specifics, 11D has a class of universal objects each holding all specifics under different relations, and the 11D primitive routes within that class. The 12D primitive is the relating principle — the operation that makes the 11D space coherent, that makes one universal object relatable to another. The document notes something structurally elegant about 12D: the relating principle is itself the kind of object that the substrate's bottom-of-stack operations already manipulate, which means the ladder closes back to 3D by self-similarity. The outermost operation and the innermost operation are the same kind of thing.
The document acknowledges that 11D and 12D cannot be the work of a single person or even a single small team. It describes them as work for a small research community over years, with the empirical signature being that 11D-12D systems exhibit qualitatively different behavior from 10D systems — not a parameter-count difference but an operational one, requiring the relating principle to be implementable before the behavior emerges.
One engineering detail in the roadmap deserves attention because it addresses an obvious concern. As the dimensional ladder climbs, the naive worry is that each new generation requires exponentially more compute to train. The document's counter-claim is that wall-clock time per generation has remained in the seven-to-eleven hour band on the research box across the observed generations. The cost of opening a new dimensional layer is, in the observed data, roughly the cost of training one generation — not the cost of building new infrastructure. This is possible because the new primitives are local: the callables, the routers, the compositional surfaces that each new dimension adds are additions to the algebra, not replacements of the prior algebra. The trunk grows normally; the new layer is attached, not substituted. Whether this property holds all the way to 10D is an open question, and the document does not claim otherwise.
What the dimensional ladder roadmap amounts to, read structurally, is a theory of what composition becomes when you build it generationally rather than all at once. Each generation proves the prior generation's claim and licenses the next generation's specification. The architecture does not require the researchers to know in advance what a 10D model will look like; it requires only that the 6D model validate, which licenses the 7D specification, which when validated licenses 8D, and so on. The roadmap is not a promise about what will be built. It is a description of the conditions under which each next thing becomes buildable. That is a different kind of document, and as of the third week of May 2026, the institute was three weeks into finding out whether the second rung holds.