Read the measurement definition
The default reading-speed check counts non-whitespace Unicode code points and divides by cue duration in seconds. It flags values over 20 code points per second. This is a clearly defined approximation: combined glyphs and different writing systems do not have identical reading difficulty, and no single threshold fits every audience or broadcaster.
Line length is checked against 42 Unicode code points per line, including spaces. The rendered width also depends on font, language, player layout and the viewer’s accessibility preferences. A line below the threshold can still wrap, while another can fit comfortably. Use the flag to find candidates for review rather than to certify a caption track.
Treat overlaps as a review queue
An overlap is detected when a cue begins before an earlier cue has ended. This includes a short cue inside a longer one, not only adjacent pairs with obvious conflicts. Intentional overlaps may describe different speakers or text on screen. The editor reports them without deciding which cue is correct.
Watch each flagged section with audio and picture. Adjust boundaries only when that improves the communication. Automatic shortening can remove reading time or hide part of a sentence. If two cues belong together, merge them and then review the line breaks; if they represent simultaneous information, retain the overlap and test how the destination player handles it.
Combine automated and human checks
A clean automated report does not check spelling, speaker identification, factual accuracy, sound-effect descriptions or the completeness of the transcript. Those checks require listening and editorial judgment. Review the beginning, end and several difficult passages even if the numeric checks show no warnings.
Save an edited caption file and a current audit after finishing. The summary above the editor describes the original import, while the edit audit reflects the current cue list. This distinction helps a reviewer compare the source with the final result. Keep a recoverable original and test one final playback before publishing the captions with the media.