# SCC Nexus Research Standards

## Purpose

SCC Nexus aims to produce outputs that are inspectable, reproducible and appropriately cautious. The standard is inspired by established academic and scientific norms: transparent methods, primary-source preference, explicit uncertainty, reproducibility, provenance and independent review where feasible.

## Evidence hierarchy

Source strength is assessed contextually, but the default preference order is:

1. Primary datasets, official registers and original research records.
2. Peer-reviewed primary research and systematic reviews.
3. Authoritative institutional reports, standards and technical guidance.
4. Reputable secondary synthesis.
5. News, commentary and discovery-only material.

Lower-tier sources may identify leads but must not silently substitute for stronger evidence when stronger evidence is available.

## Claim discipline

Every substantive claim should be classified as one of:

- **Observed** — directly represented in a cited source or dataset.
- **Derived** — calculated reproducibly from cited inputs.
- **Inferred** — interpretation supported by evidence but not directly observed.
- **Hypothesis** — plausible proposition requiring further testing.
- **Unverified** — discovered but not sufficiently corroborated.

Public presentation must not blur these categories.

## Reproducibility requirements

A research output is reproducible only when a competent third party can identify:

- source identifiers and retrieval dates;
- inclusion and exclusion criteria;
- transformations and calculations;
- software and dependency versions where material;
- validation checks;
- known missingness and bias;
- output version and build identifier.

## Statistical and scientific integrity

- Report denominators, not percentages alone.
- Distinguish exploratory from confirmatory analysis.
- Avoid causal language from observational associations unless justified by design and analysis.
- Preserve missingness and uncertainty.
- Document multiple-testing or model-selection risks where relevant.
- Do not treat statistical significance as practical or clinical importance.
- Pre-specify thresholds where feasible and disclose post-hoc choices.

## AI and automation

AI may assist discovery, extraction, classification, summarisation or drafting, but it cannot itself establish scientific truth. Material AI-assisted steps should be documented. High-impact claims require source-grounded verification independent of model output.

## Corrections

Material errors should be corrected transparently with a dated changelog. Historical versions should remain identifiable where practicable.

## Independence

Funding, commercial interests, sponsorship, personal interests or advocacy positions that could reasonably affect interpretation should be disclosed alongside relevant outputs.