Guarding conditional interview claims in an AI-edited recap

An automatic recap can retain the source words and still alter the claim. Imagine an interviewee says, “We can deliver by Wednesday, provided we get the footage Monday and lock the voice-over.” The edited excerpt keeps Wednesday and omits both dependencies. To an audience, that change matters more than a frame-accurate cut.

For a high-value claim, mark its required conditions as separate intervals. Here, the measurements are in source time, and the retained segments come from the edit's mapping back to that same source. Adjacent and overlapping segments can satisfy a condition together. The synthetic example deliberately omits the qualifier in the first edit.

def fully_covered(interval, retained):
    start, end = interval
    cursor = start
    for left, right in sorted(retained):
        if right <= cursor:
            continue
        if left > cursor:
            return False
        cursor = right
        if cursor >= end:
            return True
    return False

conditions = {
    "Monday footage": (5.1, 6.0),
    "voice-over locked": (6.0, 7.2),
}
edits = {"A": [(3.0, 5.0)], "B": [(3.0, 6.0), (6.0, 7.2)]}
for label, retained in edits.items():
    print(label, [name for name, interval in conditions.items()
                  if not fully_covered(interval, retained)])

The first list contains both conditions. The second list is empty. Interval coverage is only a review signal. It says nothing about whether a narrator changed the wording, the sound is intelligible, or the captions stay long enough to read. Play the actual export and ask a viewer what would have to happen for Wednesday to be realistic.

AI generated the draft and teaching diagram. The interview and all time ranges are fictional.