The thing it broke
was the brake.

the one ask

Find the brake.

Not a petition. Not a boycott. A test you run on your own feed, right now.

1. 01

   Open the feed you lose hours to. Go into its settings.

2. 02

   Find the brake. Turn off autoplay, force it chronological, turn off personalized recommendations, whatever it offers.

3. 03

   Set it. Close the app. Reopen it. Did your choice hold, or did the feed quietly come back?

The controls often exist. What they usually don't do is stick: they reset, they miss the Reels and the Shorts and the notifications, they nag you back. A brake that doesn't hold is decorative, present in form, absent in effect. You didn't fail the test; the brake did.

i looked for mine. it didn't hold. i kept scrolling. so did you, just now, to get here.

That is why this is an ask and not an accusation. You are not taking anyone's word that the feed is rigged; you felt the dead button in your own hand. There is no company to point at and no generation to blame, because every feed fails the same test.

the recognition, in a song

Where's the Brake.

Not a rally cry. The confession set to music: you go looking for the brake, find it's a ghost, and keep scrolling anyway. The machine has no hand on the wheel because it has yours, and the little one behind you learned not to stop by watching you not stop. It points at no company and no generation. It points at the hand holding the phone.

The song is the feeling; the test is the action. Found your brake yet? file what happened. Or open it on Suno.

wherever you're standing

The ask is the same. What it means depends on where you stand.

Below this line is the receipts: why it is the brake, who is already in court over it, and the standard itself. Read as deep as you need.

the diagnosis

A dead man's switch is supposed to stop the machine when your hand comes off.

We built the opposite: the machine that keeps running when the hand comes off. You say "show me less," "I want to stop," "I never agreed to this," and nothing changes. The controls are decorative. Your will is present in form (there's a button) and absent in effect (it does nothing).

we all screamed brake, it heard the gas.

And here is the part that changes where you aim: the harm doesn't live in the content. It lives in the delivery loop: infinite scroll, autoplay, the slot-machine reward schedule, an algorithm optimizing for time-on-site. A brilliant post and a banal one ride the same loop. The loop is what addicts. So the thing to regulate was never the speech. It was the brake.

the receipts · july 2026

This isn't a metaphor. It's in court.

The legal theory is product design, not bad content: platforms "intentionally incorporated algorithms, notifications, endless scrolling and other features designed to keep young users repeatedly returning." The defense is telling, "social media addiction is not an established psychiatric condition," which is why the case gets stuck. But the claims that advanced most cleanly weren't about addiction at all. They were about deception and consent. That's the tell.

the full record: every case, every number, dated and sourced →

why nothing on the table fixes it

Punish. Prohibit. Immunize.

None of it sets a standard for what a non-harmful product must do.

• Damages: backward-looking. Price the harm, don't stop it. A company can pay and keep the mechanism.

• Age bans: blunt, evadable, and they trade the problem for an age-verification fight.

• Immunity: the opposite of a solution. Meta is lobbying for exactly this. It removes the pressure entirely.

• “Content moderation”: government deciding which viewpoints kids see. Unconstitutional as a mandate, and it removes the very material critical thinking is practiced on.

the reframe

Addiction or not, ask if your "stop" sticks.

"Is social-media addiction a real diagnosis" is a swamp of expert-witness fights. Here is the question that is measurable and already inside the law: when a platform tells you a control does something, does it do it, and does it last? That is control integrity, and it is the whole fix. It regulates the platform's own control, not anyone's speech, which is the surface that survives the First Amendment and Section 230. This isn't a claim that addiction isn't real: the brake works whether or not anyone wins that fight, which is why it doesn't have to displace the people already litigating it.

There is a bigger, riskier idea nearby: that the feed does not just addict, it manufactures division, and that you could turn that down by changing what it optimizes for. That one is a research hypothesis, kept in the notes on purpose. The brake has to work whether or not that one holds.

the objection worth answering

"So just have the platforms detect the wedges and turn them down."

They can. Trivially. And that's the trap, not the fix. These are the best-positioned entities on earth to spot a wedge: the most compute, the most scale, and the deepest behavioral profile of every user, the same modeling that makes the feed compulsive in the first place. Detection was never the hard part.

The danger is who you'd be handing that power to. A silent, platform-owned classifier of what counts as "divisive" for every child hands the most capable and most conflicted party an unaccountable authority over speech. The instant a platform can build it cheaply, the only thing between detectable and a silent speech-authority over every kid is the platform's goodwill.

So the reason to avoid the detector was never that it can't be built. It's that a detector this cheap to build and this profitable to cheat on must never be the silent, centralized mechanism. That's why the fix is the objective, not the detector: change what the feed optimizes for, the wedges fall out with no classifier to own, and "did you change the objective?" is something an outsider can actually audit. A silent classifier's honesty never is.

the fix

A brake that works. On by default for kids.

Two layers, from most defensible outward. Neither one classifies a post or touches anyone's speech.

The through-line: regulate the loop, re-attach consent, de-amplify don't censor. The tempting next step, a parental label that names the move a post is making, is more powerful and more dangerous, so it stays in the notes as research, not policy.

the part most proposals hide

The honest limits.

A clean-sounding "solution" to a censorship-adjacent problem is exactly what gets grabbed and misused. So here's where this breaks:

• It's a frame, not a solution. The measurable thresholds, the testing, the enforcement, surviving the First Amendment and Section 230: that's 95% of the work, and it isn't here.

• “Who defines a wedge” doesn't dissolve. It moves. The parental toggle fixes who applies the label, not who forged the classifier behind it. That's why the core fix, control integrity, needs no classifier at all, and why the label stays research.

• The auto-classifier is the anti-pattern. “AI silently auto-hides wedges for kids” is the exact thing this proposal exists to prevent, not endorse.

• De-amplify, not remove. Hold that line. The moment it slides to hide or delete, it reacquires the censorship problem and stops building critical thinking. The line is load-bearing.

This is the section that makes it an honest diagnosis instead of a grudge. The target is a mechanism, not a who. It reads identically for Facebook, TikTok, or YouTube.