QC captions on a phone, not a monitor

QC captions on a phone, not a monitor. Contrast, characters per second, and safe area only show up at 9:16 in a hand. A desktop timeline is too big, too loud, and too empty of chrome. It will approve a miss the feed will swipe. Export. Hold the phone. Mute. Then ship.

A monitor is the wrong instrument

The timeline is 16:9 or a tall preview on a 27-inch panel. The feed is a rectangle in a palm, often outdoors, often compressed, often with a thumb over the lower third. Those are different contrast problems, different reading distances, and different furniture.

What the monitor hides:

Contrast. White type on a pale wall looks edged on a dim-calibrated display and disappears on a sunny phone. Local ratio is the worst frame behind the glyphs, not the histogram of the sequence. Outline or a backdrop. WCAG contrast for burned-in type is the rule. The phone is the meter.
CPS. Speech is faster than comfortable reading. A two-line block you can finish at desk distance with sound on is a flash at arm’s length with sound off. Reading speed is characters versus hold. You do not need a Versely CPS spec. You need to fail or pass the line in your hand.
Safe area. The editor does not draw TikTok’s username, Reels’ CTA, or YouTube Shorts’ 27 May 2026 AI overlay. Safe area is a snapshot of chrome. Recheck in the app. Bottom ~200px on 1080×1920 is a starting keep-out, not a contract.
Scale. Karaoke pop that looks like a polite accent on a monitor is a token that hits a chin on a phone. Snap scales. Scale plus a low baseline is a collision.
Compression. Social re-encodes. Thin strokes die. Hairline outlines that passed in Premiere are gone in the For You page. Check the upload, not only the master.

Sound-on at the desk is the other lie. You follow the mouth and forgive a late cue. Mute autoplay does not. The first impression is a silent frame. If the claim is only in the speakers under the monitor, the claim is not in the view.

Studio CSS on a webpage is a third lie. Live type on a black card is not this plate, not this chrome, not this size. A caption preview is not the burn. The phone pass is on the MP4.

What to look for in one mute hold

Export the burned file. Transfer it to the phone. Open it in the destination’s draft player if you have it — TikTok, Reels, Shorts — not only in Photos. Full screen. Brightness like a commute, not like a colour bay. Sound off.

Pass or fail:

First two seconds. Can you read the hook without unlocking audio. If not, the open is a miss, whatever the retention graph will say later.
Every cue’s worst frame. Pause. Specular flash, hand, window, pack label. If the line goes, change treatment. If it covers the SKU, move it.
Finishability. If you cannot finish a line before it is replaced, split, cut, or hold. Do not pick a punchier preset.
Chrome. Username, caption field, CTA, right rail, Shorts badge. If type sits in furniture, it was never two lines. It was two lines plus UI.
Double type. Platform auto-captions on top of your burn. Compact if you must live with it. You cannot disable theirs for them.
Freeze-frame (paid). Pause three random frames. Each should still be a sentence, not a half-lit karaoke token.

Do this in portrait, in your actual hand. A phone on a tripod next to the monitor is still a small monitor. The grip covers the lower third the way a scroller’s thumb does. That is part of the test.

Landscape ads still get a phone pass if Demand Gen can serve a vertical tile. Crop, then QC the crop. A 16:9 desk approval is not a 9:16 approval.

Desk tools that are allowed — and what they are not

Allowed before the phone:

Transcript read for names, SKUs, prices against the PDP
Split of 90-character cues into two lines
A short style sample on this clip for look, not for CPS
Waveform check for a uniform offset

Not a substitute for the phone:

Timeline playback at 50% size with sound
Caption Studio CSS
A VEED picker thumb on a black card
“It looked fine on Slack” from a laptop screenshot
A 4K reference monitor. Worse, not better: more pixels, still no chrome, still too far from a palm

BASIC 1× and DYNAMIC 2× will both fail in a hand if the plate or the hold is wrong. 165 language codes will not save a German line that is unreadable under a shop tab. 45 CSS presets will not save Classic on a white pouch until you add an edge and look at it in daylight.

The AI caption generator and add captions to a video produce the file. They do not hold the phone. You do. Burned-in captions cannot be restyled by the viewer. The hand pass is the last chance.

If a cue fails, fix the cause:

Contrast → outline or backdrop
CPS → split, cut, hold
Chrome → move up or top variant
SKU cover → move type, do not crop the label
Late karaoke → recaption the ship mix, do not recolour

Then export again. Then hold the phone again. A desk “fix” you did not re-check is a new miss.

Make the phone pass a gate

Write it into the series bible so episode fourteen cannot skip it: no upload without a mute phone hold in the destination app. Guest editors will try. The monitor is faster. The monitor is wrong.

Cadence math still applies. Thirty clips on Glass are sixty base caption renders. Thirty phone passes are thirty holds. The holds are cheaper than thirty ads whose offer lives under a button. What captions cost across a month of clips is the meter. The phone is the QC.

Native-audio generates, lipsync takes, motion-control outputs — all of them get the same gate. The mix changed; the palm did not. Caption after those passes, then QC in hand.

QC captions on a phone, not a monitor. Contrast, CPS, safe area. Mute. 9:16. Then ship.

Hold the next burn in your hand, then finish it in Versely.

FAQ

Can I QC on a phone but in Photos instead of TikTok?

Photos lacks the chrome. You will pass a bottom block the username will eat. Use the destination draft player when you can. Photos is a contrast check only.

Does a 4K monitor catch more caption errors?

It catches different ones: banding, encode. It will not catch CPS at palm distance or TikTok furniture. Do the 4K pass if you need it. Do not skip the phone.

How long should the phone pass take?

One muted playback plus pauses on the worst frame of each cue. For a 20-second Reel that is a minute. For a 12-minute tutorial, sample the open, a dense middle, and the close, then fix those, then spot the rest. Still in a hand.