Delivery Specs That Bite: Broadcast, Web, and Social

The edit is locked. The client signed off on the color, the mix is beautiful in the suite, and the whole thing plays back perfectly on the machine that made it. Then it goes to the network's automated QC, and at 4:55 on a Friday you get a one-line rejection: loudness out of tolerance, integrated -16.2 LKFS, spec is -24. The spot does not air that weekend, the trafficking deadline is blown, and the fix is a re-export you could have done in ten minutes if anyone had read the delivery sheet before the edit started.

A master that fails on delivery specs is not a finished master. It is a near-miss that costs you a weekend. The frustrating part is that none of these failures are about talent or taste. They are about a handful of numbers that were knowable on day one, and that get treated as an afterthought because the creative work feels like the real work. The delivery spec is the real work too. It is just the part nobody photographs.

The five numbers that bounce a master

Every delivery spec, no matter how long the PDF is, comes down to five decisions. Get all five right and the file clears. Miss one and the whole thing comes back.

Codec and wrapper. The codec is how the pixels are compressed. The wrapper, or container, is the file that holds them. ProRes 422 HQ inside a QuickTime MOV is a different deliverable than H.264 inside an MP4, even at the same resolution. Broadcast wants a mezzanine codec like ProRes, XDCAM HD, or DNxHD. The web wants a delivery codec like H.264. Hand one the other and it either refuses the file or silently degrades it.

Resolution and raster. Not just how many pixels, but the shape. 1920 by 1080 is landscape HD. 1080 by 1920 is vertical for a phone. 3840 by 2160 is UHD. A vertical Reel raster delivered into a 16:9 broadcast slot is not a small problem you can crop your way out of at the station. It is a bounce.

Frame rate. 23.976, 24, 25, 29.97, 30, 59.94, 60. These are not interchangeable. 25 is a European broadcast cadence. US linear television runs 29.97 or 59.94. Film-origin content is usually 23.976. Deliver a 25 fps master to a US network and it fails a validation check that has nothing to do with how the footage looks.

Color space. Rec. 709 is the standard color space for HD SDR delivery and the safe default for almost everything that is not explicitly HDR. The trap is delivering a master that was graded in a wide-gamut or log space and never correctly converted, so it lands washed out or oversaturated on a 709 pipeline. Color QC catches this, but not always before it airs.

Loudness. The one that hides. Two masters can look identical and measure completely different in perceived loudness. Broadcast, web, and social each want a different loudness target, measured in LKFS or LUFS, and this is the single most common reason a technically clean file gets rejected. It gets its own section below because it deserves it.

Broadcast is the strictest room

If you deliver to a US network or a station in a market like Dallas-Fort Worth, you are delivering into the tightest spec in the business, and the automated QC does not care that the client loved it.

The master is almost never H.264. It is ProRes 422 HQ, XDCAM HD 50, or DNxHD, usually in an MXF OP1a or a QuickTime MOV wrapper. The raster is 1920 by 1080. The frame rate is 29.97, and most US linear delivery is still 1080i59.94, meaning interlaced, which trips up editors who cut everything progressive and never set the field order. Color is Rec. 709, 10-bit, 4:2:2. And the loudness target is -24 LKFS integrated with a tolerance of plus or minus 2 LU, with true peaks held to -2 dBTP.

That -24 number is not a preference. It is federal. The CALM Act made the ATSC A/85 loudness recommendation a legal requirement for US broadcasters, which is why the network's QC bounces an out-of-spec spot without a human ever looking at it. Europe runs a hair different at -23 LUFS under EBU R128, which matters the moment a spot is built for both markets from one master.

Broadcast also asks for things web delivery never does: a slate with the spot's metadata, bars and tone at the head, a specific amount of black before and after, and often a textless tail so the network can localize supers. Leave those off and a strict trafficking house treats the file as incomplete. None of this is hard. All of it has to be on the delivery sheet before the edit, because you cannot retrofit a textless tail from a flattened master.

Web and OTT: looser wrapper, tighter loudness than you think

Delivering to a client's website, a Vimeo or a hosted player, or an over-the-top platform feels forgiving after broadcast, and on the video side it mostly is. H.264 High profile in an MP4 is the standard final codec, ProRes 422 HQ is the standard mezzanine when a platform re-encodes on its end, and 1920 by 1080 or 3840 by 2160 both clear. Frame rate follows the source. Color is Rec. 709. A reasonable bitrate for 1080p H.264 lands somewhere around 8 to 12 Mbps, higher for 4K.

