Enforcement outcomes compared: platform ban rules matter less than the grading

Two accounts post the same clip. One gets a warning, one loses the account. The rule text did not change between them, and neither result came out of nowhere. Large services grade enforcement, and the grade is set by facts that sit outside the sentence you broke.

YouTube, Twitch, X, Meta and Reddit write their ban rules in much the same shape. What differs is the grade: account history, surface and reach set it, and the grading decides the outcome.

What to take away

Reading a notice against the grade

A notice usually names a policy area, not a clause. That name is the most useful line in it, because it tells you which document governs and which family of rules you are inside. Work outward from there.

Open the policy it names on the service's own site. Find the section that plausibly covers what happened. Check whether that policy has a separate version for the surface you were on, since advertising, commerce and developer policies commonly do.

The names differ by service. X cites categories such as hateful conduct or abuse and harassment, Meta cites a Community Standards section, YouTube points to Community Guidelines and its strikes page, and Twitch cites its offense categories.

If the notice names a country, treat it as a legal route, not a policy one, and the remedy sits with a regulator or a court rather than an appeal button. The method for establishing which document governs is in working out which rule applies.

A decision has to be explicable enough to answer, which is the idea behind a general account of notice, and review by somebody uninvolved is the idea behind an explanation of appeal. Neither binds a private service. Both explain why the structure exists where it does.

If the notice names nothing, treat it as automated. Write the appeal to the most likely category rather than asking which rule was applied. The question rarely produces an answer and it spends an appeal you may only get once.

Eight variables that set the grade

Read that list again as a description of what belongs in an appeal. History, surface and reach are the facts worth contesting, because those are what the service weighed.

How YouTube, Twitch, X, Meta, Reddit, TikTok and Discord grade enforcement

YouTube strikes. Three strikes in ninety days used to end a channel. YouTube now expires each strike on its own ninety-day clock and drops the channel out of the program at the end, so the grade depends on timing as much as count. A strike on a monetized video also triggers a separate advertiser-friendly review.

A single severe violation can still end a channel without a third strike, which is the policy-category variable at work.

Twitch enforcement. Twitch publishes offense categories in its Community Guidelines, from hateful conduct to illegal acts, and says repeat offenses escalate toward indefinite suspension. It does not publish the expiry schedule for its warnings, so two streamers with the same number of actions can sit at different points on the path.

Appeals go through Twitch's own appeals form, not the notice email.

X (formerly Twitter) and the visibility filter. X restricts reach before it removes content, so the first graded outcome a poster sees is a label or a reply deboost, not a suspension. Its Help Center page on enforcement options lists post deletion, a visibility limit, reply limits, temporary locks and permanent suspension.

That is the surface variable: the same post performs differently in replies, in search and in the main timeline.

Meta's account status page. Meta shows Community Standards violations, their dates and their expiry on one page, and it separates Facebook from Instagram. It is the clearest published example of an accrual system, and it disappears when the account does, so export it while you still have access.

Pages and ad accounts carry a second record in Meta's Account Quality tool, so check both.

Reddit subreddit and site bans. Reddit grades at two levels at once: a moderator can ban you from one community, and the site can suspend the account under its Content Policy. A subreddit ban carries no site-level weight, which is why the same behavior reads as a mistake in one place and a pattern in another.

Reddit writes its site rules as a short numbered list, and each subreddit adds its own on top.

TikTok Community Guidelines. TikTok groups prohibited conduct by topic and lists content removal, warnings and permanent bans among its enforcement options. A video that stays up can still be pulled from the For You feed, which is a reach grade rather than a ban. Appeals run through the notification in the app.

Discord. Discord's Community Guidelines cover servers and direct messages, and enforcement runs from a warning to server deletion to an account ban. Server owners stack their own rules on top, so a ban from one server is not a Discord ban. Discord publishes an appeals page and a transparency report.

How these programs grade is set out in how a moderation policy is built, and the divergence between written and enforced rules is in the gap between the two.

Warnings, thresholds and the invisible clock

Most services run some accrual system. Two things matter. Individual actions usually expire on their own schedule rather than the account's, and escalation is often specific to a policy area. Three actions across four areas can behave very differently from three inside one.

The published clocks differ. YouTube states a ninety-day strike expiry. Meta shows expiry dates on the Account Status page. Twitch does not publish warning expiries at all.

None of that appears in the rule text. It lives on an account status page, if the service has one, and in an accrual policy usually described only in general terms. Look at that page while you can still see it.

Why there are so few clear examples

Rules here are written to cover conduct that has not happened yet, by people who know a published boundary example becomes a specification for getting as close to it as possible. So policies carry obvious examples and almost never carry near misses, which are the cases anybody actually argues about.

The clearest published record is a cross-platform one. Twitch disabled the account of then-President Donald Trump on 6 January 2021. Twitter suspended the same account permanently on 8 January 2021. YouTube issued a strike on 12 January 2021 that froze uploads, its first step rather than a ban.

Meta suspended the Facebook and Instagram accounts on 7 January 2021. Its Oversight Board upheld that on 5 May 2021 and sent the length back for review. Meta then set a two-year suspension on 4 June 2021, counted from 7 January 2021.

One account, four services, four different outcomes, and no rule text needed to change to produce them.

That gap is permanent. Stop expecting it to close. What fills it is your own record: what you posted, what was actioned, what was not, on which date, with what notice text. A dated record is also what a policy news timeline needs to stay useful.

After a year, that file tells you more about where the line sits for your account than any published page will. It is the only evidence that exists about how the rules are applied rather than how they are written.

Recording an example so it is worth having

Six fields, filled in the same day:

Kept consistently, that becomes a set of worked examples for your own situation, which is the only kind that transfers. What an action can be is set out in the eight outcomes.

What none of this establishes

Nothing here states what any service's rules currently say, and no pattern described above is a promise about how you will be treated. Rules change without announcement. Enforcement thresholds move with no document changing at all. A service may act under a legal requirement it does not describe.

Read the current policy on the service's own pages. Where employment or real money depends on the answer, get advice from someone qualified where you are.

Common questions

Someone else does the same thing and nothing happens to them. Is that useful?

Not as an argument. Detection is partial and sampled, so their absence from a queue is not evidence that the rule permits it. It is evidence about sampling, not about the rule.

Does having a large audience help or hurt?

Both, in different places. Reach can raise the grade on an action, and it can also open review routes smaller accounts do not have. Neither is a rule you can rely on.

Do the same rules produce the same outcome on two services?

No. A YouTube strike is graded against a ninety-day clock, a Twitch warning sits on a path the guidelines do not publish, and X grades by which enforcement option it applies. The wording differs less than the grade does.

Can I ask which specific clause was applied?

You can ask. Many notices carry no clause number at all, so the reply mostly restates the policy area. Ask instead for the notice text and the date of the action, which are the two parts an appeal can use.

Is a warning a good sign?

It means you are inside the graduated path rather than outside it. Treat it as the point at which to fix the underlying thing, not as a free action.