from information streams to stacked fields

The term topic space already has a meaning in science: a multidimensional landscape, built algorithmically, where related ideas cluster and unrelated ones sit apart — a digital map of how a body of knowledge is organized. topicspace starts from that foundation and adds two things to it: time (the same space observed at successive moments), and stacking (higher-order fields derived from the structure of the one below). The starting point is the established topic-space concept; the architecture is what gets added.

A stream, by contrast, is fundamentally a transactional view of arrival. It can tell you what showed up and in what order. It cannot tell you where the new arrivals sit relative to one another, where they are concentrating, or how the underlying structure has shifted. Those questions require a spatial representation.

Treating information as a field exposes seven operational primitives that simply do not exist in stream form:

Each of these is a measurable property of a region of the space, computable at any point in time. They give the system a vocabulary for talking about structure rather than just events.

Once information is represented as a field, a second layer becomes possible. Expectations — forward views about what the current state appears to support — are not observed directly in the raw information stream. They are inferred from the structure of the information field. The information field gives a model of the current state; the expectation field gives a model of what that state appears to support next.

The progression is not just from data to summary. It is from one field to another, with the second field operating on different objects and answering a different class of questions:

The shift between layers is structural. Each layer’s objects, operations, and meaningful properties are different from the one below it. The information field has signals; the expectation field has expectations. The information field measures density of attention; the expectation field measures support, conflict, and lifecycle of inferred views.

Expectations can themselves be tracked as a field over time. Once expectations are represented explicitly, they can be observed as structured objects with position, density, drift, support, conflict, and lifecycle. They persist, weaken, split, recombine, or disappear. That gives us a third class of object: not raw information, not just inferred expectations, but the evolution of inferred understanding through time.

The result is a stack:

At the moment, L1 and L2 are operational and L3 is partially implemented; L4 is the closing-the-loop layer that connects inferred structure back to empirical outcomes. The stack is not aspirational only — per-layer artifacts exist for L1 through L3, with L4 currently seeded by per-actor reliability scoring from a 6-month backtest.

Today, this shows up in topicspace as actor-level forward expectations scored by historical edge, plus field views over how those expectations cluster and evolve.

The reason for stacking is that each layer answers a class of question the one below it cannot. Capabilities are cumulative — every higher layer presupposes the ones beneath it and exposes new operations on top.

A stream cannot tell you where structure exists. A field cannot tell you what the structure implies. An expectation field cannot tell you how its implications change under new evidence. A revision field cannot tell you which kinds of implications are empirically useful. Each layer’s value is the question class it makes available — not just additional detail at the same level of abstraction.

The most concrete view of the architecture is the topicspace homepage. Each card on the page is composed by the four layers working together; the card is the visible product of the stack.

For any single actor, a card aggregates:

• L1 — read. The actor's current relationship between narrative pressure and price, drawn from the information field (e.g. story not being paid, price confirming narrative).
• L2 — forward view. A generated expectation derived from the field state: a headline, a direction, and a conviction score.
• L3 — revision. Daily refresh of the expectation field against the new information state. Persistent expectations stay; weakening ones lose conviction; rotational shifts surface explicitly.
• L4 — trust. A reliability classification from a 6-month backtest of directional accuracy. Determines how literally to read L2.

The figure below shows that composition explicitly. The central card is a schematic of a homepage actor card; the labels around it identify which layer drives each region. The row beneath shows the same card structure with different inputs.

The same composition produces very different outputs depending on each layer’s state for a given actor. AMD, CRM, and AMZN are engine-trusted with strong L1 reads; NVDA is uncertain at L4 even when L1 looks clean — L4 modulates how literally L2 should be read.

The feedback loop is most visible at the actor-page level. When L4 marks an actor engine_inverted, the actor page flips L2's direction before showing it — performance reaching down into the expectation layer to correct what would otherwise be a misleading signal. The same engine that generated the expectation is also the engine that, on the next refresh, will measure whether expectations of its type tend to hold.

The ambition is not a better dashboard or a better retrieval layer. It is a system for mapping how meaningful structures in a domain relate, cluster, shift, and evolve over time. The immediate use case is the AI ecosystem, but the underlying idea is more general: if a domain can be represented as a structured field of information over time, then higher-order fields may be derived from it, each capturing a different layer of understanding.

Another way to state the architecture:

Each layer adds what the layer below it cannot provide. That is the architectural property that makes the stack worth building. A stream can tell us what arrived. A field can tell us where structure exists. An expectation field can tell us what that structure implies. A revision field can tell us how that implication changes under new evidence. A performance field can tell us which forms of inferred structure are empirically useful.

The practical message: modeling information streams as a field gives us the right primitives for tracking a changing domain. That field makes it possible to infer a second layer of system expectations. Stacking one or two more fields above that opens a path toward a system that can represent not just information, but evolving domain understanding.

That is the direction topicspace is moving toward.