Why Prism is like this

What was broken

A thumbnail and a title cannot tell you whether the useful ten minutes exists inside a two-hour talk. So you gamble the afternoon to find out, and the algorithm that recommended it was optimising for a completely different outcome than yours.

Then there is the other half: you watched the right video three weeks ago and retained a vague feeling that it was good. The knowledge did not survive the week.

Everything built to help treats a symptom. Playlists organise videos you have not evaluated. Speed controls make a bad video shorter. Note apps sit in another window with no idea what is on the screen. None of them know what the video actually says.

Who it’s for

• People using YouTube to learn something specific: a framework, a craft, an exam, a trade
• People who watch long conference talks and want the relevant stretch
• People who take notes while watching and lose the link between the note and the moment

Who it isn’t for

• Entertainment viewing. Ranking a comedy channel against your goals is a category error
• Anyone after a downloader or an ad blocker
• People who want YouTube's recommendations left exactly as they are

Rules we don’t break

Decided once, so they never have to be argued about again.

• Relevance, not engagement. The signal is the transcript against your stated goal. View count, recency and channel size never enter the score.
• Always show why. A score with no explanation is just another algorithm asking to be trusted. Every match shows what matched, so you can disagree with it.
• The source stays the source. Prism grades and navigates YouTube. It does not re-host it. You are always one click from the original video at the original timestamp.
• A project, not a playlist. Work is organised around what you are trying to learn, so scoring, notes and Focus Play all share one context.
• Remove the pull, don't add another feed. Shorts, comments and the sidebar can go while you are working. Prism does not put its own recommendations in the space it cleared.

What it actually does

• Relevance scoring from transcripts. Tell it the goal; it reads each transcript, scores how useful the video actually is, and shows the reasoning next to the number.
• Section-level grading. Videos are split and colour-graded, so the sponsor read and the intro are visibly separate from the walkthrough.
• Focus Play. Stitches the relevant sections across a whole project into one continuous run. A two-hour talk becomes the fifteen minutes you needed.
• Projects with timestamped notes. Notes attach to the moment they belong to, so review is a scroll rather than a re-watch.
• A quieter YouTube. Hide Shorts, comments and the sidebar on YouTube itself while you are in a project.
• Extension plus web app. The extension cleans and saves the page you are on; the web app holds projects, scores, notes and Focus Play.

What it refuses to do

The most useful part of any note like this. Every line has a reason attached.

• Hosting or downloading video. Somebody else's fight with somebody else's lawyers, and it improves the product not at all.
• Scoring entertainment. The model assumes you have a goal. Pointed at a comedy channel it produces confident nonsense, which is worse than no score.
• Becoming a course platform. The content is YouTube's. Prism is the lens, not the curriculum.
• A social layer. No sharing of scores or projects. Comparing other people's relevance scores would be interesting and it is not the problem being solved.

How we’ll know it worked

• People finish fewer videos and learn more. That shows up as sections watched rather than whole videos played.
• Focus Play gets used on long material, because that is where the saving is measured in hours.
• Projects get revisited through notes. That is the difference between watching and learning.
• Users disagree with a score and say so. Silence would mean nobody is reading the reasoning.

Where it could go wrong

• No transcript, no score. Auto-captions vary wildly and some videos have none. Where the transcript is thin the ranking is a guess, and it has to behave like one.
• The goal sentence carries the whole system. A vague goal produces a vague ranking, and the user experiences that as the product being wrong rather than the input being thin.
• Reading transcripts costs money. Analysis has a per-video cost, which puts a real floor under any free tier. That has to be designed for, not discovered.
• It lives on somebody else's page. A YouTube layout change can break the extension overnight, and extension review sits between the fix and the user.

What’s next

• Better handling of poor or missing transcripts, including saying so plainly instead of scoring anyway.
• Sharper section boundaries on long, loosely structured talks, which is where grading is hardest.
• No dated roadmap.