I Built a Short-Form Recut Study That Has to Publish the Failures

Underperforming must be defined before the edit is seen

A creator calling a video a flop is understandable, but it is not a stable inclusion rule. The study should define underperformance against that creator's own recent same-format baseline using a named metric and fixed time window. The threshold is frozen before reviewers see the cut.

A channel with ten typical Shorts at 20,000 engaged views and one at 8,000 presents a different case from a new channel with three uploads and zero feed exposures. The first may support a structural comparison. The second may be a visibility, eligibility, or distribution question that creative recutting cannot diagnose.

Paid distribution, policy restrictions, rights disputes, materially changed topics, and missing baseline data should be excluded or reported separately. Recruitment should not select only videos where reviewers already see an obvious fix; that would bias the case series toward convenient success stories.

Participant inclusion and exclusion rules

The frozen recut workflow

Every step creates an audit trail before the next result is visible.

One primary change is the price of interpretation

A recut often improves several things at once: the opening line, first frame, caption onset, dead air, music, duration, title, and upload time. That may be excellent editing and terrible measurement. When six variables move, the result cannot identify which one mattered.

The protocol allows one primary structural change and no more than two secondary changes needed to make it coherent. A primary change might move the proof into the first second, replace a vague opening clause, or remove a setup interval. The change log preserves all downstream effects, such as the video becoming shorter.

The creator may reject the proposed recut. That rejection is data about feasibility and creative intent, not noncompliance. The study should never optimize the file so aggressively that it stops being the creator's work or changes the claim being tested.

Two evidence arms, two different questions

The viewer panel isolates perception, not platform performance

For each of the 20 pairs, at least 30 target viewers should see the original and a separate 30 should see the recut. Assignment is random, and no participant sees both versions. Showing both would create memory, demand, and comparison effects that do not resemble a cold encounter.

The first task asks viewers to state the premise after five seconds in their own words. Only then should the study ask intention to continue, promise recall, and payoff clarity. Free response prevents a generous multiple-choice option from supplying the missing context.

The panel can support a causal claim about that controlled presentation if the design and analysis hold. It cannot establish what the platform would recommend or how a creator's audience would behave. Those are separate estimands, and the report should keep them separate even when the directions agree.

Matched windows reduce noise but do not create a controlled feed

Both uploads should be read at the same elapsed windows—such as 24 hours, seven days, and 28 days—using the same metric definitions. YouTube's Shorts analytics guide distinguishes shown in feed from how many chose to view; those fields must stay separate from public views. Calendar time, weekday, season, account state, recent channel activity, traffic source, and external promotion should be recorded.

A platform can expose the recut to a different audience or no comparable audience at all. A creator's history has also changed because the original upload happened. Even identical publish times do not make the observations exchangeable.

The case narrative should therefore say, 'The recut had X under these conditions,' not, 'Moving the hook caused X.' Individual cases come first. Any pooled summary treats creator as the unit, shows heterogeneity, and remains exploratory.

Consent has to cover the uncomfortable outcomes

• Video ownership and permission for reviewer access, processing, storage, and any frame publication.
• Exactly which analytics will be collected, who can access them, how long they are retained, and how they are de-identified.
• Whether the creator may be named, described by niche only, or fully anonymized.
• The withdrawal deadline and what can no longer be removed after aggregate publication.
• The possibility that the recut performs worse or that the viewer panel prefers the original.
• The commitment to publish every completed case under the same inclusion rule.
• A clear statement that participation does not guarantee reach, retention, or a positive case study.

Preregister the decision tree, not only the hypothesis

The primary outcome, observation windows, exclusion rules, missing-data treatment, minimum exposure, and pooling decision should be timestamped before recut outcomes are known. OSF's registration guide recommends explicit hypotheses, variables, tests, exclusions, contingencies, and planned versus unplanned analyses.

A useful preregistration includes if-then rules. If the recut receives no feed exposure, report distribution separately and do not score audience response. If a creator changes the title after publication, retain the case but mark the deviation. If analytics disappear, publish the missingness rather than substituting a public counter.

Deviations are expected; hidden deviations are the problem. Each case sheet should show what changed from the plan, why, when the team learned the outcome, and whether the case remains in the primary analysis.

Why failure reporting is a study requirement

If creators can withdraw because the recut underperformed, the public case series becomes a success gallery. If ViralJury chooses which completed cases count after seeing the graphs, the estimate becomes a marketing decision. Consent and publication rules need to prevent both paths.

A peer-reviewed perspective on preregistration and calibrated confidence describes outcome-independent decision-making and transparent assessment of bias as practical reasons to register plans. For this case series, the concrete consequence is simple: eligibility, outcomes, and the publication rule are fixed before ViralJury knows which recuts look flattering.

A worse result can be more informative than a win. The new opening may improve premise comprehension while narrowing the audience, change the promise, or weaken the emotional buildup. A neutral result may show that a structurally cleaner edit met a different distribution state.

The report should preserve strengths the recut damaged, not only problems it fixed. The point is to learn which structural changes travel across formats and which tradeoffs depend on creator, audience, and story.

Claims this protocol cannot support

• ViralJury recuts improve retention or views.
• A repost is a controlled A/B test of the edit.
• The hook change caused any observed platform difference.
• A viewer-panel preference predicts recommendation or revenue.
• Only positive completed cases need to be published.
• One pooled average can replace the creator-level cases and heterogeneity.

Publication gate and product boundary

The study can publish results after 20 consented completed cases, frozen change logs, matched observation windows, preregistered missing-data rules, all failures reported, and the viewer arm reaching its registered sample. Every metric must keep its platform-native definition and collection date.

ViralJury should not be the reference that decides its recut was correct. Independent viewers evaluate perception, and creator-owned analytics record the platform case. The tool's diagnosis is one intervention in the study, not the ground truth.

Until those gates are met, the only honest product statement is that ViralJury is being built to provide pre-upload structural feedback. This protocol is not a customer case study, testimonial, or measured before-and-after win.