Skip to content

Cognisia

You read a paragraph, then read it again. Cognisia notices, and offers one small change you can keep or undo.

Try the reader

How to Live on 24 Hours a DayArnold Bennett, 1908Scripted
The Daily MiracleFrom chapter I

Philosophers have explained space. They have not explained time. It is the inexplicable raw material of everything. With it, all is possible; without it, nothing. The supply of time is truly a daily miracle, an affair genuinely astonishing when one examines it. You wake up in the morning, and lo! your purse is magically filled with twenty-four hours of the unmanufactured tissue of the universe of your life! It is yours. It is the most precious of possessions. A highly singular commodity, showered upon you in a manner as singular as the commodity itself!

For remark! No one can take it from you. It is unstealable. And no one receives either more or less than you receive.

Talk about an ideal democracy! In the realm of time there is no aristocracy of wealth, and no aristocracy of intellect. Genius is never rewarded by even an extra hour a day. And there is no punishment. Waste your infinitely precious commodity as much as you will, and the supply will never be withheld from you. No mysterious power will say:—“This man is a fool, if not a knave. He does not deserve time; he shall be cut off at the meter.” It is more certain than consols, and payment of income is not affected by Sundays. Moreover, you cannot draw on the future. Impossible to get into debt! You can only waste the passing moment. You cannot waste to-morrow; it is kept for you. You cannot waste the next hour; it is kept for you.

I said the affair was a miracle. Is it not?

You have to live on this twenty-four hours of daily time. Out of it you have to spin health, pleasure, money, content, respect, and the evolution of your immortal soul. Its right use, its most effective use, is a matter of the highest urgency and of the most thrilling actuality. All depends on that. Your happiness—the elusive prize that you are all clutching for, my friends!—depends on that. Strange that the newspapers, so enterprising and up-to-date as they are, are not full of “How to live on a given income of time,” instead of “How to live on a given income of money”! Money is far commoner than time. When one reflects, one perceives that money is just about the commonest thing there is. It encumbers the earth in gross heaps.

If one can’t contrive to live on a certain income of money, one earns a little more—or steals it, or advertises for it. One doesn’t necessarily muddle one’s life because one can’t quite manage on a thousand pounds a year; one braces the muscles and makes it guineas, and balances the budget. But if one cannot arrange that an income of twenty-four hours a day shall exactly cover all proper items of expenditure, one does muddle one’s life definitely. The supply of time, though gloriously regular, is cruelly restricted.

Which of us lives on twenty-four hours a day? And when I say “lives,” I do not mean exists, nor “muddles through.” Which of us is free from that uneasy feeling that the “great spending departments” of his daily life are not managed as they ought to be? Which of us is quite sure that his fine suit is not surmounted by a shameful hat, or that in attending to the crockery he has forgotten the quality of the food? Which of us is not saying to himself—which of us has not been saying to himself all his life: “I shall alter that when I have a little more time”?

We never shall have any more time. We have, and we have always had, all the time there is. It is the realisation of this profound and neglected truth (which, by the way, I have not discovered) that has led me to the minute practical examination of daily time-expenditure.

The Desire to Exceed One’s Programme

“But,” someone may remark, with the English disregard of everything except the point, “what is he driving at with his twenty-four hours a day? I have no difficulty in living on twenty-four hours a day. I do all that I want to do, and still find time to go in for newspaper competitions. Surely it is a simple affair, knowing that one has only twenty-four hours a day, to content one’s self with twenty-four hours a day!”

To you, my dear sir, I present my excuses and apologies. You are precisely the man that I have been wishing to meet for about forty years. Will you kindly send me your name and address, and state your charge for telling me how you do it? Instead of me talking to you, you ought to be talking to me. Please come forward. That you exist, I am convinced, and that I have not yet encountered you is my loss. Meanwhile, until you appear, I will continue to chat with my companions in distress—that innumerable band of souls who are haunted, more or less painfully, by the feeling that the years slip by, and slip by, and slip by, and that they have not yet been able to get their lives into proper working order.

If we analyse that feeling, we shall perceive it to be, primarily, one of uneasiness, of expectation, of looking forward, of aspiration. It is a source of constant discomfort, for it behaves like a skeleton at the feast of all our enjoyments. We go to the theatre and laugh; but between the acts it raises a skinny finger at us. We rush violently for the last train, and while we are cooling a long age on the platform waiting for the last train, it promenades its bones up and down by our side and inquires: “O man, what hast thou done with thy youth? What art thou doing with thine age?” You may urge that this feeling of continuous looking forward, of aspiration, is part of life itself, and inseparable from life itself. True!

