What we will not claim

This is the article the rest of the magazine gets measured against, so it is written in the flattest register we can manage. The accessibility and self-improvement markets have spent decades selling confidence that outran the evidence supporting it. We are in that market. What follows is the list of claims we have decided not to make, the list we are prepared to defend, and the test we would like applied to us.

Start with the pattern rather than the products, because the pattern is the useful thing. A product is designed to address a genuine difficulty. Some of the people who try it prefer it and say so, sincerely. That preference is then described in marketing as an improvement, the improvement acquires clinical vocabulary, and within a few years the thing is being recommended, purchased at scale by schools and services, and defended by people who have staked something on it — all several steps ahead of any controlled result. Nobody in that chain has to be acting in bad faith for it to happen. Mostly nobody is.

Coloured overlays and tinted lenses are the best-documented example. They are widely used, many readers report that pages feel calmer with them, and the strong claim — that tinting the page addresses dyslexia — has been examined repeatedly over several decades without that examination supporting it. Dyslexia-specific typefaces followed a similar arc: designed thoughtfully, popular with a lot of readers, and, when compared against a well-set ordinary typeface, generally not producing the reading advantage the design intended. Vision-training exercises sold as a remedy for reading difficulty have been through the same cycle, as have brain-training games marketed as general cognitive improvement, where the recurring finding is that people get better at the game.

Now the part that usually gets left out of this argument. Preference is real. If a reader tells you a page is more comfortable with a coloured overlay, they are reporting something true about their own experience, and dismissing it is both rude and unscientific. Comfort has downstream effects: a person who finds reading less unpleasant may read more, and reading more is not nothing. The failure here is not that people like these things. It is the category error of selling a preference as a treatment, and of allowing a sentence about comfort to be read as a sentence about ability.

The failure is not that people like these things. It is selling a preference as a treatment.

The category error, named

Our own exposure

It would be comfortable to leave that as an observation about other companies. It is not one. We sell a reading app. There is a dyslexia panel on our own home page, and it is there because we want dyslexic readers to try the product. We have a paid tier. Every incentive that produced the pattern described above is pointed at us, from the same angle, with the same force, and writing an article about the incentive does not dissolve it. If anything, an article like this one creates a fresh temptation: to bank the credibility of having been honest here and spend it somewhere less careful.

The specific mechanism worth naming is inference. We rarely need to state a false claim. It is sufficient to place a true statement about crowding next to a true statement about dyslexia and let a reader draw the line between them. That construction is legal, it is common, and it is dishonest, and it is available to us on every page we write. We have tried to head it off by stating the negative explicitly wherever the positive appears, which is why the phrase “this is not a treatment” turns up on the home page, on the science page and in this magazine more often than a copywriter would like.

The claims we refuse to make

This list is not aspirational and it is not a style guide. These are sentences that do not appear in our marketing, our documentation or our support replies, and if you find one, it is a defect and we would like to be told.

• That it treats, remediates or manages dyslexia. It does not. It is not a treatment, a therapy, an intervention, a diagnosis, a rehabilitation programme or a form of remediation. Assessment and structured literacy instruction are the things that change outcomes, and this is not either of them.
• That it improves reading ability. It changes how text is presented. Those are different sentences and the distance between them is the whole subject of this article. We have no evidence that anything about a reader’s underlying reading ability is different after using it.
• That it works for everyone. It does not, and the variation between individuals is larger than any average effect anyone is likely to find. Some readers find single-word presentation immediately easier. Others find it worse than a page, and there is no way to tell which from here.
• That our comprehension numbers come from anywhere but our own test. The recall score is produced by the in-app check, on the reader’s own material, on the reader’s own machine. It is a self-administered measurement, useful for comparing your Tuesday with your Thursday, and it is not a research instrument.
• That any figure comes from a trial we have not run. We have not run a controlled trial. The combination we are actually selling — word-by-word presentation plus automatically resolved sense imagery at speed — has not been tested by us or, as far as we know, by anyone else. It is our bet, not a finding.
• That faster is better. Comprehension declines as rate rises and past a point it collapses; what remains is skimming, which is a genuinely useful skill and is not reading. A reader who settles at 180 words per minute because the fixed point suits them has used this product correctly.

The claims we will defend

The counterpart matters as much. A refusal list on its own is a rhetorical posture; it is only worth anything if the remaining claims are specific enough to be checked and specific enough to be wrong. Each of these can be verified by a reader in a single sitting without asking us anything.

1. The word arrives at a fixed point, which removes eye movement during reading. This is mechanical rather than theoretical: there is no line, no return sweep and no place in a paragraph to keep.
2. There are no flanking letters. One word occupies the field with empty space on every side, which is the limiting case of increased letter and word spacing, and crowding is a well-established perceptual effect.
3. The image is chosen by sense, scored against an eight-word window either side, and every image is labelled with the tier that produced it — including the procedural tier, which carries no semantic content at all and is marked as such so you can catch us.
4. Nothing leaves the device. No account, no backend, no telemetry, 0 bytes uploaded. Documents are parsed on your machine and the app works offline. You do not have to trust us about this; you can watch the network panel.
5. The app contains a comprehension check that can and will tell you your recall dropped, and it ships switched on. It mixes real targets with plausible distractors, and words read counts the furthest point reached so rewinding cannot inflate it.
6. All of it is free. The complete app is the free tier, so the cost of finding out whether any of this suits you is ten minutes and a page you already meant to read.

The test we want applied to us

There is one more concession owed, and it is the largest. The strongest argument against this entire class of tool is that words in a stream vanish, and going back is often how comprehension gets repaired. Delete the backwards glance without deleting the failure that prompted it and reading feels smoother while being worse — which is the exact combination people use to judge whether they understood something. We mitigate that with step-back keys, a restart, dwell that stretches up to 2.8× at paragraph breaks, and a test at the end. We do not solve it, and a competitor pointing at it is pointing at something real.

So here is the standard, and we would rather it were applied evenly than applied to us gently. Ask any reading product three questions. What does it measure — not what does it claim, what number does it actually produce about your reading? Does it ship that measurement switched on by default, or is it an optional feature buried behind a settings panel where nobody will find it? And what result would make it admit it was wrong: is there any outcome the product could produce that its makers would accept as evidence against their own claim?

A product that cannot answer the third question is not making a claim at all. It is making an advertisement that has borrowed the grammar of a claim. Our answers, for the record: a recall score beside a rate, on your own material; yes, on by default, with an exportable session report that is deliberately unflattering; and a reader whose score is lower with the second channel than without it has produced evidence against us, which is why the science page tells you exactly how to run that comparison. If it goes against us, it goes against us. We would still rather you had the result than the subscription.