💡 Youtube's Timeline Heatmap (Histogram) enables/improves a lot...

the heatmap gives 100 values, one for each percent of the video duration
these are just before the end when going ctrl+U

						"markersList": {
								"markerType": "MARKER_TYPE_HEATMAP",
								"markers": [
									{
										"startMillis": "0",
										"durationMillis": "1234",
										"intensityScoreNormalized": 0.1
									},

this potenially enhances plans about speed-watching / skipping irrelevant scenes. ( beyond what we could do skipping intros or silences )
Because views per scenes correlate with relevance (or else with ignorant* viewers or confusing/deceptive structure of content?)

• 1.) Lows at the Start are indicating irrelvance (intro)
• 2.) Similar at the end. Besides that a slight steady decline is normal / ineviteable (average to some certain degree, simply by the viewer's attention span / patience. - Or forgetting to continue later)

3.) Only playing back the spikes-up, can sometimes save most of the time!
( while getting the gist. Especially for clickbaity videos, with an extreme diagram... )

Except: this could be inverted if the video's audience's mode/mood, can be definied as slow/meditative = Exception for forced speedup:

• Example: Linus Tech Tipps: Retro computing for 20 minutes with one fast/novel sponsor Ad from IBM...
  • so without defining this as inverted/exception this results logically/productively results in skipping most of the whole video, except for the ad. ( And this equals business advice to Linus tech tipps: a playlist (reverse chronologic with sponsors. Some / many people will watch that first or only)).

Content per second: Content difficulty( Ø-Word-length ) & content speed ( Views/WPM )

• Also: common phrases can be considered (as irrelevant) #514
• When (uncommon) text is suddenly spoken faster, viewers will commonly repeat the scene: To approach a spike's meaning it can be divided by an increased WPM ( if any ) (from transcript or audio)
• A high/increasing word-difficulty can be a clue, that the relevance of a video-section is beyond/higher than the views can tell (word-difficulty could be measured by average word-length (and the category/language in general)

( "counter-indication:" "spikes could be inluenced by "clips" - or a popular link with a start-time offset ?t=... or end-time) - yet that's rare & also real... )

• partly implemented in #3628

Duplicate of #1463 (as of #3804)

Start by reviewing duplicate issue #1463 and the partial implementation noted in #3628, then compare the related skipping proposals in #514 and #921. The issue names no files or tests and presents several speculative behaviors, so the first step is to define a single accepted heatmap use case and measurable completion criteria.