But there are degrees. A man may desire to go to Mecca. His conscience tells him that he ought to go to Mecca. He fares forth, either by the aid of Cook’s, or unassisted; he may probably never reach Mecca; he may drown before he gets to Port Said; he may perish ingloriously on the coast of the Red Sea; his desire may remain eternally frustrate. Unfulfilled aspiration may always trouble him. But he will not be tormented in the same way as the man who, desiring to reach Mecca, and harried by the desire to reach Mecca, never leaves Brixton.

It is something to have left Brixton. Most of us have not left Brixton. We have not even taken a cab to Ludgate Circus and inquired from Cook’s the price of a conducted tour. And our excuse to ourselves is that there are only twenty-four hours in the day.

If we further analyse our vague, uneasy aspiration, we shall, I think, see that it springs from a fixed idea that we ought to do something in addition to those things which we are loyally and morally obliged to do. We are obliged, by various codes written and unwritten, to maintain ourselves and our families (if any) in health and comfort, to pay our debts, to save, to increase our prosperity by increasing our efficiency. A task sufficiently difficult! A task which very few of us achieve! A task often beyond our skill! Yet, if we succeed in it, as we sometimes do, we are not satisfied; the skeleton is still with us.

A scripted reader, run through Cognisia’s real rules at 15 times speed.

Solo project. I came up with the idea and directed the design. AI wrote the code: v0 for the first version, Cursor’s agent for v2, and Claude Code for the reader on this page.

  1. First demoMar 2026Done
  2. v2 rulesOct 2026Done
  3. Opt-in readerOct 2026In progress
  4. Reader testNextNext
4 min read
The problem
Reading tools make you pick your settings. People’s favourite font is their fastest only as often as chance.
The bet
Let the page notice. Go back to a paragraph twice, and it offers one change.
Where it stands
The rules run live on this page. Not yet tested with readers.

You read it twice. It didn't stick.

Reading tools ask you to pick your own settings. That assumes you know what helps you.

I wanted to help people read faster and remember more. But what helps depends on the reader: font, line spacing, letter spacing. So my first idea was a set of fonts to choose from.

  • Expected their favourite font to work best
    73%
  • Were fastest in their favourite font
    20%
352 adults read in five fonts. Their favourite was their fastest only 20% of the time, no better than chance.Wallace et al., ACM TOCHI 2022

So I flipped it. The page notices instead. It watches one thing you'd never think to mention, going back, and compares you with your own pace, not an average.

Set it up yourself, or let the page notice?

Chosen The page notices and offers one change at a time.

  • Not chosen

    A set of fonts to pick from (my first idea)

    You'd have to know what helps.

  • Not chosen

    A settings panel

    The same guess, with more controls.

Cost I accepted

It has to wait. Nothing is offered until four paragraphs set your pace.

My first version only pretended to watch

I built it with v0 in two weeks and put it live. In October I opened the code.

The site did say you could switch states by hand.
What the site saidWhat the code did
“Watches how your brain reads”Nothing watched. States changed when you clicked.
“7 line skips”, “86% confidence”Typed in. The same on every visit.
A line strip as “a stable horizontal anchor”It drifted on a timer.
Gaze tracking, “with permission”A real eye tracker, connected to nothing.
“Install for Chrome”The button played a chime. No extension.

Its site, and its code as downloaded from v0

I wanted a basic first version, so I let the AI choose the scope. It built a demo of watching instead of watching. So v2 began by deleting 91 of its 100 files, with a new rule: every number on screen is measured, or labelled “Example”.

It notices your scrolling, not your eyes

I wanted an eye tracker at first. But a camera on you is a lot to ask, so I moved to scrolling, which the page already sees.

Scrolling is a rough signal too. Readers scroll back more on harder text. But in a study of 518 readers, scrolling back didn't predict how much they understood. So Cognisia only offers. It never decides for you.

How should the page notice?

Chosen Scrolling and timing only. No camera.

  • Not chosen

    An eye tracker, using the webcam

    A privacy cost, and too coarse to see which line you reread.

Cost I accepted

Going back looks the same whether you're lost or looking something up. In scripted checks, 4 of 8 ordinary scroll paths raise an offer.

The thresholds are starting values.

