Evidence

ASUME may have access to a large amount of information about a company, but not every piece of information is relevant to every question. A source can contain many accurate facts while only a small part of it bears on the particular property ASUME is trying to understand.

Suppose a company has published ten open positions in its implementation function. The hiring activity is observable information. If ASUME is evaluating whether the company is expanding its implementation capability, those positions may become relevant evidence.

They do not, however, establish everything that might follow from them. The vacancies alone may not reveal whether the roles are new or replacements, why the company is hiring, how important the change is, or whether the expansion creates a particular commercial need.

Evidence therefore gives reasoning a basis without automatically determining the conclusion.

At a simplified level:

Source → Relevant information → Evidence → Understanding

The purpose is not to turn every retrieved fact into evidence. It is to identify which information actually matters for the company property being evaluated.

The same piece of information can matter greatly for one question and very little for another.

Suppose ASUME observes that a company has launched a new API, opened implementation roles, expanded enterprise integrations, entered a new country, and appointed a new CFO. All of those observations may be accurate, but they do not necessarily have the same relevance if ASUME is evaluating whether the company is expanding its enterprise implementation capability.

The API launch, implementation hiring, and enterprise integrations may contribute directly to that understanding. The appointment of a CFO may be important for a different question but provide little useful evidence for this one.

Evidence is therefore evaluated in context. ASUME does not treat information as important simply because it exists in an available source. It considers whether that information actually bears on what is being evaluated.

This becomes especially important when ASUME processes large volumes of company information. More information is useful only when the relevant parts can be distinguished from everything else.

ASUME deliberately separates what was observed from what the evidence may imply.

Suppose a company publishes twelve vacancies for cybersecurity roles. ASUME can establish from the available information that the company is hiring for those positions. That observation may then provide evidence for a statement such as:

The company appears to be expanding its cybersecurity capability.

The second statement is different from the first. The company may never have said that it is expanding its cybersecurity capability. It is an interpretation supported by the observed hiring activity and, potentially, other evidence.

Other explanations may still be possible. The roles could represent replacement hiring, organisational restructuring, a temporary programme, regulatory requirements, or a combination of factors.

AUM preserves that distinction so that an inference does not silently become an observed fact. A user should be able to understand whether a statement comes directly from available information or represents something ASUME concluded from that information.

Evidence also remains separate from assumptions.

Evidence has an identifiable basis in information available to ASUME. An assumption becomes relevant where the evidence does not directly establish something that may be needed to continue the reasoning.

Consider a company hiring heavily into a newly created product function. The hiring activity can be evidence that something is changing. Concluding that the company is increasing its total capacity, however, may depend on assuming that most of those positions are additions rather than replacements for people who have left.

If the available information does not establish that distinction, ASUME should not quietly treat it as fact.

This matters because two conclusions can be based on similar evidence while requiring very different assumptions. Keeping those assumptions separate allows users to understand not only what supports a conclusion, but also where the available information stops and interpretation begins.

An assumption does not become evidence simply because it has been repeated, appears reasonable, or contributed to a useful result.

Not all evidence should contribute equally.

Several weak or repetitive sources do not necessarily provide stronger support than one highly relevant source. Ten pages repeating the same announcement, for example, may ultimately depend on a single underlying statement rather than representing ten independent pieces of evidence.

The usefulness of evidence depends on factors such as how relevant it is to the question, how directly it supports the statement being considered, whether it is independent from other evidence, how current it is, what part of the company it applies to, and whether its origin can be inspected.

This means that ASUME does not evaluate evidential strength by counting citations alone.

A recent first-party statement directly addressing a relevant company activity may be more informative than many older or derivative references. Conversely, several independent observations pointing toward the same interpretation may collectively provide stronger support than any one of them could on its own.

The objective is to establish a meaningful evidential basis, not to maximise the number of sources attached to an output.

Evidence should remain connected to where it came from.

This is its provenance.

Depending on the source, relevant provenance may include the originating website or system, the specific page or document, the relevant information within it, when it was published or observed, and the company, product, geography, or context to which it applies.

Preserving provenance allows a user to move from a statement produced by ASUME back toward the information that supports it. It also matters when the underlying information changes. A webpage can be updated, a document can disappear, an announcement can become outdated, or a newer source can provide a different picture of the company.

ASUME therefore treats provenance as part of the understanding rather than adding a source link only after an output has already been generated.

This is one of the foundations of traceability: the ability to inspect the basis of an output instead of receiving only the conclusion.

Evidence is not necessarily relevant to the entire company or valid indefinitely.

Its significance may depend on time, geography, product, business unit, market, legal entity, customer segment, or another part of the company's context.

An announcement from three years ago may accurately describe what the company was doing then while providing little evidence about its priorities today. Evidence about one subsidiary may not support a conclusion about the whole group, and information about one product may not describe the company's wider portfolio.

AUM therefore keeps the relevant scope of evidence in mind when using it to form or revise an understanding.

This prevents a locally valid observation from being applied too broadly and helps ASUME distinguish between something that was true in a particular context and something that appears to describe the company's current state more generally.

