Synaptic Structure

A fixed grammar of seven relationships. Decisions can be wired together using a closed vocabulary: SUPERSEDES, REQUIRES, CONSTRAINS, CONTRADICTS, EXTENDS, IMPLEMENTS, RELATED_TO. The vocabulary is closed for a reason. Open relationship sets devolve into folklore; every team invents its own variant of “depends on,” and the corpus stops being interoperable across teams. With seven labels in every organization, a query about CONSTRAINS means the same thing across every team that uses the platform.

Append-only by design. Relationships are never deleted. They are retired with a reason and an actor, on a timestamp. The history of why two decisions used to be connected and no longer are is itself part of the corpus. Most knowledge bases discard this layer. An organizational memory cannot afford to.

Contradiction as workflow. When the platform identifies that two decisions contradict each other, the architect has three structured paths: supersede the older decision (one decision wins, the other inherits a SUPERSEDES edge); reconcile via a third decision that explains how both can coexist; or accept the contradiction as scope-limited, with a required scope note explaining the divergence. Contradiction is the most interesting signal a decision corpus generates. The workflow exists so architects engage with it rather than route around it.

Cycle protection at the database. Directional relationships cannot form cycles. The constraint is enforced where it has to be enforced: in the database, as a trigger that walks the graph on every insert and rejects edges that would close a cycle. Application-layer constraints in multi-user systems are advisory. A decision corpus cannot afford advisory.

The engine proposes; the architect decides. As an architect captures a new decision, the platform surfaces likely relationships from the existing corpus. Suggestions are typed: the platform proposes a relationship category, not just a similarity score. The architect accepts, modifies, or rejects. The manifesto’s division of labor between the platform and the architect is enacted on this screen.

Visualized as a graph. A project’s typed-relationship corpus renders as a graph view, with the seven typed edges drawn distinctly and each node linking back to its decision. The graph is a read-only consequence of the structured corpus, not an artifact that needs separate maintenance. Architects who think structurally finally have a structural view of their decisions.