A sound effect does not begin at its impact.

You know where the hit lands. <Audio> wants to know where the file starts, and they are not the same number. A whoosh carries 180 ms before its transient, so on a hit at frame 90, from={90} puts the impact +180 ms late — a fifth of a second. The number wanted is from={85}, which is hitFrame − round(preRollMs × fps / 1000), and nothing in the API hints at it.

The engine is free and needs no account. Paste the cue sheet — the composition, the files with their pre-roll, the Sequence nesting and the frame each effect lands on. There are two roundings, not one: the hit time quantises to a frame and so does the pre-roll, and they can cancel or add — so the worst case is a whole frame (33.3 ms at 30, 41.7 ms at 24) rather than the 16.7 ms a single rounding would give, against the 20 ms a hard transient allows. A cue quoted in frames has no rounding of its own and halves the error for free. Cutting the head of a file costs level, not time — an <Audio> is anchored to the frame it is placed at, so a Sequence opening late does not delay the sound, it renders it from partway in and the run-up is simply gone. And the same from means different absolute frames in different places, because useCurrentFrame() is relative and nested offsets add. Every lane ships a worked example that costs nothing. Writing the report needs credits.