Coherence

Two systems disagree. Which one is right?

signed contract

ticketing config

The contract promises a four-hour response. The tool is set to eight. A tidy system picks one and moves on. Ontheia keeps both — because the honest answer is that the organization is, right now, holding two commitments that don't agree. Recording that faithfully is the point, not the failure.

“Coherence” is a precise word here, and it isn't a synonym for clean. This page is what it actually means — and what it deliberately refuses to become.

The distinction

Coherence, not consistency.

Consistency is a property a model has on its own: nothing inside it contradicts anything else. Coherence is a relationship between the model and the evidence — and, through the evidence, reality. Consistency is internal to a model. Coherence is how well the model and the evidence hold together.

That difference has a sharp consequence. When two sources genuinely disagree, forcing them to agree makes the model less faithful to what's true. So a contradiction, represented faithfully with both its sources, increases coherence. The model is allowed to hold tension — because the organization does.

The structure shapes the flow, while the flow gradually reshapes the structure. The model isn't a fixed map laid over the organization; it's a bed the evidence carves.

What it promises

Six things a coherent model owes you.

Coherence isn't a vibe; it's a set of commitments, each with a specific failure it exists to prevent.

Fidelity

Reality is never squeezed into the wrong box; when something genuinely new arrives, the model says so.

Failure: forced fits that a human has to correct later.

Groundedness

Every statement traces back to the document, message, or record it came from.

Failure: structure floating free of any surviving evidence.

Identity continuity

One real thing, one record — no duplicates, no records that oscillate between merged and split.

Failure: the same customer, counted three times.

Preserved disagreement

When two systems disagree, you see both, each with its source — never a silent flattening.

Failure: a false single answer — or the opposite, marking everything as a conflict.

Parsimony

The model stays as small as understanding requires; it doesn't mirror the shape of your source systems.

Failure: an entity for every table, an instance for every row.

Temporal intelligibility

The model is held so it can answer what was understood about something, and when that understanding changed.

Failure: history that can't explain a decision already made.

How it's judged

Judged against real cases — never reduced to a score.

There is deliberately no coherence gauge, no percentage, no single number to optimize. The moment coherence becomes a scalar target, it stops measuring the thing and starts being gamed.

Instead, coherence is assessed the way you'd assess judgment: against a small, fixed set of paired cases — real reconciliations with a known better and worse answer — reviewed one dimension at a time. The aggregate numbers are trends to watch, not targets to hit. It's slower and more honest than a dashboard metric, and it's the only kind of measurement that survives contact with the goal.

Why it compounds

Coherence pays forward.

A more coherent model makes the next observation easier to place. When identities are clean and relationships are explicit, each new piece of evidence has fewer plausible interpretations to weigh, needs less surrounding context to resolve, and lands with fewer corrections from a human.

The reverse is just as true, which is why it matters. Every duplicate, every silently-flattened contradiction, every source-shaped entity makes the next decision harder. Coherence isn't housekeeping — it's the thing that keeps the model usable as reality keeps moving.

For the avoidance of doubt

What this is not.

Ontheia borrows techniques from a lot of familiar categories. It isn't any of them, and it doesn't replace the systems you already run.

Not a data warehouse or catalog: The goal isn't one universal schema everything is copied into. There is no single giant model to centralize onto.
Not master data management: MDM asks “is CRM customer 42 the same as ERP account 918?” Ontheia asks the larger question: what does this new information mean, and does our understanding itself need to change?
Not a BI / dashboard layer: If it only explained the organization, it would be a sophisticated dashboard generator. Understanding has to connect to action.
Not a knowledge graph, taxonomy, or vector store: Any of those may be a technique used inside the model. None of them is the point, and none is where the truth lives.
Not an integration bus: The aim isn't to keep more representations in sync. It's to stop needing them synchronized, by cohering with the organization itself.

Where it stands

A definition to be held to.

Coherence is the bar Ontheia sets for itself; whether the prototype clears it is a question with evidence, not a slogan. See how the loop works, or watch it run against a week of evidence — including where it fell short.

Watch it run