Where web bites is loudness, and it bites in the opposite direction from what people expect. The general web and OTT target sits around -16 LUFS integrated, with true peaks at -1 dBTP. That is roughly 8 LU louder than the broadcast -24. So the editor who nails the broadcast master, then delivers that same file to the client's website, ships something that plays noticeably quiet against every other video on the page. The reverse is worse: a punchy -16 web mix pushed to broadcast is 8 LU hot and bounces on the first QC pass.

The lesson is that -16 and -24 are two different masters, not two exports of one. If a job delivers to both a network and a website, that is two loudness passes, and the mix has to be built knowing both numbers up front. This is why the delivery destinations belong in the scope of work, not in a Friday-afternoon email.

Social does not reject, it normalizes, and that is its own trap

Instagram and YouTube almost never bounce a file for loudness the way a network does. They accept the upload and then normalize it, turning loud masters down toward their own internal target, which for both platforms sits around -14 LUFS. This sounds friendlier than broadcast, and it is, right up until it quietly ruins a mix.

Here is the trap. You deliver a hot, aggressive -9 LUFS master to YouTube because loud feels impactful in the edit suite. YouTube normalizes it down to about -14 on playback, so all that extra loudness is thrown away, but the dynamics you crushed to get there do not come back. The result is a flat, lifeless-sounding video that you loudness-maximized for no reason, because the platform undid the level and kept the damage. Delivering at the platform target from the start protects the mix.

The other social bite is the raster. Instagram is a vertical-first surface: 1080 by 1920 for a 9:16 Reel, 1080 by 1350 for a 4:5 feed post, and it caps at 1080 pixels wide, so a 4K master gets hard-downscaled and a horizontal 1920 by 1080 lands letterboxed in a feed built for portrait. On the codec side social wants H.264 or HEVC in an MP4, so the ProRes master that is perfect for the network is oversized and gets re-encoded or refused at upload. The finishing move for a real campaign is often three or four separate exports from one timeline: the broadcast ProRes, the web H.264, the vertical social cut, each with its own loudness pass. That is a line item, not a favor.

Loudness is the one that hides, so measure it

Resolution and frame rate are visible. You can eyeball a vertical raster in the wrong slot. Loudness is invisible until you meter it, which is exactly why it accounts for so many rejections.

Two terms do the work. Integrated loudness, measured in LKFS on the broadcast side or the equivalent LUFS on the web and streaming side, is the average perceived loudness across the whole piece. It is the number that has to hit -24 for broadcast, roughly -16 for web, and roughly -14 for social. True peak, measured in dBTP, is the loudest instantaneous sample, and it has its own ceiling, -2 dBTP for broadcast and -1 for most web and streaming, so the signal does not clip when it is re-encoded downstream. The two are independent. A mix can sit at a perfect integrated level and still fail on a true-peak overshoot, or vice versa.

The practical move is to put a loudness meter on the master bus and check the integrated value against the destination number before you export, every time. Every serious edit and audio application has one built in. It takes thirty seconds and it catches the failure that would otherwise cost a weekend. The reason it gets skipped is that the mix sounds fine in the room, and sounding fine in the room is not the spec. The spec is a number, and the number is either right or it is not.

Build the delivery spec before you build the edit

The through-line here is timing. Every delivery failure in this post was knowable before a single clip hit the timeline, and every one of them is cheap to prevent and expensive to fix after the fact. A codec mismatch is a re-export. A loudness bounce is a re-export plus a blown trafficking window. A wrong raster on a locked social cut can mean re-framing shots that were never protected for vertical.

So the delivery spec is a preproduction document, not a post-production one. Before the shoot, you should already know every destination the finished piece is going to: which network or station, which web player, which social platforms, and the exact codec, wrapper, resolution, frame rate, color space, and loudness target for each. Those numbers shape decisions upstream, from whether to protect for vertical framing on set to which loudness the mix is built against. Put them in the scope of work so the client is agreeing to real deliverables and not a vague promise of a finished video.

A finished edit that fails delivery is not finished. It is a file that looks done and behaves like a draft. The specs are boring, they are almost never the interesting part of the conversation, and they are the difference between a master that airs and a master that comes back. Read the delivery sheet first. It is the least glamorous thing in the pipeline, and it is the thing that decides whether the work you already did actually counts.