Enterprise context, authored from process.

One. Process is the spine of enterprise context. Everything a company knows about how it operates, the people, the policies, the systems, the data, the decisions, hangs off how the work actually runs. Remove the process and the rest is a pile of facts with nothing holding them in relation to each other.

Two. The category skipped the process layer. Catalogs describe where data sits. Diagramming tools describe what someone drew. Neither describes the work, and the work is what a new joiner, an auditor or an agent is actually asking about.

Three. Context is a human prerequisite, not an AI prerequisite. The process expert retiring in March, the new hire who takes six months, the map in SharePoint from 2019: none of those problems were created by AI. AI made them fundable by making the cost of not solving them suddenly legible.

Four. Process is the unifying plane. People, org, data, systems, policy, performance, and both human and agentic decisioning all attach to it. That is not a claim about ambition. It is a claim about where those things naturally hang.

Five. We author context from artifacts you already have. The output is a queryable, evidenced, human-confirmed record. Not a diagram. Not a wiki. A record you can ask questions of, and check.

A map answers exactly one question: the one it was drawn to answer. Ask a different question and you commission another picture. That is not a criticism of mapping. It is a description of what a static artifact can do.

The workshop is not the problem either. Forty people in a room surfaces tacit knowledge and manufactures alignment, and both of those are real. What changed is the input. Most of what a workshop reconstructs from memory is already written down, in the design documents, the standard operating procedures, the runbooks, the screenshots someone made for a training session in 2019.

So the argument is not that the workshop was wasteful. It is that the 80% already written down should take days rather than months, leaving the workshop for the 20% that never was.

The record is not a map with notes on it. It is a stack, and each layer is queryable against the ones below it.

Provenance is not a fifth layer. It is a thread running through every layer above. Each item cites the page of the document it came from, so any layer can be challenged by anyone, at any time.

The process layer follows OCEL 2.0, an open standard for process records. The practical consequence is that the record is not ours to hold hostage. It exports whole, on every tier, including the free trial.

Four moves, and the words for them never change: ORGANIZE the map, INGEST what you already wrote, AUTHOR the record, USE it. Organizing starts from one of six reference maps rather than a blank canvas.

The authoring step runs in three stages.

Everything that carries meaning but is not flow lands as one of twelve types. Each attaches to the most specific thing it concerns, a single event or a single object, rather than to the document as a whole. That specificity is what makes the record answerable later.

Every element carries whether it was stated or inferred, a confidence score, the quoted sentence it came from, and its review state. Inferred elements render differently from stated ones, so the difference is visible rather than buried in a property panel.

The ranking rule is fixed: confirmed beats reviewed, reviewed beats derived, derived beats auto-placed. If the source does not say, the record says nothing. Empty beats invented.

It drafts. You confirm. Nothing confirmed is ever overwritten by a later run. That is the whole safety model, and it is deliberately boring.

Browse a full worked example, derived end to end from a synthetic document, with provenance on every element.

Three rungs, and we are on the first. Define is what ships today: the record of how the work is supposed to run, drafted from your documents and confirmed by people. Mine is next: importing event logs so the record can be compared against what actually happened. Operate is the platform step, where human and agentic decisions run against the record rather than beside it.

Naming the rungs matters because the second one is not shipped. Any diagram on this site that shows the layered record marks the event-data layer as roadmap inside the diagram itself, not in a footnote elsewhere on the page.

It is not a diagramming tool. If what you want is a picture for a slide, this is a heavy way to get one. The diagram here is one view of the record, not the record.

It is not a data catalog. Catalogs describe where data sits. This describes how the work runs and attaches the data to it.

It is not process mining. Mining tells you what happened, from event logs you already emit. This tells you what is supposed to happen, and why, from documents you already wrote. They are complements, and the second rung of the ladder is where they meet.

It is not a replacement for talking to people. It is a way of arriving at that conversation with the 80% already written down.