# Review — A-final 3 fatal · 4 major · 7 minor · 2 cosmetic · 15 measured but NOT judged engines: {"measure": {"cuts": 13, "shots": 10, "findings": 27}, "say": {"scored": 14, "worst": [["ultralight", 0.18], ["literally", 0.5], ["stripping", 0.67], ["roughing", 0.67], ["explains", 0.71]]}, "caption": {"events": 34, "ocr_calls": 103, "coverage": 0.874, "errors": 0, "sync": {"median_lag_s": -0.13, "events": 31, "p90_abs": 0.25}}, "overlay": {"boxes": 14, "undeclared": 0, "expected": 4}, "fit": {"devices": 2, "pacing": {"devices": 2, "per_minute": 2.4, "coverage": 0.093}, "major": 0}, "pair": {"body_overlays": 1, "not_hers": 0}, "click": {"clicks": 0, "discontinuities": 0, "seams_hit": 0, "elsewhere": 0}} | At | Kind | Severity | What | |---|---|---|---| | 27.92s | identity | fatal | the face in the shot from 27.92s to 29.96s does not match the rest of the ad: SFace cosine 0.702 against the other shots (every labelled control sits above 0.82) | | 30.4s | identity | fatal | the face in the shot from 30.4s to 35.36s does not match the rest of the ad: SFace cosine 0.69 against the other shots (every labelled control sits above 0.82) | | 35.36s | identity | fatal | the face in the shot from 35.36s to 39.28s does not match the rest of the ad: SFace cosine 0.698 against the other shots (every labelled control sits above 0.82) | | 15.4s | editing | major | the picture changes inside the spoken word "stripping" (15.36-15.74 s). The track runs on through it, so the line is whole and the EDIT lands mid-word, which reads as clumsy rather than broken. | | 19.12s | editing | major | the picture changes inside the spoken word "wet," (19.08-19.46 s). The track runs on through it, so the line is whole and the EDIT lands mid-word, which reads as clumsy rather than broken. | | 19.88s | editing | major | the picture changes inside the spoken word "grabby," (19.76-20.12 s). The track runs on through it, so the line is whole and the EDIT lands mid-word, which reads as clumsy rather than broken. | | 30.4s | editing | major | the picture changes inside the spoken word "let" (30.34-30.48 s). The track runs on through it, so the line is whole and the EDIT lands mid-word, which reads as clumsy rather than broken. | | 4.32s | editing | minor | the cut lands 100 ms after the word "it." ends, so the line has no tail before the shot changes | | 15.4s | editing | minor | the cut lands 40 ms after the word "are" ends, so the line has no tail before the shot changes | | 19.12s | editing | minor | the cut lands 40 ms after the word "causing" ends, so the line has no tail before the shot changes | | 21.52s | editing | minor | the cut lands 0 ms after the word "you" ends, so the line has no tail before the shot changes | | 30.4s | editing | minor | the cut lands 60 ms after the word "So" ends, so the line has no tail before the shot changes | | 35.36s | editing | minor | the cut lands 0 ms after the word "And" ends, so the line has no tail before the shot changes | | 39.28s | editing | minor | the cut lands 80 ms after the word "through." ends, so the line has no tail before the shot changes | | 0s | lipsync | cosmetic | sync in window: MFCC best lag -40 ms (peak R2 0.218, at zero 0.209); loudness correlation 0.174 at zero lag, best 0.223 at -320 ms, 1 frames loud-with-closed-mouth, 77 open-over-silence, NOT judged (the donor reads 0.087 / 26 / 1); LipLeak 0.583 | | 27.92s | editing | cosmetic | the cut sits in a silent pause that the transcript spans as the word "Just" (27.58-28.08 s): a cut in a gap, not in a word | | 27.88s | skin-tone | unknown | BODY OVERLAY skin a*b* (137.8, 140.6) vs the presenter's cheek (143.5, 149.0): distance 10.1 (a*-5.6, b*-8.4), reported only until an accepted pair is measured | | 0s | cadence | unknown | CADENCE, for the eye only: the first 10 s carry 1 visible change(s), one every 5.0 s; the general lean in short-form is one every 1-2 s. The whole ad runs one every 3.5 s. A yapper or an older audience wants longer holds, so this is case by case and never a defect. | | 0.0s | skin-tone | unknown | skin NOT JUDGED: the shot from 0.0s to 4.32s sits a*+2.5 b*+5.3 from the other shots (distance 6.8). The donor reaches 7.1 at its own cuts, so this scale cannot separate a light change from a skin change. A human decides; positive a* and b* together read as warmer and more yellow. | | 0s | caption | unknown | CAPTION SYNC, reported only: median -0.13 s from the first spoken word to the caption appearing over 31 events (90th percentile 0.25 s); the correct cut reads +0.00 / 0.19 and unshifted controls read -0.24 to -0.29, so no verdict is drawn | | 4.32s | audio-seam | unknown | the track steps +14.3 dB here, and one side of the cut is silence, so this is a speaker starting or stopping rather than a seam in the sound | | 10.64s | audio-seam | unknown | the track steps +16.4 dB here, and one side of the cut is silence, so this is a speaker starting or stopping rather than a seam in the sound | | 21.52s | skin-tone | unknown | skin NOT JUDGED: the shot from 21.52s to 27.92s sits a*-0.9 b*-3.1 from the other shots (distance 4.5). The donor reaches 7.1 at its own cuts, so this scale cannot separate a light change from a skin change. A human decides; positive a* and b* together read as warmer and more yellow. | | 22.98s | pronunciation | unknown | REDUB SHORTLIST: "ultralight" is heard by an acoustic model with no vocabulary as "r" (overlap 0.18). Whisper reports the right word because its language model supplies it. The donor control slurs its own words to 0.38, so this is a ranking for a human, not a verdict. | | 27.92s | audio-seam | unknown | the track steps +26.8 dB here, and one side of the cut is silence, so this is a speaker starting or stopping rather than a seam in the sound | | 29.96s | audio-seam | unknown | the track steps -21.3 dB here, and one side of the cut is silence, so this is a speaker starting or stopping rather than a seam in the sound | | 30.4s | skin-tone | unknown | skin NOT JUDGED: the shot from 30.4s to 35.36s sits a*-6.9 b*-7.5 from the other shots (distance 11.1). The donor reaches 7.1 at its own cuts, so this scale cannot separate a light change from a skin change. A human decides; positive a* and b* together read as warmer and more yellow. | | 35.36s | skin-tone | unknown | skin NOT JUDGED: the shot from 35.36s to 39.28s sits a*-6.9 b*-7.5 from the other shots (distance 11.7). The donor reaches 7.1 at its own cuts, so this scale cannot separate a light change from a skin change. A human decides; positive a* and b* together read as warmer and more yellow. | | 37.98s | pronunciation | unknown | REDUB SHORTLIST: "literally" is heard by an acoustic model with no vocabulary as "ter" (overlap 0.5). Whisper reports the right word because its language model supplies it. The donor control slurs its own words to 0.38, so this is a ranking for a human, not a verdict. | | 39.28s | skin-tone | unknown | skin NOT JUDGED: the shot from 39.28s to 44.68s sits a*-1.8 b*-3.0 from the other shots (distance 3.7). The donor reaches 7.1 at its own cuts, so this scale cannot separate a light change from a skin change. A human decides; positive a* and b* together read as warmer and more yellow. | | 44.68s | audio-seam | unknown | the track steps -38.8 dB here, and one side of the cut is silence, so this is a speaker starting or stopping rather than a seam in the sound |