Find anything on this site69 of 69 records

69 records indexed: every page, every jurisdiction with its reconciliation result, the guides, the rules cited here and every correction.

Provenance · four of four

Corrections log

Every figure we have got wrong, with the date, the page it was on, what it said, what it says now, and how it surfaced. Publishing your own mistakes is the only version of a trust signal that costs anything.

Both entries below were found by machine rather than by a reader complaining, and both are stated with the money attached. We do not write that a figure “has been updated for accuracy”. We write what it was, what it should have been, and how much the difference was worth.

corrections published

rule at fault, in every one

£628.50

largest single overstatement

reported by a reader

quietly edited

ProvenanceSource of record 995 Open data 1 file Rate changelog 3 datesCorrections 2 loggedMethodology

The log

Newest first. Each entry keeps the wrong figure in full: the point of a corrections log is that the reader can see what they might have been told, not a sanitised note that something changed. The rule id links the error back to the source of record, so you can read the document we misread.

2 entries

0106 September 2026/

Rule at faultgb.income_tax.bands.ty2026

What it said

The homepage calculator started the additional rate at £137,710 of gross income, understating income tax by £243 at a £130,000 gross (£44,460 against £44,703). The error ran from £125,140 to £137,710 of gross and was invisible below that.

What it says now

The additional rate starts at £125,140. The homepage calculator lifts each taxable-income boundary by the personal allowance so it can run against gross, but it baked in the standard £12,570 at build time instead of using the tapered allowance it already computes at runtime. By £125,140 the allowance is zero, so there was nothing to add. The boundary is now emitted as an expression in the runtime allowance, which is correct at every income.

How it surfaced

Searching the published homepage for the corrected £125,140 boundary after fixing the rule. It was absent, and £137,710 was there instead.

0206 September 2026/countries/united-kingdom/

Rule at faultgb.income_tax.bands.ty2026

What it said

The additional rate was applied from £112,570 of taxable income, overstating income tax on a £150,000 gross by £628.50 (£54,331.50 against £53,703). Incomes with taxable income below £112,570, including the £60,000 case the site had a test for, were unaffected.

What it says now

The additional rate applies from £125,140 of taxable income. The ingest had subtracted the £12,570 personal allowance from every published band boundary. That is correct for the basic rate limit but wrong at the top, because the allowance has already tapered to zero by £125,140: GOV.UK states "You do not get a Personal Allowance on taxable income over £125,140." The bands are now read from the GOV.UK table stated on income after allowances, which needs no arithmetic.

How it surfaced

Cross-checking our own figures against the OECD Tax Database over SDMX, which publishes the UK higher-rate threshold as £125,140 where our rules held £112,570.

The pattern in both entries

One misreading, two places, found two different ways

Both corrections come from the same rule and the same mistake: subtracting the personal allowance from every published band boundary. That is right at the basic rate limit and wrong at the top, because by £125,140 the allowance has already tapered to nothing, so there was nothing left to subtract.

The entries exist separately because the error reached the reader twice by two routes, once through the rule and once through the build step that lifts boundaries onto gross, and each route had to be fixed on its own. Collapsing them into one entry would have hidden the fact that fixing the rule did not fix the homepage.

Neither was reported. The first was caught by the OECD cross-check, the second by checking the built page for the corrected boundary and not finding it. Both checks now run on every build.

What counts as a correction

Any published figure that was wrong: a misread band, a threshold applied to the wrong year, a rate that moved and we did not notice, an allowance subtracted where it had already tapered away. If a reader could have made a decision on the number and the number was wrong, it belongs here.

Typos in prose are not corrections and do not appear here, because burying real errors in a list of them is its own kind of dishonesty. A log padded with comma fixes is a log nobody reads, which is a convenient outcome for whoever is padding it.

A figure that has never been published cannot be corrected. Where the store holds no rule for a jurisdiction, the calculator says so on the page instead of estimating, so there is no wrong number to log in the first place. That gap shows up in the methodology as a refusal, not here.

Telling us

Point at the page and the line. The fastest thing you can send is the official clause you think we misread: the arithmetic is open on the source of record and downloadable from the open data, so a disagreement can usually be settled without either of us guessing at the other’s method.

You do not need to be right, and you do not need to be polite about it. A reader who says “this looks wrong at £150,000” and nothing else is still more useful than a check we did not run, because it tells us where to point the next one.

Where to send it

There is no address on this page yet.

Inviting corrections and printing a mailbox that bounces would be worse than the gap, so the gap is stated instead. The dashed border on this box means the same thing it means everywhere else on the site: a statement about our own immaturity rather than a claim we want to make.

When it exists, anything sent to it that turns out to be a correction appears on this page with the date you told us and how it surfaced, and the wrong figure stays printed in full. We do not quietly edit a number and move on.

Source of record Rate changelog

On reading a short log

For most of this site’s life this page said “nothing logged yet”, and that was a statement about how long the site had been up rather than about how careful we are. 2 entries is not much better. The ruleset is young, most of it has been read once rather than twice, and only 7 of its 995 rules have reached verified status. Treat a short log as a reason to check us, not a reason to trust us: the sites with no corrections page at all are not the ones making no mistakes.