Your pace
The middle of your last ten paragraphs.
Going back
Twice to one paragraph in two minutes, or three times anywhere in one.
The offer
One change, at the next paragraph, so your line stays put.
After
Read on and it stays. Undo it and it isn't offered again.

Bold starts have to earn their place

I assumed bolding the start of words would help everyone. For most adults, it doesn't.

Word starts

Philosophers have explained space. They have not explained time. It is the inexplicable raw material of everything. With it, all is possible; without it, nothing. The supply of time is truly a daily miracle, an affair genuinely astonishing when one examines it. You wake up in the morning, and lo! your purse is magically filled with twenty-four hours of the unmanufactured tissue of the universe of your life! It is yours. It is the most precious of possessions. A highly singular commodity, showered upon you in a manner as singular as the commodity itself!

Long words get a bold start. Short common words stay plain until Strong, because readers skip most of them.

Word starts bold the first part of each word, like Bionic Reading. They helped me and some friends, so I built v2 around them. Others said they did nothing.

Four studies of adults found no gain, and two found a slowdown. So after six paragraphs with word starts and six without, Cognisia turns them down if you read slower.

Keep a feature most readers don't gain from?

Chosen Keep it, and let each reader's pace decide.

  • Not chosen

    Drop it, as v2's first brief did

    An average can hide the readers it helps.

  • Not chosen

    Turn it on for everyone

    Two studies found it slower.

Cost I accepted

A reader's pace swings between passages by about as much as the 15% slowdown it looks for, so it can misjudge.

You stay in charge

Try “Read it yourself”, then switch on “Learn from my reading”.

Until you do, nothing is measured. Each offer says what happened to the text, never what's wrong with you. It's a reading tool, not a diagnosis.

On a phone: the offer, and Why as a sheet. Its numbers are examples.Figma, drawn with Claude from the built reader, October 2026
On a phone: the offer, and Why as a sheet. Its numbers are examples.Figma, drawn with Claude from the built reader, October 2026
On a phone, the offer sits over text you've already read. Its numbers are examples.Figma, drawn with Claude from the built reader, October 2026
On a phone, the offer sits over text you've already read. Its numbers are examples.Figma, drawn with Claude from the built reader, October 2026

When should it start paying attention?

Chosen Only after you switch it on.

  • Not chosen

    Named states, applied silently (first version)

    “Mild Struggle” named a condition, and moved the line you were on.

  • Not chosen

    On from the start (v2)

    It changes your reading before you say yes.

Cost I accepted

Fewer people will switch it on.

Where it stands

Checked
When the type changes, your line moves under half a pixel in Chromium.
Tried
15 people close to me tried an earlier version. 13 liked it, 2 found it useless. Which version isn't recorded.
Can't see
Rereading inside one paragraph.
Not tested
Whether any change helps someone read.
Next
A Chrome extension that adapts the pages people already read. Still an idea.

What it changed in me

The first version looked finished, and it wasn't. Now I don't call anything done until it does what it says, the idea is truly mine, and nothing on the screen looks machine-made.

Read more: what v2 retired, the app in Figma, why spacing comes first, the five studies, and how still the line staysLess

What v2 retired from that look: gradient text, glows and glass, pill labels with a dot, uppercase monospace labels, a fake browser window, “live” pulsing dots, sound effects, a splash screen and the “Built with v0” badge. They read as generated templates.

The reader as an app, in Figma. The same states drawn as a full-page app, in Light and Dark from one set of colour variables. The top bar and the toolbar's status line are designed, not built.

The offer, on the reading line, at 1440 px. Its pace is an example.Figma, drawn with Claude from the built reader, October 2026
The offer, on the reading line, at 1440 px. Its pace is an example.Figma, drawn with Claude from the built reader, October 2026

Why the first offer is wider spacing. Much wider spacing halved reading errors for dyslexic children in one study, though another found no gain in speed, so for adults it's a guess.

The five studies of word starts.

A selection of five published studies of bold word starts. None found a gain for typical adult readers.
StudyWhoWhat it found
Snell, 202432 studentsNo clear effect. Inconclusive on its own.
Spear and others, 202560 studentsSlightly slower: 282 against 291 words a minute
Parker and others, 202690 skilled adult readersEvidence of no difference in reading time
Možina and others, 2025114 adultsSlower, on screen and on paper
Vascelli and others, 202626 young readers, 20 with learning disordersMore letters read correctly. A small pilot.

