Evidence Standard

What the evidence supports — and no more.

aiclavis research separates direct evidence from calculation, interpretation and uncertainty. A conclusion should never sound more certain than the evidence supporting it. Every material finding is therefore classified as:

OBSERVED
DERIVED
INFERRED
UNKNOWN

These labels are not confidence scores. They describe the relationship between a statement and the evidence behind it.

OBSERVED

Directly supported by evidence.

An observation is something the available source explicitly establishes. Examples include:

An observation may still be incomplete or later corrected. Its classification means only that the statement itself is directly supported by the cited evidence.

Standard: A material observation should be traceable to its source. When multiple sources disagree, the disagreement is not silently resolved. It is surfaced.
DERIVED

Calculated from observed evidence.

A derived finding is not directly stated by a source. It is produced through an explicit calculation or transformation of observed data. Examples include:

A derived statement must be reproducible from its stated inputs.

Standard: The source data and transformation should be visible enough that another analyst could reproduce the result. If assumptions are required, they must be identified. A calculation does not become an observation simply because the arithmetic is straightforward.
INFERRED

Supported by evidence, but not directly established by it.

An inference is an explanation, interpretation or structural conclusion that fits the observed and derived evidence. Examples include:

Inference is where much of the analytical value is created. It is also where analytical overreach most easily occurs.

Standard: An inference must be clearly distinguishable from the evidence that supports it. Alternative explanations should be considered when they are materially plausible. A compelling narrative is not evidence.
UNKNOWN

The available evidence does not support a conclusion.

Unknown is not a failure of the research. It is a valid research outcome. A question remains unknown when:

Standard: aiclavis does not fill evidence gaps with narrative. When the evidence is insufficient, the conclusion remains unknown.

The evidence ceiling

A conclusion cannot be stronger than the evidence beneath it.

A material finding inherits the limitations of its weakest necessary assumption. If an argument depends on an inference, the final conclusion cannot be presented as directly observed. If a calculation depends on uncertain inputs, the uncertainty remains part of the result.

The analytical chain should remain visible:

source → observation → derivation → inference → conclusion

At each step, certainty may stay the same or decrease. It should never increase without new evidence.

Hypothesis discipline

The initial hypothesis does not have to survive.

aiclavis research is not designed to confirm an initial concern. A hypothesis may be:

A negative result remains a result. If the available evidence does not support the suspected gap, the correct outcome is to say so. The objective is to resolve the question — not to preserve the original narrative.

Falsification

Every meaningful hypothesis should have a way to fail.

Where possible, aiclavis identifies what evidence would weaken or overturn a working hypothesis. A research conclusion is stronger when we can state not only why it may be true, but also:

What would make us change our mind?

This may include:

If no imaginable evidence could change a conclusion, the conclusion is not being treated as a research hypothesis.

Contradictory evidence

Contradiction is evidence too.

When credible sources conflict, aiclavis does not simply select the version that best supports the working argument. The contradiction is investigated. Possible outcomes include:

If the conflict remains unresolved, the relevant conclusion may remain UNKNOWN.

Source hierarchy

Not all sources carry the same evidentiary weight.

aiclavis generally prefers the most direct and authoritative source available for the claim being made. Depending on the question, this may include:

  1. regulatory filings and official disclosures
  2. audited financial statements
  3. company announcements and investor materials
  4. government and procurement records
  5. court, registry and ownership records
  6. industry and market datasets
  7. reputable specialist reporting
  8. secondary commentary

A secondary source may identify a useful lead. It should not replace a stronger primary source when one is available.

Time matters

Evidence is always evaluated within a time boundary.

Every research object should define a data cutoff. A conclusion describes what the evidence supports up to that point. Later events may:

A later outcome does not retroactively change what was knowable at the original cutoff.

Public evidence and internal evidence

Public research is conducted only from information available outside the organisation unless explicitly stated otherwise. Client work may introduce:

New evidence may change earlier conclusions. When it does, the conclusion should change with it. The purpose of the method is not consistency with our previous view. It is consistency with the strongest available evidence.

Materiality

Not every inconsistency deserves a conclusion.

aiclavis distinguishes between anomalies that are merely interesting and those that could materially affect a decision. A finding becomes strategically relevant when it may change:

Research should reduce decision uncertainty, not simply increase the number of facts.

Revision

Research is versioned.

Material factual corrections or new evidence may require a research object to be updated. Where appropriate, revisions should document:

Silent correction of material findings should be avoided.

Right of reply

Subjects of public research may provide factual corrections or official context.

aiclavis welcomes:

A company response does not automatically override an independently supported conclusion. It becomes additional evidence and is assessed under the same standard. Material corrections are incorporated and documented.

What this standard does not claim

The Evidence Standard does not eliminate uncertainty. It does not guarantee that every inference will later prove correct. It does not turn incomplete public information into internal knowledge.

Its purpose is narrower:

to make the boundary between what is known, calculated, inferred and still unknown explicit.

That boundary is part of the research result.

This research was conducted independently using public information unless otherwise stated. It does not constitute investment research, investment advice or a securities recommendation. Factual corrections and official company context are welcome and, where material, documented in the revision history.

The operating principle

Evidence before narrative.

We do not need the original hypothesis to be true. We do not need every question to resolve cleanly. We do need the conclusion to remain inside the limits of the evidence.

OBSERVED · DERIVED · INFERRED · UNKNOWN