Check their sums

The States built a calculator to show us how the tax plan would affect our households. (https://www.gov.gg/taxreform – calculator at https://2026-impact-calc.gov.gg). It does not match their own proposals, and the pattern of the errors suggests that those who built and checked it did not understand the policy they were working up.

The Policy & Resources Committee has asked the States to approve a tax reform package that it says will improve public finances by around £59m a year, with a 3% GST at its centre. To reassure islanders, the Committee produced and published an online calculator. You enter your income and your outgoings, and it tells you whether you would be better or worse off. The Committee has urged people to use it, and has treated its results as evidence, most recently at the Scrutiny hearing (07/07/2026) (https://www.youtube.com/watch?v=RLOfrv_Dv2g&t=869s) held before the debate. Its Vice-President has cast the argument as one in which the claims made against the package do not survive contact with the evidence.

We decided to test that evidence. This is what we did, what we found, and why it matters.

What we did

We examined the calculator as it is actually served to the public. First, in fairness, the good news: the privacy claim holds. The tool runs entirely within your own browser; nothing you enter is transmitted, there are no trackers, and when you close the page your figures are gone. On that narrow question the engineering is sound, and we are glad to confirm it.

We then took the calculator’s own code and built a second copy alongside it, changed in one respect only. Where the live tool departs from the published proposals, our copy follows the proposals. We did not estimate the arithmetic or rely on our own reading of it. We ran the States’ actual code and the corrected code side by side on identical households, so that every difference can be traced to a particular line and a particular figure. To guard against error on our own part, the corrected engine is tested on each run against the original, with every correction switched off, and must reproduce the States’ output exactly before any correction is applied.

We have published the result as a page anyone can use, with the two answers shown together and each difference explained, at thepeoplestrustgsy.com/wp-content/GST/guernsey-gst-calculator-audit.html.

What we found

For a single person on an ordinary wage, the calculator is correct. The income tax and the social security match the policy to the penny. We say so plainly, because it makes the rest harder to dismiss.

For a good many other households it departs from the policy, and it does so in ways that are specific and demonstrable.

Mortgage interest is removed from the GST base twice, and without the cap the policy sets. The base for GST is disposable income, reached by taking gross income and subtracting income tax, social security, housing payments and pension contributions. The full rent or mortgage payment is therefore already removed, correctly, because housing does not bear GST. The code then subtracts the mortgage interest a second time, as a separate deduction from the base, so the same interest is relieved twice. That second deduction is applied in full, whereas the mortgage interest relief given against income tax is correctly capped at £3,500. The understatement of GST is 3% of the interest wrongly deducted, which is £600 on £20,000 of interest. The error flatters the package.
An employed person on a low wage who also has investment income is reclassified as not employed, and charged accordingly. Under the proposals an employee pays the 8.5% contribution on employment income above the £11,122 allowance, and the unearned income of an employed or self-employed person falls outside the new contribution; the policy letter states on page 8 that unearned income “will not be brought into scope for employed and self-employed people”, and its rate table gives the rate as 0%. The code computes the employee contribution first, which is correctly nil where the wage is below the allowance. It then reads that nil result as meaning the person is not employed, and levies the 8.5% non-employed charge on the person’s whole income, investment income included, above the allowance. A person with £10,000 of wages and £30,000 of investment income, who should pay nothing on the investment income, is charged roughly £2,455 that the policy does not impose. The classification turns on whether the computed employee contribution happens to exceed zero, rather than on whether the person is in fact employed. The error overstates the cost of the package.
The additional self-employed contribution band is absent altogether. The proposals add a 2.5% contribution for the self-employed, and for employers, on income between the upper earnings limit of £196,560 and £300,000. The code stops the self-employed calculation at the upper limit and applies nothing above it; the band is simply not present. The configuration for the self-employed under the package does not carry its own upper limit or allowance, and silently borrows the employee figures, which is itself a sign that the section was never built out to the policy. A self-employed person earning £300,000 is undercharged by up to £2,586 a year. The error flatters the package, though only for high earners.
The transfer of unused personal allowance is given to every couple, married or not, and marital status is never even asked. Under the law, unused personal allowance passes between spouses and civil partners; a couple living together unmarried can obtain the same only by a formal election, subject to conditions. The code grants the transfer to any two-adult household, testing only that there are two adults, and not whether they are married or in a civil partnership. The calculator never asks. Its data model carries a marital-status field, initialised and passed into the calculation, but no question on the form sets it and no formula reads it. Because the transferred allowance is worth more against the current 20% system than against the package, in which it largely displaces income that would be taxed at 15%, granting it to a couple who should not receive it depresses the current baseline by more than it reduces the package cost, and so makes the package look worse than it is, by around £490 to £570 for an unmarried couple with one earner. The error overstates the cost of the package.
The vehicle tax runs outside the published schedule. The published annual charge runs from £25 to £280, with a median of £132. The schedule in the code runs from £40 to £300, so a large petrol vehicle is charged £300, which is £20 above the stated maximum.
A single boundary is treated inconsistently between the two halves of the tool. At the £24,960 lower limit for the non-employed and pensioner contribution, the current-system side charges a person whose income sits exactly on the limit, while the package side exempts the same income, because the two calculations apply different tests at the identical figure. The amount is small, but a tool cannot be internally consistent while applying two rules to one boundary.

