Research methodology

• We prefer primary research, official documentation, public datasets, and directly observed product data. Sources that can change over time include a publication or access date.
• Secondary sources are used for context or synthesis when they add value. The article still links to the underlying primary source when it is available.
• A source is chosen for relevance and authority, not because its conclusion supports Tetriz. Conflicting credible evidence is represented rather than silently excluded.

• Quantitative claims state the population, time window, inclusion rules, material exclusions, and unit of analysis. Comparisons use like-for-like cohorts and periods where the data permits.
• Each metric is defined before interpretation. Rates keep their denominator visible, and calculated examples disclose the inputs and assumptions that materially affect the result.
• Data-quality limits, missing observations, and plausible confounders are named beside the conclusion they constrain. Correlation, association, and causal evidence are not presented as interchangeable.

A benchmark identifies its source, cohort, and applicable period. Customer examples, internal observations, industry benchmarks, and hypothetical scenarios remain labelled as different kinds of evidence. A modelled calculation is not described as a measured customer result.

• Conclusions stay within the observed population unless broader evidence supports generalisation. Material uncertainty remains visible in the conclusion rather than being reduced to a footnote.
• When supporting data can be shared, the article links the source, calculation, or relevant documentation. Confidential inputs are described at a safe level and narrow the claim when readers cannot inspect them.
• The author checks whether another plausible explanation could produce the result. A substantive reviewer is credited when a technical or methodological review is performed.