Original research protocol
I Counted Progress, Not Just Seconds, in the Payoff Distance Study
A proposed annotation study that measures both elapsed time and intervening beats, because equal waits can contain very different amounts of progress.
Two videos can delay their result for twelve seconds and feel nothing alike. One uses the time to run a test, reveal evidence, and narrow the answer. The other repeats 'wait for it' over three beauty shots. A stopwatch gives both the same number.
The Payoff Distance Study adds the missing structure. It marks the opening promise, the first proof that bears on it, the event that supplies the strongest resolution, and every meaningful beat in between. Time remains useful, but it stops pretending to explain the journey on its own.
The study is descriptive by design. It can map how a registered sample constructs payoff. It cannot declare that six seconds is ideal, that earlier always wins, or that the platform punished a long setup.
The promise-payoff thread is the unit
A video is the container. The analytical unit is a thread: one identifiable expectation linked to the evidence and event that resolve, reject, complicate, or reframe it. One video may contain several threads. A ranked list can have a parent promise plus a smaller thread for every item.
Do not force a thread onto a mood piece, performance, ambient clip, or intentionally open ending. The registered state 'no discrete thread' is a result. It is better than inventing a promise because the schema expects one.
For every thread, store frame boundaries and exact evidence. Later knowledge can make an opening seem more specific than it was, so reviewers mark the promise before watching or coding the payoff whenever the interface permits staged annotation.
Four events that define the path
| Event | Operational definition | Common mistake |
|---|---|---|
| Opening promise | An explicit or strongly implied question, claim, task, tension, comparison, or transformation the video sets up. | Treating a generic teaser such as 'wait for it' as a complete promise when the viewer cannot state what is awaited. |
| First proof | The earliest concrete evidence that can support, challenge, or narrow the promise. | Counting a repeated assertion or unrelated beauty shot as evidence. |
| Primary payoff | The event that most directly resolves, demonstrates, answers, or meaningfully transforms the recorded promise. | Choosing the last event merely because it appears last. |
| Intervening beat | A distinct event that adds context, progress, evidence, complication, stakes, preparation, repetition, diversion, or transition. | Counting every camera cut as a new narrative beat. |
| Payoff distance | Frames, seconds, normalized duration, and beat count between the registered promise boundary and payoff onset. | Publishing elapsed time without the progress record. |
| Payoff density | The number and type of progress-bearing events across the thread, reported with duration and uncertainty. | Equating more cuts or more text with more value. |
Progress is coded before delay is judged
Each intervening beat receives a functional label: context, process progress, evidence, complication, stakes clarification, emotional preparation, repetition, unrelated diversion, transition, or cannot determine. The label says what changed in the thread, not whether the reviewer personally enjoyed it.
This makes suspense visible without granting it a free pass. A reaction hold may prepare emotion. A test step may narrow the answer. A repeated promise may add nothing. Reviewers record the function and confidence first; any later classification of earned progress or avoidable delay uses the frozen rule.
The source material's goal matters. In a restoration video, the process may be the promised experience. In a one-question product test, repeated setup before the first evidence may be less defensible. The dataset preserves this distinction instead of imposing one fast-edit template.
Edge cases the codebook must keep
| Case | Treatment | Why |
|---|---|---|
| Payoff shown first | Record a zero-distance reveal and a later explanatory or causal thread if one exists. | Result-first editing is not missing data. |
| Several promises | Create linked threads and identify the primary thread under a registered emphasis rule. | Collapsing them hides partial and unmet promises. |
| Several payoffs | Store local proof, primary resolution, and consequence or callback as separate linked events. | The strongest resolution need not be the latest event. |
| Ambiguous promise | Record at least two plausible interpretations and do not compute one definitive distance until the rule resolves them. | Vague anticipation can support several incompatible threads. |
| Missing payoff | Retain the thread with a missing-payoff code; exclude only from calculations that require an observed endpoint. | Deleting it creates survivorship bias. |
| Loop | Annotate the first complete narrative cycle and record the loop construction. | A loop boundary can carry the payoff and the next promise at once. |
| Multi-part story | Include only a self-contained local thread or code the payoff as external and incomplete. | A future episode should not be invented as an observed resolution. |
Annotation workflow
Verify rights, consent scope, source version, timebase, language, intended audience, and whether the unit is self-contained.
Mark opening promise existence, type, boundaries, literal words or visuals, and confidence without public performance data.
Mark first proof and every intervening beat, with frame boundaries and functional labels.
Select the primary payoff or a registered ambiguous, absent, incomplete, or no-discrete-thread state.
Link events into threads, then derive elapsed frames, seconds, normalized distance, beat count, and progress mix.
Lock independent ratings before adjudication; preserve boundary and category disagreement in the released analysis.
A worked payoff annotation before any timing judgment
This synthetic 18-second recipe example shows why the clock alone is not the diagnosis.
| Event | Time | Annotation decision |
|---|---|---|
| Promise | 00:00.8 | The opening promises a crisp result without deep frying. |
| Progress beat | 00:03.4 | The coating choice supplies new causal information, so the interval is not empty delay. |
| First proof | 00:08.7 | The cut reveals the crisp interior before the full serving shot. |
| Primary payoff | 00:14.2 | The finished result resolves the original promise; later seasoning is an optional secondary beat. |
Analysis without an ideal timer
Permitted outputs include distributions of observed distances, beat counts, promise types, payoff types, ambiguous cases, missing payoffs, and reviewer agreement in the registered sample. Medians and quantiles should sit beside the full distribution because narrative timing is likely to be skewed and multi-modal.
A model may examine whether longer threads contain more progress beats or whether observed structures differ by declared format. It should not correlate payoff distance with ViralJury's overall structural score as primary validation if that score already penalizes delay. That would prove the product agrees with itself.
Any link to comprehension needs a separate viewer task. Any link to native performance remains observational unless a stronger design justifies causal language. Platform recommendation is a third question again. The study stays credible by keeping those targets apart.
Publication package
- Preregistration, codebook, rights summary, sampling and exclusion flow, data schema, and every deviation.
- Separate category agreement and temporal-boundary agreement, with raw ratings preserved.
- Distributions for ambiguous, absent, and multiple-thread cases rather than a cleaned sample that contains only neat stories.
- Versioned, consent-safe event tables and analysis code when permissions allow.
- Synthetic examples labeled as synthetic and never counted as observations.
- Headlines that describe the sampled structure without promising an ideal timer or audience outcome.
Sources and protocol foundations
The event model is informed by research on continuity editing and event segmentation and a broader review of event segmentation in reading and film. Those studies motivate event boundaries; they do not validate this exact codebook.
Research on curiosity and information gaps supports studying unresolved information. It does not establish a universal payoff deadline or justify vague withholding.
The agreement plan follows the multi-rater approach described by Hayes and Krippendorff.
The confirmatory codebook and analysis should be frozen in a timestamped repository such as OSF Registrations before the main sample is annotated.
Frequently asked questions
What is the best payoff time for a Short?
This study does not assume one exists. The useful timing depends on the promise, content type, intervening progress, audience, and goal. The protocol maps those elements before it compares durations.
Is an immediate payoff always better?
No universal claim is justified. Result-first editing can clarify the value, while some stories, jokes, tests, and transformations need build-up. The study records first proof and progress so delay is not judged by time alone.
How is payoff distance different from video length?
Video length describes the whole file. Payoff distance describes one promise-payoff thread and includes both elapsed time and the number and function of the beats between those events.
Has ViralJury collected a payoff-distance dataset?
No. This page publishes the proposed methodology. There is no median, niche table, platform comparison, or optimal timing statistic to cite yet.