Build a Documentary Paper Edit Before Calling It a Rough Cut

A paper edit is a proposal. It is assembled from material another person can find — transcript passages attached to a recording and a timecode, actions written up in a scene log, documents you can point at — and arranged in the order you think the film might go. It is real work. It is not a rough cut. A rough cut plays: picture, sound, a duration someone can sit through. A paper edit is text, and text cannot tell you how any of it sounded.

The confusion usually arrives through a good sentence. A subject says something clear and well formed. You move it into position. It reads beautifully. The document starts behaving like a film that already exists, and from there it is a short step to telling a colleague the sequence is basically there, or putting a runtime in a pitch deck that nothing has been cut to support.

A paper edit earns its place anyway, because it does two things early that an edit suite cannot. It tests whether your material supports the argument you intend, using a text file and someone's patience. And it produces a precise list of what you do not have.

Build from material another editor can locate

An edit instruction is only as good as its address. "Dana says the program is ahead of schedule" is not an instruction. It is a memory. Whoever receives it cannot find the line, cannot check what came before it, and cannot tell whether "ahead of schedule" was said about the whole program or about one street.

So every held beat in the document carries its address. Which recording, which version of it, which transcript file and page, which in and out timecode. Scene logs get the same treatment: log identifier, date, observer, card or file name, and the approximate time of the action. If a recording was re-synced, re-transcribed, or replaced, say which version you are citing. Card names get reused. Files get renamed. A timecode is a locator, not a location: it points at something that still has to exist and still has to play.

Transcripts are usually imperfect, and the paper edit should be imperfect in the same places. Overlapping speech, a word nobody can quite hear, a passage where the room noise ate the sentence — keep those visible. Writing [unclear] or [overlap] in the document is more useful than quietly smoothing the passage into a confident line, because the smoothed version reads like a quotation and will be treated as one. A colleague who repeats your tidy version of a mumbled sentence has now put words in a subject's mouth, in a document, in writing.

One more thing belongs in this section even though it feels like a different department. A documented source location is not permission. Knowing that a line sits at 00:31:02 in INT-03 does not establish that you may publish it, circulate it, or put it in a pitch. Clearance is a separate question with a separate answer, and the paper edit is not the place to assume it.

A public write-up of paper editing from the video provider Reduct — recorded here on 8 September 2026 under the title "Elevating the Paper Edit for Documentary-style Filmmaking," and re-checked on 18 September 2026, when the same URL carried an August 2026 date and a different title — describes the practice as arranged transcript quotations linked to source video and timecodes, sometimes accompanied by notes about visuals and sound. That is one provider's description of one workflow, not an independent assessment of it, and it is not proof that any proposed join will work. The workflow description is useful. The page's claims about saving time are the provider's own, and no transcript, footage or application was tested for this article. The page does name the limit that matters most here: a text arrangement is a place to plan, not a place to judge how the video plays.

Arrange a progression, not a folder of favorite quotations

Before the beats, write one sentence stating the question the sequence pursues. Not a theme — a question. Does the program's schedule slip onto the people living on these streets? Then each beat has a job: it changes what the viewer knows or believes about that question. A beat that does not change anything is decoration, however good the line is.

This is where the folder of favorite quotations fails. It is easy to collect twelve remarkable statements and put them in ascending order of remarkableness. That is a highlight reel with citations. A sequence needs a beat that establishes, a beat that complicates, a beat that turns. A powerful statement may be entirely incapable of doing the transitional work you assigned it, and no amount of rearranging will teach it.

Keep enough of each selection's context to be inspectable. A quotation trimmed to one sentence cannot be checked by anyone, including you in three weeks. Pull the surrounding question and answer into the paper edit, mark the part you intend to use, and let the reviewer see what you left outside the brackets. That outside material is where most of the surprises live — which is the subject of the example below.

It also helps to write a rough duration beside each held beat. Not because the document has a runtime, but because a beat that will occupy twenty-five seconds of screen time should be chosen with that cost in view. Proposed and missing beats do not get durations yet, because nobody can measure them.

Make the invented joins and missing pictures visible

Three categories, marked in the document itself rather than carried in your head:

Held. Recorded speech and recorded action that exists, with an address. A quotation, a described action on a card, a document in hand.

Proposed. Narration you have written but nobody has recorded, bridge text, intended pickups, questions to re-ask, joins you want to make. A join is proposed material. So is a shot you intend to grab.