There are lesser items besides. The hardship payment figures are written into the code with their derivation not shown, and the eligibility cut-off is set at £32,400, which is the income-tax band threshold from the previous version of the package rather than any figure in the current one. The engine carries no protection against invalid entries, and it credits the 1.9% pension and benefit uplift in full, without the GST that would fall on it when spent.

Two assumptions run through all of it. The tool treats every remaining pound as spent, which loads the GST as heavily as it can, unless the user happens to fill in a single savings field. And it looks at one year only: incomes are held at today’s level on both sides, so there is no allowance for the erosion of the fixed reliefs against a GST that will apply to rising prices, and no allowance for the pressure that GST will place on rents over time.

The most telling case is the Committee’s own. P&R have used a single person earning £50,000 as their showcase, and have said such a person would be better off. Put that exact case into the live calculator, on the settings with which it is published, and it returns the opposite: worse off by £39 a year. The tax and the contributions are right; the reversal comes entirely from the assumption that the household spends everything. Enter any meaningful saving and it turns positive again. The Committee’s headline example does not survive the Committee’s own tool.

This would have been checked

None of these is a subtle numerical slip. Each is the policy itself rendered wrongly: a whole contribution band left out; a statutory condition on an allowance dropped; a relief granted twice; a published schedule exceeded. That distinction is the heart of the matter. A calculator of this kind, issued under the States’ name, produced by or for the Civil Service, and put forward by the Committee as evidence, is not a private draft. It would have been reviewed before it was published and before it was relied upon in front of Scrutiny. We are entitled to assume that it was checked.

If the instrument built to explain the policy cannot match the policy, no one is obliged to take the rest of the work on faith.

If it was checked, and these departures remained, then those who checked it did not recognise them as departures from the policy. To see that the self-employed band is missing, or that the marital condition has been dropped, a reviewer must first know that the band and the condition are part of the policy. The errors survived review because the policy was not understood by those doing the work. The only alternative, that a public tool bearing on a £59m decision was published and cited as evidence without being checked against the policy at all, is no more reassuring.

The pattern supports this reading, rather than any suggestion of bad faith. The errors do not run one way. Some overstate the cost of the package and some understate it; the tool flatters the reform for a high-earning self-employed person in the same breath as it penalises an unmarried couple. A tool built to sell the package would lean in a single direction. A tool built by people who had not fully grasped the policy would scatter its errors on both sides, which is what this one does.

How we checked our own sums

An accusation of unchecked work invites the question of who checked ours, so we set out the answer.

First, identity. The audit runs against the calculator’s code exactly as the States served it, captured on 7 July 2026. A cryptographic fingerprint of the captured page, which would change if a single character changed, is recorded on the audit page, and every line of the calculation engine was matched against the page as served before anything was measured.

