Chrome extension – accessibility
Plainsight
Ten reading tools on every page you visit. Free, no account, no paid tier – and it tells you which of its own features the research actually supports.
Version 0.3.5 is live on the Chrome Web Store as of 15 September 2026. Free, no account. The full privacy policy is in the Privacy section below. Questions: dhseadev@gmail.com
The ten tools, and what the evidence says
Every verdict below is shown inside the product, next to the switch that turns the tool on. No other reading extension surveyed for this build tells you which of its features are supported by research and which are popular but unproven.
Type and spacing
Letter, word, line and paragraph spacing at the WCAG 2.2 minimums, text size, line length, page colour.
Line focus
Five ruler styles – bar, lightbox, shade, underline, and a mask that snaps to real rendered lines.
Read aloud
On-device voices, 0.5x to 4x, word and sentence highlighting that mutates nothing in the page.
Focus reader
Full-screen distraction-free reading, in the page, with every other tool applying inside it.
Word tools
Syllable splitting and parts-of-speech colouring, with an underline cue so it is never colour-only.
Vision filters
Colour-vision correction for protan, deutan and tritan. Not a simulator – it re-maps confusable colours.
Tint and comfort
Any tint with a live contrast guard, brightness reduction, blue attenuation, stop page animation.
Look-up
Alt-click a word for syllables, part of speech, a prefix and root breakdown, and a definition. Offline.
Colour gradient
Alternating colour across each line, measured against real wrap points. Widely sold, and we are not aware of independent evidence it helps.
Fixation bolding
Bolded word-starts. Snell 2024 found no reading-time benefit; a 1,916-reader replication found it slower. Shipped anyway, labelled.
It declines rather than guessing
Two tools could have shipped confident and wrong. Instead both refuse the cases they cannot do safely, and both measure themselves inside the product against hold-out sets asserted to share no words with the tables they ship with, so the measurement is not circular.
| Tool | Coverage | Accuracy | What it does when unsure |
|---|---|---|---|
| Syllable splitting | 100 percent of words | 97 percent on the hold-out set | Leaves the word whole |
| Parts of speech | 77 percent of words coloured | 96 percent of those correct | Leaves the word uncoloured |
Two limits, stated
Text inside open shadow roots is not reached – sites built from web components are left alone rather than partly decorated. On YouTube that is 1,767 custom elements and about two thirds of visible text reachable. Nothing breaks; some text is simply untouched.
Read aloud plans a whole page up front. A 100,000-character article becomes 786 chunks in memory before the first word is spoken. Fine at that size, but it is not a sliding window.
Questions
Does it work on any page?
It works on ordinary http and https pages. Chrome does not allow any extension to run on browser pages like the new-tab page, the settings pages or the Web Store, and Plainsight says so on those pages instead of doing nothing quietly.
Does it need an account?
No. There is no account, no sign-in, no paid tier and no server of ours.
Why does it say one of its own features does not work?
Because it does not, and you deserve to know that before you turn it on. Fixation bolding is popular, widely sold, and the replication evidence is against it. It ships because comfort and preference are real reasons to use something.
How accurate is the syllable splitting?
97 percent on a hold-out set of words deliberately excluded from the table it ships with. Words the safe rules cannot split are left whole rather than guessed at.
Privacy
What Plainsight does and does not send anywhere
Plainsight itself makes no network requests. There is no fetch, no XMLHttpRequest, no WebSocket, no analytics, no remote script and no remote font anywhere in the shipped code. A build gate scans every shipped file for all of those, checks the manifest, and fails the build if the content security policy does not lock connect-src to none. That gate is itself proven to fail on seeded defects before any pass from it is reported.
Two Chrome features Plainsight uses can put data on the network, and both are off or local by default. Some of Chrome’s voices synthesise on a server, which means the text being read is sent there – Plainsight uses on-device voices only unless you change that, ranks local voices above network ones even when you allow both, labels which is which, and refuses to speak rather than quietly using a network voice. Chrome’s settings sync is off unless you turn it on; your per-site rules never sync.
What is stored: your settings, and a list of hostnames you gave your own rules to. Hostnames only – no addresses, no queries, no page content. Page text is held in memory while the reader or read-aloud is running and released when you close them. Nothing is ever transmitted.
Permissions. At install: storage, activeTab and scripting. No access to any site until you grant it. Running on every site is one button in Settings, and one button beside it takes it back. Deliberately not all-urls, which would also cover local files.
Contact: dhseadev@gmail.com
What is ours, and what is not
The extension is original work. The research it cites is not ours and is named in the product so you can check it: Zorzi and colleagues on spacing, Niklaus and colleagues on reading rulers, Wood and colleagues on text-to-speech, Griffiths and colleagues on coloured overlays, and Snell on fixation bolding. Where a finding goes against a feature we ship, the product says so.
A DHSeaDev project. Built and verified in the open – 69 unit assertions, 53 mutants killed, 57 real-browser assertions and a four-shard stress battery.
More from DHSeaDev: Haven, a privacy-first wellbeing journalling extension built the same local-only, no-account way.
Why it costs nothing: Why accessibility tools should not have a paywall.