The data layer decides what the system actually knows.
The Data category answers one question: how do years of messy company information become usable by software and machine intelligence? Everything above it — retrieval, reasoning, automation — inherits whatever this layer gets right or wrong.
The model is not the whole product.
A capable model reasoning over records that disagree with each other produces confident, well-written, wrong answers. Real enterprise systems are constrained far more by data quality, identity, chronology and provenance than by model choice, which is why this is the first category rather than an implementation detail buried under the others.
The work here is unglamorous and mostly deterministic. Records arrive from systems that were never designed to be read together, in formats that changed over the years, describing the same customer or employee or transaction under different names. Making that material queryable — without inventing anything that was not in it — is the whole job.
Four problems, in the order they bite.
- Access. Getting records out of the systems a company already runs, repeatedly and without asking its staff to export anything by hand.
- Identity. Deciding when two records describe the same real-world thing, and recording the decision rather than silently merging.
- Chronology. Keeping the difference between what is true now and what was true when the event being analyzed occurred.
- Provenance. Capturing, at the moment of ingest, where every value came from — so a later answer can be traced back to a source rather than defended by assertion.
Ingestion and cleaning make records readable. Entity resolution decides what they refer to. Temporal data preserves when each statement applied, and provenance records which system asserted it. Labeling and evaluation datasets exist so the interpretive steps can be measured rather than trusted, and retrieval preparation shapes the result into something the knowledge layer can actually search.
Two of these have close counterparts in Knowledge Architecture, and the split is deliberate: this category is responsible for capturing chronology and lineage at the source. Reasoning over them — deciding which source wins a conflict, or reconstructing the state of the business on a past date — belongs to the knowledge layer.
Eight capability pages sit in this category. Availability in the platform is not a claim that every one is used on every engagement.