Make the comparison
inspectable.
A database recommendation should show its work. This is the editorial standard Database Lens intends to apply to future comparisons.
01 / Define a workload, not a winner
Every comparison starts with a specific use case, data model, transaction boundary, latency objective, availability requirement, deployment model, and operating budget. We describe what would disqualify a candidate before assigning any preference.
02 / Identify the kind of evidence
| Evidence | What the article must show |
|---|---|
| Documented capability | A link to the primary source, the relevant edition, and the version or date. |
| Hands-on observation | The environment and steps used, including what was and was not tested. |
| Measured benchmark | Hardware, dataset, configuration, test scripts, repetition, and the reported distribution of results. |
| Editorial judgment | The assumptions behind the interpretation and where it may not apply. |
03 / Compare the whole operating picture
Our evaluation framework covers correctness, query behavior, failure and recovery, scaling, day-to-day operations, migration effort, and cost. The relative importance of each dimension depends on the workload. We do not present an unexplained universal score.
04 / Make measured results reproducible
Before labeling an article a benchmark, we require a documented test protocol and accessible supporting materials. We distinguish vendor-reported results from our own measurements and avoid converting a single workload into a general product ranking.
05 / Explain the limits
Every comparison should identify the scope of its evidence, relevant assumptions, and what remains untested. A documented feature is not a measured performance result. A conclusion should stay within the workloads and conditions the evidence supports.
06 / Keep a correction trail
Substantive revisions should update the article date and explain what changed. Product claims should be checked against the edition and version being discussed. If evidence is incomplete, the conclusion should say so.