A growing subtitle error suggests that one timeline is running at a different scale, but it does not prove a frame-rate problem. Compare verified spoken moments before applying a conversion. A constant offset needs a shift; a uniform speed change may need both scaling and a shift; internal edits can require separate sections instead of one global formula.
Establish which kind of change occurred
The Subtitle Edit frame-rate documentation describes converting timings between rates. Its separate Point Sync guide documents alignment using reference points. Those operations change subtitle timing; merely changing a video export's frame-rate label does not prove its real-time duration changed.
Keep the exact video version that the subtitles describe. Pick identifiable words near the start and end, then a third in the middle. Record the subtitle's current time and the actual word's time in that video. Avoid using a soft breath or a long vowel whose onset is ambiguous. Check that the language track and edit are the intended ones.
Fit a relationship, then try to disprove it
For a truly uniform change, write target time as t_new = a × t_old + b. With two distinct verified anchors, calculate a = (new2 − new1) / (old2 − old1) and then b = new1 − a × old1. Times must use the same unit. This arithmetic is an editorial explanation, not a claim that every drift is linear.
In a fictional example, old times 100 and 500 seconds correspond to 104 and 520 seconds in the new version. The scale is 1.04 and the offset is zero. An independent old time of 300 should therefore correspond to 312. If the actual word is at 318, the single-scale explanation fails and needs investigation.
When the same fixed set of frames really is replayed from exact 25 fps to exact 24 fps, duration scales by 25/24. Ordinary normal-speed conversion can instead preserve duration by changing which frames are displayed. Do not apply the ratio just because file properties list different rates.
Apply only after the third anchor agrees
Work on a copy, convert both start and end of each cue, and inspect rounding, negative starts and overlaps. Recheck first, middle and last passages against actual speech. A change that aligns only the final line can conceal a wrong beginning or a deleted scene.
If the residual error jumps after one cut, investigate that edit boundary. If it varies irregularly, stop applying a global scale and use supported sectional or point-based alignment. Do not force a nonlinear edit to fit a convenient frame-rate story.
For an Imagild film assembled externally, keep the reviewed subtitle file tied to the precise finished video version. Acceptance means independent spoken anchors agree and the cues retain correct text and readable duration. The formula diagnoses a particular timing relationship; it cannot recover missing dialogue or certify synchronization without playback.