Inference

AUM distinguishes between what ASUME can observe and what it concludes from those observations. Suppose a company publishes twelve new vacancies for cybersecurity roles. ASUME can establish from the source that the company is hiring for twelve cybersecurity positions. That is directly supported by the available information.

ASUME may then consider a statement such as:

The company appears to be expanding its cybersecurity capability.

That statement is different. The company may never have said that it is expanding its cybersecurity capability. The conclusion is inferred from the hiring activity and, potentially, other relevant evidence.

A further conclusion might be that cybersecurity is becoming a higher priority for the company. That requires a stronger basis because it moves further from what was directly observed. Additional information about investment, organisational changes, product development, regulation, or management statements may strengthen or weaken that interpretation.

A simplified distinction is therefore:

What was observed → What the evidence supports

Inference operates between the two without making them indistinguishable.

The complete state of a company cannot be observed directly. Some properties can be established from available information, while others become visible only through the signals they produce.

A company may never publish a statement saying that it has an implementation-capacity problem, that a particular capability has become strategically important, or that an existing process is creating a constraint. It may instead increase hiring, change its product, reorganise a function, introduce new integrations, alter documentation, or behave differently across several areas.

Those observations may collectively tell us something meaningful about the company even when no single source states the conclusion directly.

Inference allows ASUME to move from the question “What information is available?” to the more useful question “What does this information currently support about the company?” This is the role inference plays inside AUM.

Evidence and inference have different roles. Evidence provides the observable basis for a conclusion; inference is the reasoning applied to that basis. A correctly cited source can therefore support an inference without making the inference itself an observed fact.

Consider a company that has introduced enterprise deployment features, increased implementation hiring, and published several new enterprise customer cases. Those observations may together support the conclusion:

Enterprise expansion appears to be a current priority.

A user should still be able to inspect the evidence separately from the conclusion and ask whether the interpretation makes sense. The underlying information may be accurate while the proposed conclusion remains debatable, incomplete, or compatible with another explanation.

This separation is important because evidence quality and inference quality are related but not identical. A reliable source can be interpreted badly, while a reasonable inference can still remain uncertain because the available evidence is incomplete.

Inference sometimes requires something that the evidence does not directly establish. When that happens, ASUME uses assumptions to keep the missing step visible rather than silently presenting it as fact.

Suppose a company has substantially increased hiring in one function. The hiring activity may support the possibility that the function is expanding. But that conclusion may depend on something the available information does not establish, such as whether the vacancies represent additional capacity rather than replacements for people who have left.

If that distinction cannot be observed directly, it matters to the interpretation. The conclusion should therefore remain connected to the assumption on which it depends.

This makes inference easier to inspect and revise. If later information shows that most of the positions were replacements, the assumption can be challenged and conclusions that depended on it can be reconsidered.

Assumptions therefore allow ASUME to reason under incomplete information without pretending that the missing information was already known.

Inference rarely depends on one observation in isolation. What matters often becomes clearer when several related changes are considered together.

Suppose a company launches a new enterprise product, increases implementation hiring, expands its enterprise deployment documentation, and begins publishing more enterprise customer cases. Any one of those observations may have several explanations. Seen together, however, they may provide a stronger basis for understanding the company's current direction.

This does not mean that several related observations automatically establish a conclusion. It means that their relationship can be informative. A product launch may give hiring activity a different meaning, while new documentation may provide additional context for both.

ASUME therefore considers relevant information in relation to other available evidence rather than treating every fact independently. This is one of the ways inference contributes to understanding rather than simply summarising what has been retrieved.

The same evidence can support more than one explanation.

A company that substantially increases engineering hiring may be expanding its business, replacing departing employees, developing a new product, bringing outsourced work in-house, entering a new market, or doing several of those things at the same time. The fact that one explanation sounds plausible is not enough to establish it.

ASUME therefore considers what the wider evidence supports before settling on a stronger interpretation. Additional information may favour one explanation, weaken another, or show that several remain possible.

Where the evidence does not justify choosing between them, the appropriate result may remain qualified. ASUME can preserve uncertainty rather than forcing the available information into one confident explanation.

Inference is therefore not about finding a story that fits the evidence. It is about determining which conclusions the available evidence can reasonably support and how strongly they should be represented.

Company inference normally takes place under incomplete information. Evidence can be indirect, outdated, contradictory, limited in scope, or compatible with different explanations. The quality of the conclusion therefore depends not only on whether some supporting information exists, but also on what remains unresolved.

AUM does not require every inference to resolve into a definitive statement. Depending on the evidence, the appropriate result may be a strongly supported conclusion, a qualified conclusion, competing interpretations, an unresolved assumption, or an explicit evidence gap.

