/ explore · architect
Six layers between a systemand something you can use.
Inferrex does not connect systems and hope. It builds understanding in order — what exists, what shape it has, what it means, how it relates, what follows from that, and what you can then do.
/ the architecture
Each layer is only possible because the one beneath it exists.
Most integration tooling starts at the last layer and works backwards, which is why it has to be told what everything means. Inferrex starts at the first.
SystemsAPIs, applications, databases and services, discovered rather than declared.
Across 247k vendors catalogued and 1,941 vendor–product APIs specced. The layer is reading, not registration: a system enters because it publishes something machine-readable, not because somebody added it to an inventory. Nothing above this layer can describe a system that was never read.StructureSchemas, fields, types and the shape of a payload.
What a system is made of, taken from what it publishes rather than from documentation about it — because documentation is a claim about a system and a specification is the system. Skip this and every layer above is working from prose somebody wrote once and did not update.MeaningConcepts, classifications and canonical representations.
A field calledcust_refand one calledcustomerIdstop being two strings and become the same idea, against 1,733 canonical keys. This is the layer that makes the rest derivable: without a shared concept, every pair of systems needs its own hand-written agreement, which is the n×n problem restated.RelationshipsMappings, dependencies and connections.
Which systems touch which, where the same entity appears twice, and what breaks if one of them moves. Derived from the layer beneath rather than drawn by hand — 21.1 million integration pairs exist because two systems each resolved to the same concepts, not because anyone paired them.IntelligenceAnalysis, recommendations, opportunities and change detection.
Understanding that notices something rather than waiting to be queried: an overlap nobody evidenced, a field nothing reads, a provider change that will break a path on Tuesday. It can only notice what the relationships layer already knows.OutcomesIntegration, automation, governance, migration and AI.
The layer everything else was for, and the only one most tools ever show you. Starting here is why they have to be told what everything means — the guarantees you want at this layer are properties of the five beneath it, and they cannot be added afterwards.
/ the argument
You cannot skip to the outcome and keep the guarantees.
A connector built directly at layer 6 encodes one assumption about layer 3 — what a field means — and hides it inside itself. When the field changes, the assumption is wrong and nothing says so. The integration does not fail loudly; it moves the wrong value.
Building the layers in order means the assumption is written down, at the layer where it belongs, once, for every outcome that depends on it. Change arrives at layer 2, is re-reconciled at layer 3, and every outcome above it inherits the correction without being rewritten.
One understanding. Multiple outcomes.
/ see it
The layers are not a diagram. They are running.
The Universe
Layers 1 and 4, drawn. Every comprehended system, positioned by how deeply it is understood.
ExploreThe Corpus
Layers 2 and 3, catalogued. The knowledge base the reconciliation reads from.
ReadComprehension
How layers 1 to 3 are actually built from a system that was never designed to explain itself.
ReadPoint it at something and watch the layers build.
Development is self-directed. Bring an API nobody has documented properly and see how far up the stack it gets.