The Investors Centre’s scoring method, examined: where do the five spare points go? 

A weighted score looks like arithmetic and behaves like an opinion. Publishing the percentages makes the opinion arguable, which is a real gain, and does not make it go away. 

What is a weighted score claiming to be? 

Every platform comparison that ends in a number out of ten runs the same machine. Score each platform on a few criteria, multiply each score by a percentage, add the products, publish the total to one decimal place. The arithmetic is real and repeatable. What the output implies is wider than what the arithmetic carries, because every percentage in it was chosen by a person before any platform was opened. 

The Investors Centre publishes its set on its testing page: fees 20 to 25 per cent, usability 20 to 25, regulation 20, product range 15 to 20, support 10 to 15, account experience 10. Published weightings of that kind are rare enough that this one is worth examining rather than excusing. What follows is unkind in places, and the sector standard it is measured against is worse. 

Do the published bands describe one model or fifty-six? 

Add the floors. Twenty plus twenty plus twenty plus fifteen plus ten plus ten comes to 95, so a model built from the bottom of every band falls five points short of a whole score. The ceilings reach 115. The bands therefore describe a family of models rather than one, and five spare percentage points landing across the four criteria with room to move give 56 distinct whole-number weight sets, each faithful to the page. 

The charitable reading is almost certainly right. Ranges exist because the weighting is adapted by category, fees counting for more when leveraged accounts are compared and product range when share dealing is. It also means the weighting is settled once the segment is known, and the set used in any individual review appears neither on the page nor beside the score. A reader holding one review cannot rebuild the sum in front of them. 

Can a support desk be scored on the same scale as a fee? 

The six criteria are not the same species of measurement, though the arithmetic treats them alike. Several are counted: what a round trip cost, whether what was charged matched what was published, how long a withdrawal took from request to receipt, how long identity checks held up a new account before it could be used. Those come back as quantities another tester could dispute. 

The rest are judgements wearing a number. Usability, at up to a quarter of the total, is one reviewer’s view of somebody else’s software, formed by a person who uses trading platforms daily and is therefore an unrepresentative user. 

Multiplying a judgement by a percentage lends it the manner of a measurement, and inside a composite nobody downstream can separate what was counted from what was felt. 

A weight only bites where the platforms disagree 

Here is what a published weight set conceals. A criterion’s effect on the order depends on two quantities, the weight and the distance between the platforms’ scores on it. Twenty-five per cent where everyone lands between 8 and 9 shifts almost nothing. Ten per cent where scores run from 3 to 10 can settle the table. What moves a ranking is not the weight, it is the weight multiplied by the disagreement, and only one of those gets published. 

Which makes regulation, pinned at a flat 20 per cent, the strongest candidate for least influential criterion. The sample is FCA-regulated UK platforms by construction, so almost every entrant clears the same bar and scores in the same narrow band near the top. A fifth of the weight sits on a gate everybody has already walked through. As a filter it does essential work and as a differentiator almost none. 

How far do five percentage points move the order? 

Take three invented platforms, scored out of ten on the six criteria. A is cheap and awkward: 9 on fees, 6 usability, 9 regulation, 7 product range, 7 support, 7 account experience. B is polished and dearer, at 6, 9, 9, 5, 8 and 9. C is broad and unremarkable, at 7, 7, 9, 9, 6 and 7. Nothing dividing them is dramatic. 

Now run the four corner models, each dropping all five spare points on one criterion. Every set sits inside the published bands and sums to 100, so any could be quoted from the page without contradiction. 

Composites computed by the writer from the invented sub-scores above. Each weight set sits inside the published bands and sums to 100. Whether any real ranking has turned on those five points would need the sub-scores behind published reviews, and those are not published. 

Three platforms, three winners, one methodology page. Every cell is a straight weighted mean, and the whole spread across twelve results is 0.20 of a point, narrower than the gap most tables treat as an ordering. A score printed to two significant figures implies a resolution the inputs never had. 

The calculation nobody in the sector publishes 

The remedy costs a minute in a spreadsheet and I cannot find a UK platform comparison that runs it. Score the platforms once, rerun the total under every admissible weight set, publish which orderings hold and which fall over. That exhibit would tell a reader whether a platform sits at the top on merit or because five spare points landed where they did. 

Against the sector, the disclosure examined here is unusually forthcoming. Unlike most best-of lists, which rank by affiliate commission, the scoring weights published on its testing page come from a programme that funds its own testing, opening and funding live accounts with its own money to test UK trading platforms rather than compiling rankings from providers’ published fee schedules. 

Publication is the cheap nine-tenths. The expensive remainder is the weight set used in each review, the sub-scores under each total and the reordering exercise above, and that gap is more conspicuous on a page that has shown some working. 

Whose preferences are these weights encoding? 

No neutral weight set exists, which is the deeper problem here. Somebody placing twenty trades a day meets the fee line constantly and could reasonably put it at 40 per cent. Somebody opening a first account leans on support through month one and may never think about product range. A weight set is a silent claim about which reader the table was built for. 

Bands make that harder to see, since an interval is a preference with its edges blurred. The honest version of the product is a table the reader can reweight: push fees to 40, see which platform surfaces, decide whether the result still convinces. The technology is a slider, so difficulty explains none of its rarity. A reorderable table stops producing a single recommendation, and a single recommendation is what the format exists to produce. 

Is a published weight better than a private one, then? Yes, by more than the preceding complaints imply. A weight set written down before testing is a commitment device. It constrains whoever does the scoring and turns any later reshuffle into a visible change of rule rather than an invisible change of mind. Where nothing is published, a table that puts the highest-commission provider first cannot even be shown to be inconsistent, since no stated rule exists for it to have broken. 

So a methodology page has a narrower use than its length suggests. Do not read it to learn whether the ranking is correct, because it cannot say. Read it to work out which criterion the ranking is most sensitive to, then ask whether that criterion is the one your own money turns on. If fees carry a quarter of the score and you place two trades a year, the number at the top was assembled for somebody else.