Research and Science

The purpose is to make research claims traceable from the bounded object of inquiry to the observations, measurements, data, models, inferences, and publication choices supporting them. The document does not decide every disciplinary dispute; it shows what a particular finding rests upon and where its limits lie.

The scope includes empirical, historical, qualitative, quantitative, theoretical, experimental, and computational research insofar as the relevant chains can be documented. Requirements must be adapted to the field: a laboratory experiment, a corpus study, historical source criticism, and a mathematical demonstration do not have the same grounds and must not be reduced to one methodological form.

Reality Audit does not presume that research is captured, false, or dishonest. It begins with mandate, material, and distinctions. It may end in F1 — supported correspondence — just as readily as in uncertainty, insufficient grounds, error, capture risk, or capture identified.

A research claim is rarely produced in one step. It passes through choices concerning question, scope, instrument, registration, cleaning, coding, analysis, model, visualisation, interpretation, and formulation. Each stage may add information, but also selection, loss, assumptions, and uncertainty.

The audit must therefore distinguish what is investigated from the traces and representations used. It must also distinguish a rule-governed inference from the claim about what that inference shows concerning actuality.

Before method and result are assessed, the question the study actually tests must be clear. An investigation of one dataset, selected group, or period cannot without further grounds become a claim about all persons, all conditions, or a permanent mechanism.

The mandate must distinguish the original research question from the Reality Audit question. Reality Audit may, for example, test traceability, domain of validity, or publication language without reopening the entire study.

Data are not simply found; they are registered through instruments, documents, categories, selection, and procedures. This does not make data arbitrary, but it makes the registration chain relevant to what the data can support.

Measurement error, missing data, sampling bias, coding, rounding, filtering, and cleaning must not be collapsed into one vague category. They operate differently and require different documentation and correction.

Models and statistical methods may reveal patterns, compare explanations, and produce precise predictions. They do not acquire ontological content merely by being mathematically consistent or fitting a dataset.

An inference must be reconstructable: which assumptions are used, which alternatives are tested, which parameters are identifiable, and how does the argument move from result to conclusion?

Validation is not one yes-or-no stamp. It may concern instrument, measurement, construct, code, model, prediction, external validity, or practical effect. What is validated, against what, and under which conditions must be stated.

In this document, reproduction means obtaining a result again from the same material and analysis, while replication means renewed testing with independent material, sample, or performance. Fields use these terms differently; the local definition must therefore be visible.

Peer review, editorial control, disciplinary consensus, and institutional status may be important quality and trust signals. They remain social and professional control arrangements, not direct identity between authority and truth.

A published article may later be corrected, retracted, or bounded. An unpublished or dissenting result may be weak, but status alone does not decide its grounds. Reality Audit examines both the material and the control system around it.

Figures, tables, maps, images, diagrams, and summaries are not neutral windows. They select scale, crop, order, category, colour, aggregation, and language. These choices may be professionally necessary, but they must answer to the material and claim they carry.

Public communication may require simplification. Simplification becomes misleading when decisive uncertainty, scope, alternatives, or the distinction between observation and interpretation are removed so that the recipient receives a stronger assignment than the research supports.

Research and science are an application domain, not one audit type. A case may contain a correct model but misleading visualisation, sound measurement but overextended language, or a weak institutional correction system without the result itself being false.

Each audit-type finding therefore requires its own material. A finding in one track must not automatically transfer to another.

Findings must remain bounded to what the audit examined. An error in a figure is not automatically an error finding for the entire theory; a failed replication is not automatically proof of misconduct; and a robust result is not automatically a complete explanation.

Complex research cases should be divided into partial findings: data grounds, measurement, model, inference, representation, authority system, and practice may receive different F categories and confidence levels.

Correction must answer to the finding. Minor documentation errors do not automatically require retraction, while a decisive data or code error cannot be corrected merely through new wording. The measure must reach the point in the research chain where the deviation arose.

Open science may strengthen traceability through sharing protocols, code, materials, preregistration, versions, and negative results. Openness is not unlimited: privacy, source protection, Indigenous and ownership interests, safety, copyright, and legitimate confidentiality must be considered.

This document is not an anti-science argument and does not justify treating all expertise, methods, or research findings as equivalent. Corrigibility is a strength of research where institutions and practices genuinely permit correction.

Reality Audit must not be used to demand impossible certainty before knowledge can inform action. Decisions often must be made under uncertainty. The requirement is that uncertainty, grounds, alternatives, responsibility, and opportunity for renewed testing remain visible and proportionate to consequence.

Further development should include field-specific appendices for measurement science, statistical inference, qualitative methods, historical source criticism, computational science, research ethics, and open science. Such appendices must be developed with relevant domain competence.

• Develop field-specific requirements without making one discipline the measure of every other discipline.
• Build public, anonymised, or synthetic cases demonstrating the difference among data, model, language, and system problems.
• Develop a research audit template containing protocol, data flow, analysis plan, findings matrix, and correction log.
• Test the methodology on both disputed and ordinary research findings — including cases ending in F1.