Published studies, 2024 to 2026

How still the line stays. The type changes with a short crossfade, then the text scrolls back by however far it moved. In the reader on this page, the paragraph on the reading line moved 0.03 px at 1440 px wide and 0.41 px on a 390 px phone, in Chromium. It varies with the layout's width. v2's own tests measured 0.38 px in Chromium and 0.75 px in WebKit on a phone.

  • Allowed
    4 px
  • This page, phone, Chromium
    0.41 px
  • v2, phone, Chromium
    0.38 px
  • v2, phone, WebKit
    0.75 px
How far the line you're reading moved when the type changed.This page’s reader, and v2’s browser tests

Later I checked whether a laptop camera could do the job. The best webcam trackers are off by about 1.4 degrees, and telling one letter from the next takes about 0.3. A webcam sees roughly where you are, not which line you went back to.

The reader’s session

13570:001:002:003:00Went backBackAgainAgainOffer: wider spacingOffer
Which paragraph is on the reading line, minute by minute.

How often it's wrong. Real reading wasn't available, so scripted scrolling went through the real rules on this page's layout. The first run found a bug. A quick scroll back counted every paragraph it passed, so a look-up could name the wrong paragraph. In 3 of 8 ordinary paths, the offer landed and was kept within 0.12 seconds, unseen. Now a paragraph counts only once your reading line rests in it for 0.3 seconds. That fixed all three without missing any struggle the rules caught before.

Scripts written to stand for ordinary reading, not people. They show what the rules do with each path, not how often people do it.
Scripted scroll pathOffer?
Read straight throughNo
Glance back once at the paragraph aboveNo
Look up two names, 40 seconds apartNo (yes before the fix, naming the wrong paragraph)
Check the same definition twice in 2 minutesYes
Hunt for a quote: up, up, down, upYes
Two small nudges back at a paragraph's startYes
Flick past by mistake, scroll back onceNo
Glance at the line above at three paragraph starts in a minuteYes
Struggle: back to one paragraph twiceYes
Struggle: back two paragraphs, twiceYes
Struggle: three quick looks at the paragraph aboveYes
Struggle: reread one long paragraph from its startNo. The rules can't see it

This page’s false-alarm test, on its 1440 px layout, October 2026

The decision log. v2's log holds 74 entries. 11 are marked as my calls, including word starts as the core. Most of the rest are the agent's starting values, like the thresholds above, still waiting for testing.

Sources

Each claim, quoted from the page, with how sure it is.

Show every sourceLess

Research findings

My own account

  • The first version's goal (read faster, remember more), the set of fonts, the eye tracker set aside for privacy, and the Chrome extension as the direction. My own account of how the idea developed, 2026-10-12.

What people told me

  • “15 people close to me tried an earlier version. 13 liked it, 2 found it useless.” Friends and family, asked informally, before the reader on this page existed. Which version each person tried isn't recorded. Not a test.
  • Word starts “helped me and some friends,” and “other people told me they did nothing for them.” From my own reading and conversations. Not recorded.

My estimates

  • “The spread in readers' pace between two passages was about 18%.” From Readwise's 2022 test with 1,916 online readers, a blog post, not peer-reviewed: the spread (standard deviation) of each reader's difference between two passages was 58.2 words a minute, about 18% of the average speed of 327.9. The typical difference is smaller, roughly 14% if the spread is normal. Either way it is close to the 15% the check acts on.

Measured from the build

  • The rules in the reader above are v2's, file for file. A test compares them with v2's code and fails if one line differs. The reader's scripted mode only scrolls; the rules see that scrolling as they would a person's.
  • The table of what the first version claimed and did. The claims are quoted from its site. The behaviour is read from its code as downloaded from v0: states set by a click, numbers typed in, a line strip moved by a timer, an eye tracker not connected to the reader, and an install button that only plays a sound.
  • “91 of the first version's 100 files” deleted. From v2's history.
  • “4 of 8 ordinary scroll paths raise an offer.” Scripted scroll paths through the real rules on this page's layout at 1440 px, in a test that fails if the number changes. Scripts, not people.
  • The line moving 0.03 px and 0.41 px. Measured in this page's reader in Chromium, at 1440 and 390 px wide, after the offer and after Undo, on 12 October 2026. It changes with the layout's width (0.28 px at 360). Not yet measured in Safari.
  • The line moving 0.38 px and 0.75 px. v2's browser tests on a 375 px screen, against a limit of 4 px.
  • v2 on my laptop starting on, with word starts on. From v2's code.