Missing. Images not shot, access not granted, a question nobody asked, a recording you have not yet located.

Give each beat a status label, and make it visible in the layout: a status column, a bracket, a different typeface. If your proposed narration is formatted identically to your transcript quotations, the document will lie to every reader who skims it. A proposed image described in the same language as a shot reference is a promise you have not made and cannot keep.

Two subtleties are worth the extra line. First, a paraphrase is proposed material. If you have rewritten a subject's muddled sentence into a clean one, that clean sentence is not a quotation, and the only honest options are to ask the person to say it plainly on camera or to move the idea into narration — where it stops being the subject's claim and becomes yours. Second, a planned image is not a shot reference. Write "we want a wide of the street at dawn; not shot" and not "wide, Larch Street, dawn," which reads like it exists.

Return the sequence to sound and image

The example below is invented, so the method can be shown step by step. The project, the names, the timecodes, the transcript pages and the log entries are all fabricated. No recordings were selected, played, or cut.

The invented project is a short documentary about a municipal water crew replacing lead service lines. The held material is an interview, INT-03, recorded on 11 March: Dana Whitfield, the utility's operations lead, camera A, running 00:14:02 to 00:58:31, with a transcript at version 2. There is also scene log SL-07, "Larch Street service tap," observed on 18 March between 07:40 and 13:10, pointing at card A002.

Here is the paper edit, version 1. The sequence is trying to argue that the program is working, that the crew's labor is steady and unglamorous, and that Dana's judgment closes the section.

Now suppose the recordings are played, as they eventually must be. Three things the text concealed become visible, and each one changes the sequence.

The nine seconds outside the selection. The transcript page 11 continues past beat 1's out point. At 00:31:18, immediately after "than the plan called for," Dana adds: "On the streets where we could get the permits. Two of the nine are still waiting on easements." That qualification sits nine seconds outside the chosen span. In the paper edit, beat 1 was the film's premise of success. In the recording, it is the opening of a problem. Nothing about the words changed; the selection did.

There are three defensible responses, and the paper edit's job is to list them, not to choose blindly. Extend the selection to 00:31:27, which makes the beat 25 seconds and gives it a different job. Keep the 16 seconds and stop describing it as a statement about the program's progress. Or find a line that is genuinely unambiguous and let this one go. Whichever you pick, beat 2's narration has to change to match, because it currently sets up a success story that beat 1 no longer supports.

The pause before the verdict. On the page, "It was the right call" reads as a settled judgment, and it was assigned 8 seconds. In the invented recording, the transcript shows eight seconds of silence first — 00:44:04 to 00:44:12 — and during that silence Dana looks off frame to the left. Then the line arrives quietly. What the on-page verdict reads as, the recording delivers as a concession, and the silence is part of the line rather than dead air in front of it. Two consequences follow. The beat costs 16 seconds, not 8. And the look goes somewhere the frame does not contain, so using this moment means deciding how to handle a glance at something the audience cannot see.

The join nobody could hear on paper. Version 1 never states how the section plays, and that omission is doing a great deal of concealing. Beats 1 and 4 are two passages from the same interview, twelve minutes and forty-five seconds apart: beat 1 ends around 00:31, beat 4 begins at 00:44. Cut literally as the table reads, they are two separate segments, with the narration and the Larch Street picture between them. Tightened, as the section wants to be, they are laid together as one continuous audio spine, with beats 2 and 3 as the covering picture carried across the join. Only one of those can be the film, and the second is the one the page hides, because two rows of a table sitting one above the other read as one continuous thought.

Laid together, the two passages announce themselves as two passages. In the earlier one Dana is seated at a desk with a window behind; in the later one, standing beside a wall map. The room changes, the framing changes, and the join has to be covered. This is the concealed seam that most often reaches a rough cut and gets discovered there, expensively.

It also has a productive side effect, though a smaller one than it first looks. The beat that would carry the audience across the seam is beat 3, the Larch Street tap on card A002 — it is already the intervening beat in the design, the only picture in the section that could cover the join. So the seam adds no new requirement to the sequence; it raises the stakes on one the sequence has already taken on. Card A002 was assigned 20 seconds of screen time and labelled "Held, recorded action, unviewed," and its open question is whether the camera was rolling for the tap itself. The log notes a battery change mid-morning and a camera repositioning, and it does not say. If the answer is no, the covering picture does not exist, and the join has to be solved another way — a cutaway from a different day, or a shorter version of one passage that plays clearly as a separate moment. Which is the whole point of the next section.

