How Ontology works

Turn messy material into answers you can inspect.

Ontology names the important things in a domain, makes their relationships explicit, and selects only the context needed for the question at hand.

The result can be an explanation, a map, a next action, or a memory—but it always carries enough evidence to show why it exists.

The recurring move

01
Name the material
Stories, projects, events, evidence, decisions, and observations.
02
Connect what matters
Typed relationships preserve distinctions a flat document would blur.
03
Select bounded context
The question determines the smallest useful neighborhood.
04
Return an answer with a receipt
The result exposes its source, path, snapshot, and limits.

One question, all the way through

A worked Parallax answer

Question

“How is Hannibal’s campaign framed across perspectives?”

The question is not answered by searching for a paragraph. It is answered by following the relationships that frame the campaign.

Source
The existing story stays the source of truth.
9 portable Parallax story files are projected into a versioned snapshot with 219 typed objects and 333 relationships.
Structure
Perspectives, chapters, events, and places remain distinct.
A perspective frames an event. A chapter contains it. A waypoint places it on the route. Those are different claims, so Ontology keeps them separate.
Selection
Only the relevant neighborhood enters the result.
The selector parsed “hannibal,” ranked 3 candidates, and retained 3 objects within a 900-token budget.
Result
3 perspective objects are selected.
carthaginian, numidian, roman enter the bounded context slice together with their traceable relationships.
Boundary: selection identifies relevant framing objects; it does not decide which perspective is historically true.

The model transfers

The value is not one universal interface. It is a reusable way to make a domain answerable without erasing what makes that domain specific.

Parallax

history

Stories become navigable through chapters, events, places, and points of view.

Core

software factory

Projects become explainable through goals, evidence, blockers, and dependencies.

PK memory

agent context

Past encounters become bounded context rather than an undifferentiated history dump.

Evidence, not ceremony

A useful result should be inspectable. The receipt makes the answer’s construction visible and keeps confidence proportional to the evidence.

That is the difference between a plausible answer and one a person—or another agent—can responsibly act on.