A source trail you can follow.

Establish what the inquiry must answer.

Agree on the question, intended use, deliverables, and deadline. Identify which claims need checking, which sources are relevant, and what falls outside the work.

This establishes the boundaries against which the final answer can be judged.

Examine how the evidence was produced.

Prioritize original research and primary documentation where available. Examine methods, definitions, source dates, and relevant versions. Record access limitations that could affect the findings.

Several reports repeating one source do not provide several independent confirmations. Source relationships matter alongside the number of citations.

Test the connection between claim and source.

Trace important claims to the material supporting them. Compare accounts and look for qualifications, contradictions, or changes in meaning between the original evidence and its later use.

A source can be credible without supporting the particular conclusion attached to it. The synthesis must make any additional reasoning visible.

Make the answer usable and reviewable.

Provide the agreed synthesis, references, and supporting material. Distinguish direct findings from interpretation, identify consequential gaps, and explain where the evidence does not resolve the question.

The handoff should let you follow a conclusion back to its basis and identify questions that need further work.

State what remains uncertain.

A scoped inquiry cannot guarantee that every relevant source has been found. Coverage limits, unavailable material, conflicting findings, and unresolved questions are part of the result.

Confidence should reflect the evidence available. A firm conclusion requires stronger support than a plausible interpretation.

Five checks behind a conclusion.

A genuine source and an accurate quotation can still lead to an unsupported conclusion. Our working standard examines the source, the reported fact, the method, the inference, and its application.

Source verification

Confirm the author, title, publication status, and appropriate version. Record the source location and access date.

Factual verification

Locate the reported finding. Check exact values, denominators, rounding, and qualifications against the cited passage.

Methodological verification

Record the sample, task design, metric, tools, comparison group, and exclusions that produced the finding.

Inference verification

Ask whether the evidence warrants the conclusion. Separate direct findings from interpretation and consider competing explanations.

Transfer verification

Check whether the conclusion applies to the intended use. Record relevant differences between the studied setting and the client’s work.

Does the evidence fit the intended work?

Before using an external evaluation to support an operational decision, we examine six points of correspondence.

Task

Does the evaluated work resemble the actual task distribution, including routine and difficult cases?

Success

Does the metric capture the required quality and consequential failures?

System

Are the model, prompts, tools, permissions, and review process materially comparable?

Workflow

Does the evaluation include context gathering, iteration, coordination, and integration?

Reliability

Are repetition, variation, deferral, recovery, and serious failures assessed at the required level?

Time

Does the evidence apply to the current system, with version changes and contamination risks accounted for?

A mismatch identifies evidence still needed. This is an analytical check, not a certification or a universal passing score.

Make the strength of the claim visible.

Directly establishes

The study design directly measures the proposition being stated.

Evidence reasonably supports

An inference is needed, but it is proportionate to the design and accounts for alternatives.

Complicates

The evidence changes the scope, confidence, mechanism, or applicability of the conclusion.

Contradicts

The evidence is materially inconsistent with the proposition under relevant conditions.

Not established

Available evidence does not justify the proposition. This does not prove it false.

Unresolved

The evidence is insufficient, conflicting, inaccessible, or too sensitive to version changes for a stable judgment.

Try to disprove the emerging conclusion.

Review asks what the strongest counterevidence is, whether different studies are being treated as one measure, and whether a bounded search has been mistaken for proof that evidence does not exist.

Important claims should have a traceable record of their evidence, methods, qualifications, and intended use. Material discrepancies must be resolved or disclosed before handoff.

Based on Diagnostic Research Verification Standard v0.1, September 2026. This provisional working standard is revised as practice reveals its limits.