Second, authority. Each departure is tied to the words of the policy documents rather than to our reading of them. The missing contributions band is in the letter in terms: an additional rate of 2.5% for employers and the self-employed between the Upper Earnings Limit and £300,000. The treatment of an employed person’s investment income is in the letter’s own rate table, which gives it as 0% under this package and 8.5% under the abandoned 2024 package, the very rate the calculator still charges. The vehicle schedule of £25 to £280 with a median of £132 is quoted from the letter. The £32,400 ceiling is recited in paragraph 2.1 as the previous package’s tax threshold. The double deduction of mortgage interest rests on the calculator’s own question labels, which ask for the full annual rent or mortgage payment and then for the interest expressly “for relief calculation”.

Third, arithmetic. Every figure was derived twice by different routes. Working only from the rates and thresholds in the letter’s Appendix 7, without reference to any code, we recomputed the tax, contributions and GST for a single employee at £25,000, £50,000, £100,000 and £200,000; the live calculator matches that independent derivation to the penny at all four points, which is what entitles us to say its single-person arithmetic is sound. Every amount quoted in this article, from the £2,455 overcharge on a low earner with savings to the £39 by which the Committee’s £50,000 example fails, was re-derived the same way.

Fourth, the check that never stops. The published audit page re-runs the comparison every time it is opened. It executes the States’ engine and the corrected engine with every correction switched off, across more than a thousand randomly generated households, and requires the two to agree exactly before any correction is shown; the result is displayed at the top of the page. If they ever diverge, the page says so on its face.

This is offered as method, not as proof of infallibility. The process caught and corrected two of our own descriptions after initial publication, which is precisely what checking is for. Everything here can be repeated by anyone, because the code is published inside the page itself. That is the standard we ask the States to meet.

A note on the document itself.

The policy letter, as offered to the public on gov.gg, is not fully searchable. On 68 of its 172 pages, some or all of the paragraphs are pictures of text rather than text, rendered line by line as images, and on those pages are the linked footnotes, which matters because the footnotes are where the sources are cited and where figures such as the £1.6m already spent on implementation are disclosed. A search of the public copy for these things returns nothing. The copy lodged with the parliament.gg papers is a different build of the same letter, and it is markedly more searchable than the one given to the public: the case-study tables and several footnotes that are images in the public copy are text in the deputies’ copy. Yet in both, the passage recording that unearned income will not be brought into scope for employed and self-employed people survives only as a picture, invisible to any search. The pattern, in which the same document is rasterised differently by different publishing runs, is characteristic of exporting from machines whose fonts are not licensed for embedding; whatever the cause, the effect is that the public and their deputies were given copies of one letter that answer the same searches differently, and neither answers them completely. The remedy would cost 10 minutes: re-export the letter from its source files on a machine with correctly licensed fonts. Until then, we have prepared a fully searchable copy, produced by optical character recognition, which can be found at https://www.thepeoplestrustgsy.com/wp-content/GST/2026-07-10 TPT-proposition-P2026-49-MS_ocr.pdf.

Why it matters, and what must follow

The States are to decide, this summer, on a permanent change to the way the island is taxed, worth some £59m a year. The public and their deputies have been directed to a calculator in order to understand what that change would mean for them. If the officials and the politicians cannot produce a calculator that agrees with their own proposals, on a matter of this importance and after this long, the question is no longer a technical one. It is a question of competence, and therefore of trust. If the instrument built to explain the policy cannot match the policy, no one is obliged to take the rest of the work on faith.

That question presses hardest on the parts we cannot see. Behind the calculator sits the States’ own household model, from which the claim that a majority will be better off is drawn. On the Committee’s own answers to Deputy Curgenven, that model rests on data from 2019 and 2020. The distributional figures in the policy letter appear as bare results, with no method attached. The model and the data behind it have not been released, and requests to examine them have been refused. We were able to check the calculator only because its code, of necessity, runs on your own device, where it can be read. The model that produced the headline claims enjoys no such exposure, and there is no good reason for it to remain closed.

We ask two things of the States. Correct the calculator, so that it matches the policy it describes, or withdraw it. And publish the model and the base data, in a form that can be checked, so that the figures underpinning this decision can be tested in the way the public calculator now has been. We have done our part in the open. On a decision of this size, the island is entitled to expect the same.

ByJames Collings