Creators watch the first hour after publication closely, and it is the least informative period available. The audience it measures is not the audience that determines the video's outcome.
The first hour is a subscriber measurement
Early views arrive predominantly from notifications and subscription feeds, which reach people who already follow the channel and would have watched almost anything published.
That group is roughly constant between uploads, so the first hour tends to produce similar numbers regardless of the video, which is exactly why it discriminates so poorly.
A video that performs unusually well or badly with subscribers is telling you something about the title and thumbnail, not about whether the video will find a wider audience.
Recommendation operates on a slower clock
Broader distribution begins by testing the video with small groups of non-subscribers and expanding gradually where the response justifies it, which takes considerably longer than an hour.
Each expansion requires enough data to be confident, so the process advances in steps rather than continuously, and a video can sit apparently flat before moving sharply.
This is why videos sometimes accelerate days or weeks after publication, having been judged a failure by everyone watching the first evening's numbers.
Different traffic sources mature differently
Search traffic accumulates slowly and steadily, because it depends on people asking a question rather than on being shown something, and those questions are asked continuously.
Suggested placement alongside other videos builds as the system establishes which content this video pairs well with, which requires viewing history that does not exist at publication.
A video weighted toward either source will therefore look weak on day one and continue growing long after a subscriber-driven video has finished its run.
Acting on early numbers causes damage
The common response is to change the title or thumbnail within hours, which discards the data the system has just gathered and restarts the process of learning who responds.
Repeated changes compound this, leaving the video permanently in an early testing state and producing the poor performance the changes were meant to prevent.
What is worth watching instead
Retention within the first minute is meaningful immediately, because it reflects the video itself and is measured on whoever has arrived rather than on how many.
Comparing the shape of the audience curve against previous uploads after several days gives a far more reliable signal than any absolute number gathered on the first evening.