Screenburn: What a Retention Score Cannot Prove

A caretaker locks a greenhouse while a neighbor holds seedlings

A caretaker has to close a community greenhouse for repairs. One opening reads the notice aloud. Another gives her the key while a neighbor is still outside. Both openings deliver the same facts. Only the second makes her the person who has to act. That distinction matters when a service assigns a short-drama script a retention score before any viewers have seen a cut.

Create a short-drama opening you can screen

Screenburn’s Writers Room and Retention Engine give teams different ways to work with scripts. Here is a screening worksheet for separating three kinds of evidence, with two original greenhouse openings you can use to practice. The openings carry no vendor scores or observed audience results.

Put the new Writers Room report on the right timeline

Put dates beside each statement you take into a production meeting. The latest article, the marketplace launch, and the present product description answer different questions.

November 12 industry coverage

Broadcast’s November 12 report describes Screenburn Group’s Writers Room and the use of a writer’s plan as input. For a small team, the relevant question is what a draft from that process gives an editor to read and revise. The report is coverage of the product, with the product claims attributed to the company.

September 30 scored-script marketplace announcement

Screenburn’s September 30 marketplace announcement described Screenburn Originals as a marketplace for scripts with scores. Treat the scoring and marketplace statements as a separate event on the timeline. If a producer sends you a screenshot with a score, ask which script version it came from and what the team hoped to learn from it.

What the current official page confirms

Screenburn’s official site describes Writers Room and Retention Engine separately. The distinction is useful: a writer can evaluate a draft workflow and a scoring workflow with different questions. Keep the official descriptions in the production notes and give any later viewing data its own source and date.

In a working folder, save the product description with the date it was read. Put it next to the actual script revision if you try the tool. That small record prevents a later discussion from treating an earlier marketing announcement as an account of what the current interface did. If the script changes after scoring, keep the old score on the old revision.

Keep the draft score outside the audience-results column

A single “retention” spreadsheet column invites people to compare entries that were collected in different ways. Use these three columns and label their units.

A software judgment about a script

Column 1 is the software’s judgment about one script revision. Enter the provider, the submission date, the revision ID, the score as displayed, and any explanation you actually received. The unanswered question is what a viewer will understand or feel once the scene has been performed and edited. Leave the viewer fields empty at this stage.

A person reacting to a cut

Column 2 is one person’s response to a particular cut under stated conditions. Enter what they said, the point they identified, and whether they watched with sound. Save their words before adding an interpretation. A response can suggest where to look in the edit, while the unanswered question is how other viewers will respond.

Recorded viewing behavior after release

Column 3 is viewing behavior from a released version, with platform, reporting window, and relevant traffic context. YouTube’s retention help page describes a report about how moments held attention in a specific video and explains how to interpret patterns such as spikes or dips. A dip tells you where to investigate. A viewing metric for one release does not explain the reason on its own.

When a producer asks for a single summary cell, use a sentence with provenance instead: “Revision B is readable enough to screen. We have one product score, no screening comments yet, and no released viewing data.” Rewrite the sentence when evidence arrives. The sheet should make missing evidence obvious without leaving people to infer it from a zero.

Write an opening whose two versions deserve different questions

Keep the same facts in both versions. The door must shut today because a damaged pane needs repair. A neighbor is waiting outside. The caretaker has the key.

A ceremonial announcement with no human stake

Version A: “The community greenhouse will close today. A pane was damaged, and repairs are required.” Lian reads the notice off the clipboard while a neighbor waits at the door. The key sits in Lian’s palm. The scene can end with the closure clearly announced and the neighbor still waiting.

A small decision with a personal cost

Version B: The neighbor puts a tray of seedlings down outside. “Can I just water them first?” Lian looks at the cracked pane, then slides the key into the lock. “Leave them with me. I’ll water them in the yard.” Lian shuts the door. The closure, reason, neighbor, and key are all still present. The scene gives Lian a decision the viewer can name.

The offer to water the seedlings is deliberately small. The change in the opening is who carries the inconvenience. In the first version, a voice announces a rule. In the second, the actor handling the key has to enforce that rule and offer a practical next step. Keep the cracked pane visible in both cuts if you film them so the viewer has comparable context.

The question a reviewer should answer unaided

Ask “What did Lian decide, and what did that decision cost her?” A participant who says only “the greenhouse closed” may have understood the facts without noticing that Lian had to turn someone away or take on their work. Keep the answer in the participant’s words. The question is about the opening’s legibility and stakes, independent of whether someone prefers that kind of story.

Collect a reaction before revealing the tool label

A viewer who has already seen a vendor score has been told what to look for. Keep the score off the screening notes until you have captured the responses.

Show a comparable cut without the score

Use the same recording conditions for the two cuts where possible, including volume, captions, and image quality. Mark any variation you cannot control. The GOV.UK research guide offers a useful principle from moderated testing: state the task in neutral language and avoid giving away the answer. Here, a simple invitation to watch the clip is enough.

Set a viewing order before the first session. If you can show only one cut to each participant, record which one they saw. If someone sees both, ask about the first cut before showing the second and record the order. The second viewing may benefit from the information the viewer learned in the first. Do not collapse both answers into a single reaction.

Ask where interest changed and why

Try “Where did your interest change?” followed by “What was on screen or being said then?” Accept a beat such as “when she turns the key” if a timecode is inconvenient. Keep direction separate from cause. Someone who paused at the lock might have noticed the decision, needed to read the subtitle again, or been interrupted away from the screen.

Record skipped or unanswered responses

Use one row per answer: cut ID ___, viewing condition ___, response ___, time or beat ___, reason given ___, uncertainty ___, skipped or unanswered ___. Preserve blanks and interruptions. Add any notes about order if someone saw both versions. A handful of people responding in one setting can inform an edit within that setting.

Decide what to revise when the evidence disagrees

A disagreement can point to a concrete revision. Keep the score, the viewing response, and the release record visible when you choose what to change.

Before the next meeting, put an owner and date beside the proposed revision. A writer might take the line, an editor might take the cut, and the person gathering responses can keep the release data separate. That way a change to one column does not quietly invalidate the others.

Strong score but an unclear personal stake

Hypothetical result: the vendor score is favorable, and a viewer says “a building was closing” without naming a decision. Review the beat where Lian takes responsibility. You could hold on the key as she offers to water the seedlings. Keep the existing ending until you learn whether the problem begins with the opening or later.

Readable scene with limited viewing data

Hypothetical result: a viewer can explain Lian’s choice, and there is too little release data to interpret a curve. Save the cut, record what was understood, and give the public data its own observation window. If people instead understand the choice and lose interest near the cut to black, the cliffhanger guide can help diagnose the episode’s final beat.

A revision that needs another actual viewing

Hypothetical result: feedback repeatedly points to the key turn but for conflicting reasons. Revise the specific shot or subtitle, then show the new cut under comparable conditions. The APOB short-film workflow can supply a scene asset for such a prototype. Record the exact file reviewed and give any new response its own row so it remains possible to see what changed.

One useful revision log reads “changed only the hold on the key; kept the dialogue, order, and sound.” Another reads “changed both the line and the camera angle, so the next review cannot isolate the cause.” Both are legitimate production choices. The notes let a collaborator understand what a later comparison can establish.

Sources

HappyHorse 1.1 vs HappyHorse 1.0: Feature Analysis for AI Video GenerationAPOB AI generated AI imageSoft gradient background of the APOB AI call-to-action banner

Create a dreamlike vision with APOB

No credit card needed

CREATE FOR FREE

You may also be interested in