Give every minute of a long video a job

Long videos drift when the middle has no assignment. You can feel it in the retention curve: a gentle, constant decline with no event, because nothing in minutes three through seven was responsible for anything. The opening promised a result. The ending eventually delivered it. The middle was "more of the video."

The leaked MrBeast production document splits the runtime on purpose. Minute one confirms the thumbnail's promise. Minutes one to three carry visible escalation. Later blocks have their own briefs, and re-engagement beats are scheduled (the crane arriving at 0:30 is the example they wrote down) rather than discovered in the graph afterwards. You do not need their scale to steal the method. You need a sheet that names a job for every block before the script exists.

A job is not a chapter title. "History of the format" is a topic. "Make the thumbnail's claim visible on screen" is a job. If you cannot say what a minute is for, that minute is padding.

The jobs, in order

Use this as the default spine for an 8–15 minute video. Shorter videos keep the same jobs and collapse the clock. Longer videos split the middle, they do not invent a fifth kind of job.

Block 0: the first frame (0:00–0:15). Confirm that the click was not a mistake. The hook is the first frame, not the first line of voiceover. If the thumbnail showed a cracked housing, the cracked housing is on screen before you introduce yourself. Channels that lose this block usually lose it to a logo sting, a "hey guys," or a wide establishing shot that could belong to any video. The craft of that opening window is the same problem as the first three seconds, just with a longer fuse.

Block 1: the promise check (0:15–1:00). Minute one has one job: prove the title was not bait. You do not deliver the whole answer. You deliver enough of it that a viewer who clicked "I rebuilt this in 48 hours" has seen the rebuild start, not heard a lecture about why 48 hours matters. If minute one is context, the people who came for the result leave, and they leave before your best material.

Block 2: escalation (1:00–3:00). Something has to get bigger, stranger, or more specific than the thumbnail implied. "Crazy progression" in the leaked doc is this block. For a tutorial, that is the first full worked example, not a second round of framing. For a commentary video, that is the first claim you could not have fit on the thumbnail. If this block restates Block 1, you have already started the slow bleed.

Block 3: the paid-for middle (3:00–6:00, or 3:00 to the two-thirds mark). Name a new responsibility: the mechanism, the second example, the thing that would be a video on its own. A list can continue here only if the items are heavier than Block 2. A review uses this block to test, not to recap the spec sheet.

Block 4: the close (last third, often 6:00 to end on a 10-minute cut). Resolve the original promise, then leave one open loop that is not "subscribe." The leftover loop can be Part 2, a related video, or a question the comments can actually answer. End-screen CTAs that fire before this resolution read as exits. The hook / body / CTA script shape still applies; the CTA sits after a payoff, not in place of one.

Scheduled resets, not "energy." Independently of the blocks, plant at least one visible reset inside each block after minute one: a reveal, a location change, a number on screen, a burst of faster cuts. Put them on the clock at draft time. "We will feel it" is how resets arrive 40 seconds after the dip they were meant to catch.

Put the job on the storyboard card, not just the shot. The storyboarding guide is the visual half of the same discipline.

The block sheet (fill this before the script)

Copy the table. Fill the right two columns in a sitting. If a cell in "Job in one sentence" takes more than a sentence, the block is doing two jobs and should be split or cut.

Block	Clock (10-min video)	Job in one sentence	Proof the job happened (what is on screen)	Reset inside the block
0 First frame	0:00–0:15	Confirm the thumbnail in the picture, not the voiceover	Thumbnail object or result is already visible	—
1 Promise check	0:15–1:00	Viewer can tell the title was earned	A piece of the result, started or shown	One specific number or constraint
2 Escalation	1:00–3:00	Bigger than the thumbnail implied	First full example / first claim you could not thumbnail	Pattern interrupt at ~2:00
3 Paid-for middle	3:00–6:00	New responsibility, not a restatement	Mechanism, test, or second example	Reveal or location change at ~4:30
4 Close	6:00–end	Original promise resolved	The result, complete	Open loop after the resolution

