Methodology

Datasets

Every analysis names its dataset, its source, its period and its size. Where a series is a reconstruction or an estimate rather than a direct measurement, that is stated. Cleaning steps are reported, including what was removed and why.

Methods

The method is chosen for the question, not for novelty. Where several methods would be defensible, the alternatives are named. Where a simpler method would have produced the same conclusion, that is said.

Claims and benchmarks

Numbers are either measured here — with the procedure described — or sourced with a citation. Vendor benchmarks are labelled as vendor benchmarks. No statistic on this site is invented for illustration.

Software evaluation

For every tool review: whether the software was actually run, on what dataset, which features were exercised, what version, and what could not be tested. Untested claims are attributed to the vendor.

Uncertainty and assumptions

Point estimates are reported with intervals wherever the method produces them. Model assumptions are listed, and the places where they are most likely to fail are named. An analysis that reports only its best model is treated as incomplete.

Corrections

Errors are corrected in place with a dated note. Substantive changes to a conclusion are flagged at the top of the page.