Here is the revised sequence, with its gaps stated as questions.

The gap list, which is the most durable thing a paper edit produces:

• Does card A002 contain the tap itself, or only the setup and the notice left at the door?
• Does any coverage exist across the 00:31 to 00:44 camera change, or does the join need a cutaway from another day?
• Is the homeowner willing to have the basement filmed, and if not, what does beat 5 become?
• What is off frame during the pause at 00:44:09?
• Do the releases covering INT-03 and SL-07 permit use in a pitch, as distinct from use in a finished film?

Notice what the revision did not do. It did not invent a better line, restore a missing shot, or decide the question of the pause. It made the options and the costs legible, which is what a text document can honestly do.

Describe the result accurately in the pitch

What you have in hand is a written assembly. Say so, in those words, and then say the useful things.

This is a paper edit — a written sequence assembled from identified sources. It asks whether the program's delays land on the people living on these streets. It draws on about forty seconds of recorded interview across two passages of one interview, an unmeasured amount of narration not yet written to final, one scene log whose footage has not been viewed, and an image nobody has shot.

That is a real editorial proposal, and it is checkable. A producer can read a sequence like this and push back on the argument, question a source, or see immediately that the missing basement image is the thing standing between the pitch and a shooting plan. Nobody has to take your word for anything.

What not to say is easier to list. Don't call it a rough cut. Don't attach a runtime — you can total the held seconds, but narration and unshot images have no duration, so the number would describe nothing. Don't describe the proposed bridge as footage, or the intended pickup as recorded. Don't imply that unseen material supports the argument. And don't let anyone circulate the document without its status labels, because a paper edit that looks finished is worse than no paper edit at all.

Then hand it over in a state another editor can use. Every held beat addressed to a file, page and timecode. Every proposed beat labelled as proposed. Every missing piece written as a question with an owner rather than a vague ambition. When those three things are true, the document can survive being argued with — which is the only test worth putting it through at this stage.

Expect it to change. In the invented example above, the sequence did not break; three of its five beats changed status or meaning the moment the recordings were played, and the argument came out narrower and more defensible for it. Most paper edits do something of the same. Plan for that, rather than treating the written arrangement as a result it was never able to be.

Frequently asked questions

What is the difference between a paper edit and a rough cut?

A paper edit is a written proposal arranged from locatable material such as transcript passages with timecodes, scene logs, and documents. It is not a rough cut because a rough cut plays picture and sound and has a duration someone can sit through. Text cannot tell you how any of it sounded.

What address information should each held beat carry?

Each held beat should identify the recording and version, transcript file and page, and in and out timecode. Scene log entries should carry log identifier, date, observer, card or file name, and approximate time. If a recording was re-synced, re-transcribed, or replaced, say which version is cited. A timecode is a locator, not proof the material may be published.

What do Held, Proposed, and Missing mean in the document?

Held is recorded speech or action that exists with an address. Proposed is narration not yet recorded, bridge text, intended pickups, questions to re-ask, or joins you want to make. Missing is images not shot, access not granted, a question nobody asked, or a recording not located. Give each beat a visible status label so the document does not lie to skimmers.

Why keep unclear or overlapping transcript passages visible?

Smoothing a passage into a confident line makes it read like a quotation and can lead someone to repeat it as if it were the subject's exact words. Writing unclear or overlap keeps the document imperfect where the transcript is imperfect. That is more useful than quietly putting words in a subject's mouth in writing.

What did the invented water-crew example reveal when recordings were played?

Three changes appeared. A qualification about permit delays sat nine seconds outside beat 1's chosen span, turning a premise of success into an opening problem. A pause before the verdict cost eight extra seconds and included a look off frame, making the line read as a concession. And two passages from the same interview, twelve minutes and forty-five seconds apart, had been read on the page as one continuous thought, though they would need covering picture if laid together.

Further reading

• Quote a Documentary Participant Without Changing What They Answered — Before relying on a held quotation, return to the exchange it came from and check whether the question, hedge or scope around it was cut away.
• Translate Documentary Testimony for a Pitch Without Changing What Was Claimed — When testimony passes through translation, keep the recording, transcript, translation version and pitch sentence as separate records so the pitch cannot assert more than the source did.