How We Verify
Accuracy claims are cheap. This is the process behind ours — including what it does not cover.
1. One Data Layer, No Hand-Typed Figures
Every number on this site — calculator, tables, guides, CSV exports — is generated from a single source-of-truth JSON tree. No figure is typed into a page template. The build fails if a figure written into a page disagrees with the dataset, which is what stops a site quoting two different band 5 salaries on two different pages.
2. Cross-Checks against the Publisher's Own Arithmetic
The strongest test available is comparing our calculation against numbers the publisher worked out independently. The England pay-scale tables publish basic-plus-HCAS totals for every band, step and London zone — 435 figures across five years. Our clamp-and-round logic reproduces every one exactly.
That check is what identified the rounding convention: the supplement is rounded up to the whole pound, not rounded to nearest. Plain rounding matched 399 of the 435 and missed 36 by £1.
3. the Provenance Contract
Every scale carries a status, a named source and a date. A scale is only verified when a formal circular or published rate table exists. Where an award has been announced but not formally issued, it is provisional and labelled everywhere it appears — currently Northern Ireland 2026/27.
4. Golden Test Cases
The engine is tested against cases worked out by hand from the published rules. The expected values are not produced by running the code — a test that records the code's own output can only confirm it still does what it did, never that it does the right thing.
The corpus covers the actual-pay pension tier including exact boundaries, NI charged on gross with an explicit assertion it is not reduced by pension, the personal allowance taper, both HCAS bounds, six-band Scottish tax, split-year NI blending, stacking student loans, pensionability, tax code parsing, and internal consistency of the explanation against the figures it explains. It also computes every published pay point across all five years and four nations and asserts each produces a sane result.
A worked example of why this matters. An earlier version of this calculator contained hand-worked tax tests that were wrong, because the same misunderstanding produced both the engine and the test: the higher-rate band was taken to run to £125,140 of taxable income when it actually runs to £112,570. Both were corrected when the figures were rebuilt from the published band tables. Cross-checks against independent published figures exist precisely because hand-worked tests can inherit the author's error.
5. the Explanation Is Generated, Not Written
The line-by-line explanation is emitted by the same function that produces your figures. Nobody writes those sentences by hand, so the explanation and the number cannot drift apart.
6. the Annual Cycle
- When an award is announced, the new scale is entered and marked provisional if no circular has been issued.
- Once the circular is published, figures are checked against it and promoted to verified with the date recorded.
- Previous years are retained rather than overwritten, so back pay works and history stays available.
- The change is published on the change log.
Figures are updated within 24 hours of a pay award being confirmed by the relevant government or NHS body. Each update is timestamped here so the commitment can be checked rather than taken on trust.
7. When We Get It Wrong
Errors are published, not quietly patched. Every correction goes on the corrections log with what was wrong, who it affected, how it came to light and what stops it recurring.
What This Process Does Not Do
- It does not make the tool a payslip. Real payroll applies your actual tax code and your employer's local arrangements.
- It does not cover the scenarios listed under what is not modelled.
- It has not been independently audited. If that changes, the auditor will be named here.