Method

Grouping (the same-work screen)

Titles are normalised (case and punctuation folded), grouped on the EXACT normalised title text, level words included. Titles whose only difference is a level word (Senior, II, Trainee) are marked as equal-value questions and never counted as same-work flags.

The basis

Figures read on FTE-adjusted gross annual base pay, the proxy for the pay level the directive names (Art 3(1)(b)); where a row carries FTE, the adjusted pay is base / FTE; rows without an FTE read as 1.0, labelled on the result. Variable pay is read separately where a column exists.

The gaps

Per category and overall: mean gap = (mean male pay - mean female pay) / mean male pay x 100, and the median version on median pay levels; both wordings are Art 3(1)(c) and (e), percentages of the average male pay level. Negative figures read women ahead; they carry the same sign rules.

The trigger

A category reaches the joint pay assessment trigger where the mean gap reads at least 5 percent, inclusive of exactly 5.00; the figure compared is the one displayed (two decimals). All three Art 10(1) conditions are shown per category.

Quartiles

The gendered rows sort by adjusted pay; four groups from the lowest, per quartile as equal as the count allows (a remainder lands in the highest band): the Art 9(1)(f) proportions read from those groups.

The exposure

A scenario, not a ruling: per category, the lift each worker below the category median would need to reach the median, summed per worker and in total. The remedy the employer owes is its own decision under Art 10(1)(c) (remedy within six months of submission) and Art 4(4) (criteria); the figure only prices the size of the spread.

Genders not stated

Rows without a gender figure are carried, counted, marked not stated, and excluded from the gap arithmetic, which compares female and male pay levels; the count is shown next to the workers figure, and nothing is inferred from a name. The gap items stay marked not computable where the paste gives no gender at all.

Method last reviewed 14 October 2026.

Determinism

Same paste, same corrections, same duty inputs: same figures, byte for byte, verified by the engine gate on every deploy. No AI, no randomness, and no figure invented where a column is missing (the item reads not produceable).