This matters because fluent language can easily make weak and strong conclusions sound equally certain. ASUME aims to preserve the difference between them. If the available basis is limited, the conclusion should reflect that limitation rather than becoming more certain merely because it can be expressed clearly.

Uncertainty is therefore not something inference is designed to eliminate. It is something the understanding needs to represent when the evidence requires it.

An inference is only as broad as the evidence and assumptions that support it. A conclusion may apply to a particular time period, product, geography, business unit, market, customer segment, or other part of the company's context without applying to the company as a whole.

For example, evidence from one division may support the conclusion that enterprise expansion is a priority within that division. It may not be enough to conclude that enterprise expansion is the company's global priority. Similarly, evidence that strongly supported a conclusion two years ago may provide much weaker support for the same conclusion today.

Keeping this context matters because companies are not static or uniform. Different parts of the organisation can move in different directions, and the significance of evidence changes over time.

ASUME therefore keeps relevant scope connected to its conclusions so that an inference is not applied more broadly or for longer than its basis supports.

An inference is not permanent simply because it was reasonable when it was formed. Companies change, new information becomes available, and earlier assumptions can turn out to be wrong.

Suppose ASUME currently understands that a company is expanding an internal cybersecurity capability. That conclusion may be supported by increased hiring, new roles, and changes to technical documentation. Later, the company announces a major outsourcing agreement covering the same function. The new information may require the earlier conclusion to be weakened, qualified, or revised.

The important point is that ASUME does not need to keep both conclusions as disconnected answers. Because the inference contributes to a maintained company understanding, changes in its evidential basis can change the understanding itself.

At a simplified level:

Existing understanding + new evidence → updated understanding

This allows ASUME to preserve continuity while still changing its view when the information justifies it.

New evidence does not always point neatly in the same direction as existing evidence. A public announcement may suggest one priority while hiring activity indicates another. Recent product changes may conflict with older documentation, while Customer-provided information may introduce context that changes how a public statement should be interpreted.

ASUME should not discard conflicting evidence simply because it makes the current understanding less certain. Nor should it automatically favour information that confirms an existing conclusion.

Where the conflict cannot yet be resolved, it may be more appropriate to weaken a conclusion, narrow its scope, preserve competing interpretations, or leave the disagreement visible until additional information becomes available.

Contradiction can therefore improve the understanding by revealing where an earlier interpretation no longer explains all of the evidence.

An inference produces a conclusion about some part of the company, but an individual conclusion is not the same thing as understanding the company. Understanding requires conclusions to remain connected to the evidence, assumptions, context, uncertainty, and other information that affect them.

ASUME therefore does not treat inference as a sequence of independent questions that disappear once an answer has been generated. Relevant conclusions can contribute to the company's maintained understanding and remain available to later reasoning.

This means a new question can benefit from what ASUME already understands about the company, while new evidence can also revise that existing understanding. Different objectives can then use relevant parts of the same maintained company state rather than requiring the company to be reconstructed independently each time.

Inference is therefore one step in the larger process. It helps establish what the available information supports; maintained understanding keeps that result connected to everything necessary to use and revise it later.

A user should be able to distinguish the conclusion ASUME reached from the basis on which it reached it. For an inferred statement, useful questions include what information was observed, which evidence supports the conclusion, which assumptions matter, whether conflicting evidence exists, what context the conclusion applies to, and what could cause it to change.

This does not mean exposing every internal computational operation involved in producing an inference. What matters is preserving enough of its basis that the conclusion can be understood, questioned, and traced back toward the information that supports it.

A user may agree with the evidence but disagree with ASUME's interpretation. New evidence may reveal that an assumption no longer holds. A conclusion may remain reasonable for one part of the company while being too broad for another. Keeping those distinctions visible makes the inference inspectable rather than asking the user to accept it as an unexplained model output.

This provides one of the foundations for ASUME's approach to traceability.

Inference does not mean guessing what a company might be doing, and it does not mean turning the first plausible explanation into a fact. Generated text is not evidence simply because a model produced it, an assumption does not become true because it is useful, and a conclusion does not become certain because supporting information exists.

Inference also does not mean that ASUME claims access to a company's complete internal state. The conclusions ASUME forms remain bounded by the information available, the strength and relevance of the evidence, the assumptions required, the context in which the evidence applies, and the uncertainty that remains.

The purpose of inference in ASUME is narrower and more useful: to reason beyond what is directly stated while preserving the difference between what was observed and what ASUME concluded from it.

That distinction allows ASUME to form useful company understanding without presenting interpretation as fact.