Sure, this was the command I used:

    ./yt-dlp --skip-download --write-sub --write-auto-sub --sub-lang en --sub-format json3 "https://www.youtube.com/playlist?list=PL8mG-RkN2uTw7PhlnAr4pZZz2QubIbujH"

That'll grab all the WAN shows (that's the playlist LTT provides for the WAN) of the autogenerated subs in a rather, uh, unique, JSON based format. From there I wrote some throw away quality Python to find razor or blade near percent:

    import os, json

    hits, files = 0, 0
    for cur in [x for x in os.listdir(".") if x.endswith(".json3")]:
        files += 1
        with open(cur) as f:
            data = json.load(f)
        words = []
        for event in data["events"]:
            at = event["tStartMs"]
            for seg in event.get("segs", []):
                words.append((len(words), at, seg["utf8"].strip().lower()))

        span = 10 * 1000
        for i, at, word in words:
            if word in {"razor", "blade", "razors", "blades"}:
                span_min = min(x[0] for x in words if abs(x[1] - at) <= span)
                span_max = max(x[0] for x in words if abs(x[1] - at) <= span)
                for other_i, other_at, other_word in words[span_min:span_max+1]:
                    if other_word in {"percent"}:
                        print("-" * 100)
                        loc = min(at, other_at) // 1000
                        loc = f"{loc//3600}:{(loc%3600)//60:02d}:{loc%60:02d}"
                        print(loc + ": " + cur)
                        phrase = " ".join(x[2] for x in words[min(i,other_i): max(i,other_i)+1])
                        print(phrase)
                        hits += 1

    print(f"{hits} hits in {files} files.")

It found some false positives (maybe one, don't remember), and the hit shared above.