Progress Report on Caption Before Video
Reading the caption first has become a filtering behavior that extends rather than limits short-video consumption.
Progress report covering the period in which reading the caption before watching the video was identified as an efficient filtering strategy for short-video feeds. The strategy has performed according to its design. The aggregate outcome was not anticipated.
Caption before video extends the break by one clip. This is the first finding. The filter is real: captions that indicate low relevance do result in skips. The filter is also effective at increasing confidence in captions that indicate high relevance, which results in plays that would not have occurred without the pre-read. The net effect on time spent is positive, not negative.
The feed removes clear boundaries between the next clips. This is the context in which caption-before-video operates. The filter was designed to improve clip selection. It was not designed to interrupt the feed structure. The feed has no pause between clips. The caption pre-read occupies the pause that does not exist. The next caption has already loaded.
Under the governing rule, caption before video becomes procedure once repetition feels safer than reconsidering it. A person who has pre-read captions across thirty clips has not made thirty decisions. They have made one decision, on the first clip, and executed a reflex thirty times since. The filter is now part of the consumption behavior, not a modifier of it.
Each clip is easy to leave. The feed removes a natural stopping point. These two facts interact poorly with caption-before-video because the pre-read extends engagement exactly long enough for the next clip to begin.
The break becomes a session requiring a separate act of will, and the act of will does not arrive on its own. The progress report notes this outcome is consistent across all measured cohorts.
Recommendation: none available within the current feed design.