#!/usr/bin/env python3
"""Caption cues for the clip, in SOURCE seconds of vid_v2855698113_2560.mp4.

Every line here was decoded twice by small.en under different settings
(vad/beam) and then read against the battle record on wavewarz.info, which is
where the spellings come from: the songs are "Hypnotic" and "Dolphin Flippin",
the poll went 3-1, and the AI judge is called DJ Wavy. Words the two decodes
disagreed on, or that scored below ~0.5, are marked in the comments — those are
the only places a caption is a judgement call rather than a transcript.
"""

# (start, end, text) — text is what goes on screen, two short lines max
CUES = [
    (0.00,  6.60, "My god, speaking of Dolphin Flippin —\nwe've had a flippening on the charts."),
    (7.06, 13.20, "We had a last-second buy on\nDolphin Flippin to take the dub."),   # "dub": heard as "dubbers"
    (13.24, 14.60, "My god."),
    (14.82, 19.60, "And Dolphin Flippin also got\nthree votes on our polls."),        # "polls": heard as "pulse", 3-1 in the record
    (20.30, 23.50, "And let's see what DJ Wavy\ngot to say."),
    (34.92, 36.10, "That Hypnotic was crazy."),
    (38.00, 43.60, "Track A is a masterclass in sonic\ncohesion and mood."),
    (44.66, 47.00, "While B brings high energy creativity,"),
    (47.76, 51.60, "A's professional mix and focused\narrangement feel more complete."),
    (52.08, 56.50, "A built a war. B threw a party\nin a room."),                     # "built": low confidence
    (57.00, 61.00, "The polish on Hypnotic gives it\nthe definitive edge."),
]

# The edit. The span dropped between 23.52 and 34.92 is an aside that has
# nothing to do with the battle and does not belong on a brand channel.
KEEP = [(0.00, 23.52), (34.92, 62.40)]