A Speedrun Record Is Not Official When the Run Ends. Moderators Verify It Frame by Frame First. Here Is How.

Watch what happens when you finish the run of your life. The timer stops, you clip the moment, you post the time — and none of it's official yet. You've made a claim, not a record. The record happens later, at a desk you'll never see, when a volunteer moderator for that game opens your video and checks three things in order: whether you entered the right race, whether your clock is the clock the leaderboard trusts, and whether the footage proves both. Here's the whole chain, in the order it hits you.

So when does a record actually become official?

A record becomes official when a moderator accepts it onto the leaderboard — not when you finish, and not when you submit. On Speedrun.com, every game carries its own board, its own rules, and its own named moderation team. Super Mario 64's board, to take the worked example running through this piece, lists named super moderators, moderators and a bench of verifiers, and it's a big job: the site's own counts for that one game stand at 54,762 runs from 8,575 players (site counts, read off the game's Speedrun.com page). Every one of those runs passed through a human check like the one you're about to read. Your time sits in a queue until somebody with the verifier tag works through it.

That's the part newcomers miss, and it's worth sitting with for a second: there's no central speedrunning authority handing down one rulebook. There's a site, thousands of game communities on it, and each community's own written rules. The chain below is how those communities run the check.

Which race did you actually enter?

Before anyone times anything, your run has to belong to a category — and the category is the race. The three labels you'll meet everywhere:

• Any% means you reach the game's finish by any route the category allows. You're not collecting everything; you're getting to the end, fast.
• 100% means you meet that community's written definition of complete — every item, collectible or objective the rules name, then the finish.
• Glitchless means you run under that community's ban list for glitches and major skips. What's banned is written down per game, because "glitch" is a line each community draws for itself.

Watch how concrete that gets in practice. Super Mario 64 doesn't even use those three labels as its main boards — it splits by star count, and its submission guide asks 16 Star runners to declare whether they used the "LBLJ" route or the "No LBLJ" route (community submission guide, Ukikipedia's RTA Guide). Same game, different races, different leaderboards. Submit a brilliant run to the wrong category and you haven't set a record — you've filed your time in a race you didn't run.

Which clock is the real clock?

Here's where the frame-by-frame part stops being a figure of speech. Communities time runs two main ways, and they don't mix:

• In-game time (IGT) trusts the timer inside the game. Celeste's rules on Speedrun.com state it in one line — "Timing is done using the in-game timer" (official category rules, quoted in the game's forum) — and everything else follows from that choice.
• Real-time attack (RTA) trusts a clock outside the game, running from a defined first moment to a defined last one. Mario 64's board times full-game runs from power-on to the frame the big star is collected at the end, and its retiming standard pins the first frame of the Super Mario 64 logo at 1.33 seconds (official leaderboard rules). A moderator retiming your run counts frames against fixed points like that one — which is why the top of a board can turn on a single frame.

The Mario 64 submission guide adds the detail that tells you how seriously the clock is taken: you're encouraged to re-time your own run before you submit it, and you shouldn't include milliseconds unless yours is a top time in a short category (community submission guide). Translation for your couch: for most of the board, a tenth of a second is bragging rights. At the top, it's the whole argument, and it's settled by counting frames in your video.

Does your machine count?

Your hardware is part of the run, and the leaderboard splits by it. Mario 64's submission form makes you declare your platform up front — Nintendo 64, Virtual Console, or emulator — because each one lands you on a different part of the board (community submission guide). Emulator isn't a free pass, either. It's a regulated lane.

Yoshi's Island's rules on Speedrun.com show you how tight the lane gets (official community rules). Allowed emulators are named by version — Snes9x 1.51 and 1.53 through 1.56.2. Runahead is banned. So are flash carts, third-party consoles and any emulator with fewer lag frames than the original console release, because running faster than the real hardware is an advantage the real hardware can't answer. Controller tricks a stock controller can't do — macros, turbo, pressing left and right at once — are banned, and so are outside tools, modified game code, cheat devices, savestates and rewind where the platform offers them. If you're picturing a settings page, you're picturing the crime scene: the rules are mostly a list of ways software could quietly make you faster than the machine the record was set on.

What do you have to hand over?

Video. That's the short version, and there's no version without it. Mario 64's guide is blunt: a link to a video of your run "is required," and if you recorded on Twitch you should submit a highlight rather than the stream recording, because stream recordings get deleted automatically after a set time (community submission guide). Your proof has to outlive your stream.

Near the top of a board, the standard jumps again. Yoshi's Island requires every submission that'd place in the top 15 to carry full video proof that stays up permanently — and full video has a definition: one segment, with full audible game audio (official community rules). Two split stream recordings don't count. A recording with partly muted audio doesn't count. Emulator runners on that board also have to show their input display and frame count on screen, and show their settings after the run — emulator version, framerate, runahead. You're not just proving you played well. You're proving the machine behaved.

Who checks it — and what gets a run thrown out?

A moderator for that game checks it, and the first honesty rule is built into the furniture: in Yoshi's Island's own words, "Mods cannot verify their own full game runs" (official community rules, verbatim). Nobody grades their own homework, even at the top of the board their name sits on.

Runs get rejected for reasons that are almost boring, which is the point — boring is checkable:

• Wrong race. You broke a category rule, used a banned glitch in a restricted category, or filed under the wrong category. Mario 64's guide warns in bold that if the rules aren't followed, "your run will be rejected" (community submission guide).
• Broken proof. No video, a video that disappears, split recordings where one continuous recording's required, or audio that's missing or partly muted on a top run.
• Illegal setup. A banned emulator version, runahead left on, a tool or modified code in play, or emulator settings you didn't show.
• A clock that doesn't retime. Your claimed time doesn't survive a frame count against the board's fixed start and end points.

If your run's coming up, do the boring parts before the brilliant one: read your category's rules page first, record locally as well as streaming, keep the full audio on, and re-time the run yourself before a moderator does it for you. And when launch week starts throwing three different numbers at you and calling all of them "sales," read our explainer on copies, players and concurrent players next — it's the same lesson in a different costume: a number only means something once you know who counted it, and how.