Worked example, filled for a video titled "I ran the same prompt on four video models":

Block	Job in one sentence	Proof on screen
0	The four outputs are visible before I talk	2-second grid of the four clips
1	I show the identical prompt and the first obvious miss	Prompt card, then the worst failure, unedited
2	One model is clearly ahead on motion; I say so	Side-by-side of the motion pass, no commentary padding
3	I rerun the loser with a tighter brief and show what changed	Before/after of one model, not all four again
4	I pick a winner for this brief, and name the brief it would lose	Winner clip, then a one-line "don't use it for X"

If you cannot point to the proof column, the job is still a wish. Channel-trailer energy and a pause at 5:00 with nothing to say are how Block 3 dies.

How undefined blocks show up later

Undefined blocks leave a signature.

• Slow bleed from 2:00 to the end. Blocks 2 and 3 were topics, not jobs. Raise the density or cut the runtime. Pacing can be part of this; the cut-rate ranges by platform are a check on habituation, not a substitute for a job.
• A cliff at a chapter marker. The previous block finished too cleanly. End Block 2 on a question Block 3 must answer, not on a summary.
• A cliff at 5–6 minutes across several videos. That is a length problem, not a block problem. The idea was a 6-minute idea. Publish it at 6.
• Healthy hold until the last 90 seconds, then a slide. Block 4 started recapping. Recaps are for people who already left.

When a single beat inside an otherwise named block fails, recut that beat. On an EDL timeline that is a range delete and a re-render: the editor's 480p preview pass is free, it carries a short per-user cooldown, and a single charge applies to the final export only. A one-second offender is a segment retake, not a reason to scrap Block 3.

Write the jobs first, the sentences second

The production order that keeps blocks honest:

1. Write the title and describe the thumbnail in one line. That pair is the promise Block 0 and Block 1 have to confirm. If you cannot describe the thumbnail, you are not ready to assign jobs.
2. Fill the sheet. Five jobs, five proofs. If Block 3's proof is "more examples like Block 2," you do not have a Block 3. Cut the runtime or find a second responsibility.
3. Only then write voiceover. Each paragraph must serve the job of its block. A good sentence in the wrong block is still a leak.
4. Mark resets on the clock. If your last eight videos dip around 2:20, the Block 2 reset sits at 2:05, not "somewhere in the middle."
5. After publish, score the sheet against the curve. A miss in a block you named is an execution miss. A miss in a block you fudged is a planning miss. Log which one it was or you will rewrite the same middle forever.

Generated b-roll will happily fill three minutes that have no job. The sheet is the constraint that stops you taking that deal.

FAQ

What if the video is four minutes, not ten?

Keep all five jobs and shrink the clock. Block 0 is still the first frame. Block 1 is still the rest of minute one, or the rest of the first 45 seconds. Blocks 2 and 3 collapse into one paid-for middle. Block 4 is still a resolution plus an open loop. What you must not do is drop Block 1 because the video is "short enough to skip the promise check."

Is a job the same as a chapter title?

No. A chapter title is a label the viewer sees ("The test," "Results"). A job is the reason that minute exists ("Show the failure case before the winner so the comparison is fair"). You can have chapters without jobs; that is how you get tidy markers sitting on top of a bleed. Write the job first. Name the chapter after.

Do I have to use these exact timestamps?

No. Use percentages if runtime jumps around: first 10 percent, next 20, next 30, last 40. Keep the order: confirm, escalate, pay off, close. Starting with context and confirming at minute four spends your most expensive viewers on a preface.

Can I reuse the same sheet every week?

Reuse the jobs. Rewrite the proof column every time. A sheet that still says "first full example" with no named example is a template, not a plan. The week the proof column is hard to fill is the week the idea is a Short, not a 10-minute video.