Company information does not always agree.

A company may publicly describe one priority while its hiring behaviour points in another direction. New documentation may contradict older material. Customer-provided information may add context that changes how a public statement should be interpreted.

ASUME should not remove contradictory evidence simply to preserve an existing conclusion. Nor should it automatically choose whichever source supports what it already understands about the company.

Sometimes new evidence resolves the disagreement. In other cases, the appropriate understanding may need to remain qualified until there is enough information to distinguish between competing interpretations.

A contradiction is therefore not necessarily a failure of the system. It can be useful information about the limits of what is currently understood.

Keeping disagreement visible is part of maintaining an understanding that can change when better evidence becomes available.

Sometimes the most important thing about the evidence is what it cannot establish.

An evidence gap exists when ASUME does not have enough relevant information to adequately support or evaluate something that matters to the question being considered.

An evidence gap does not mean the statement is false. It means that ASUME does not currently have a sufficient basis to present it as established.

For example, ASUME may find reliable evidence that a company uses a particular technology but have no reliable evidence that the technology is creating an operational constraint. The first statement can be supported while the second remains unresolved.

The appropriate response is not to invent the missing connection. ASUME may instead leave the conclusion qualified, make a necessary assumption visible, seek additional information where appropriate, or preserve the gap as part of the understanding.

This is what Never hide the gaps means in practice. Missing support is information in its own right and should remain visible when it materially affects a conclusion.

Evidence can support a conclusion without making it certain.

Company information is often incomplete, indirect, changing, or compatible with several explanations. A conclusion supported by several independent and directly relevant observations should therefore not be treated in the same way as one that depends on a single indirect source and several unresolved assumptions.

ASUME reflects these differences in how an understanding is represented. Where the evidence is strong, a statement may be supported more firmly. Where the basis is weaker, the appropriate result may remain qualified or uncertain. When different explanations remain plausible, those alternatives do not need to be hidden simply to produce a single confident answer.

The goal is not to eliminate uncertainty from company understanding. It is to make the relationship between the available evidence and the strength of the conclusion more honest and useful.

Evidence continues to matter after an initial company understanding has been created.

New information can strengthen an existing conclusion, weaken it, contradict it, change where it applies, challenge an assumption, or introduce an entirely new view of part of the company.

Suppose ASUME currently understands that a company is building an internal capability based on hiring, documentation, and organisational changes. A later announcement that the same function is being outsourced may materially change what the earlier evidence means.

The new information should not simply sit beside the old conclusion as another disconnected fact. The relevant part of the understanding can instead be reconsidered.

At a simplified level:

Existing understanding + new evidence → updated understanding

This is one of the reasons ASUME maintains evidence rather than using it only to generate a one-time answer.

Evidence must remain distinguishable from ASUME's interpretation of that evidence.

ASUME uses automated evaluation and sampled human verification to evaluate parts of the inference pipeline. Where evidence is reviewed, the purpose can include checking whether an observation corresponds to the underlying information, whether its source and context are preserved, whether it is relevant to the conclusion it supports, and whether unsupported assumptions have been introduced.

Verification does not mean that every piece of evidence or every ASUME output is manually reviewed. Human verification is a sampled quality-control mechanism used alongside automated evaluation.

It also does not mean that verifying the evidence automatically verifies every conclusion derived from it. A source can be real while being interpreted incorrectly. An accurate observation can be irrelevant to a particular conclusion. Relevant evidence can still support several explanations.

Verification therefore helps evaluate the quality of the evidential basis and the way it is used. It does not turn an inferential conclusion into a proven fact.

ASUME deliberately uses the word evidence rather than proof.

Many of the questions ASUME helps answer concern properties of companies that are not directly observable. The available information may support an interpretation strongly without logically establishing that it must be true.

A series of job posts may provide strong evidence that a company is developing a capability, but it does not prove the company's internal strategic intention. A pricing change may provide evidence of repositioning without proving why management made the decision. A company announcement may strongly support one interpretation of its priorities while leaving another explanation possible.

The useful question is therefore not always whether something has been proven. It is whether the available evidence is strong enough for the conclusion being made, what assumptions the conclusion depends on, and what remains unknown.

This distinction is important because ASUME is designed to help users understand companies under incomplete information, not to convert uncertainty into certainty through confident language.

Evidence is one part of the larger process through which ASUME builds and maintains company understanding. It provides the observable basis from which assumptions can be made explicit, inferences can be formed, and conclusions can be evaluated.

The relationships are not always linear. One observation may be relevant to several parts of the company understanding, several pieces of evidence may support one conclusion, and new evidence may challenge something ASUME previously understood. Evidence can also remain useful across different objectives because its relevance is not limited to the output for which it was first considered.

What matters for the user is that evidence does not disappear once an answer has been produced. It remains part of the basis of ASUME's maintained understanding, making it possible to inspect where a conclusion came from, see where support is weak or missing, and revise the understanding when the information changes.

That is the role evidence plays in ASUME: it keeps company understanding connected to what can actually be observed.