How it's
made

Every asset — still or video frame — passes four gates before it goes anywhere near Instagram. They exist because the first few weeks found four different ways for a good-looking reel to be wrong.

Text. Every word on screen is read back by a machine and compared to the brief character by character. Any mismatch fails. This gate exists because of a number, and the number has its own page below.

Aspect and safe area. 1080 by 1920, and every line of type inside the band Instagram doesn’t cover with its own chrome. A headline that runs to the frame edge loses its last letter; that happened, and the field name in the closing card came out one letter short.

Kit. The Eagle has to match the Eagle from last week — beak, head, jersey, feet. Any drift fails, and it fails before the motion render, because motion is the expensive part.

The share test. Would a BA parent send this to another BA parent? Would someone from the other school send it to a friend? Both have to be yes. It’s the one check that predicts whether a reel travels.

The Wednesday reel for St. Joseph’s Prep was about the 690 miles between Philadelphia and Brentwood. The model rendered 600.

It went into a locked cover and a finished master. It got past three separate human reviews. Every note in the folder said 690. Nobody caught it because the eye reads a plausible number as the number it expects, and 600 is plausible.

The fix wasn’t a better reviewer. It was taking the words away from the model entirely. Every headline, date, matchup and score is now rendered locally as a card — deterministic, safe-area checked, versioned, free to redo. The model makes the picture. Nothing it produces is trusted to spell.

The one exception is a brush-lettered display graphic that gets generated once, checked once, and then placed the same way forever. A one-time gated event is a different risk from a per-render gamble.

The wrong master is in the cutting room, with the wrong number on it. It’s the most useful artifact the series has produced.

The Eagle is a character. He wears 26 because it’s the season. He has taped forearms, bare talons, no helmet, and a closed beak. That description is the easy part.

The hard part is that every generation is a re-roll. Ask a model for the Eagle twice and you get two eagles who are cousins. Ask for him in motion and he changes mid-shot. Every still made before the end of August had him in white pants; the real uniform is light grey. Nobody noticed for two weeks because each still, on its own, looked fine.

So there’s a canon now — one approved image, attached as the reference to every future generation, and a written kit spec down to the seam stripe. And the motion rule changed: animate a still you’ve already approved, rather than generating motion from text. It’s cheaper, but that’s a side effect. The point is that an approved still can’t drift.

The opponent mascots get the same treatment. The St. Joseph’s hawk has rust-red tail feathers because without them, at thumbnail size, he reads as a second eagle in a dark jersey. That’s not a style choice. That’s the thing that makes it a rivalry.

Each week starts with research on the opponent — records, history between the schools, what’s at stake. That research is done by one process and then checked by a completely separate one that is never shown where the first one looked. It gets the claim as a sentence and has to find its own support. If the two agree by different routes, that’s evidence. If they agree because they read the same page, it isn’t.

This sounds paranoid until it catches something. In the first real run, a real quote from a real coach arrived wrapped in a context he never said it in — true words, invented frame. A checker reading the same source would have waved it through. A checker working blind noticed the frame wasn’t there.

Every number and date in a brief cites a claim, and a script fails the week if any cited claim didn’t survive. Then a final pass argues against the recommended concepts — is this a repeat of last week, would a parent actually send it, is the opening frame interesting with no title on it — and writes a memo.

Nothing in that memo is a decision. The memo recommends. A person locks the week, every week. The whole system exists to make that person’s Sunday shorter, not to replace it.

Instagram’s Story editor shows you the whole canvas. The live Story doesn’t — the bottom of it is covered by the reply bar, and the top by the username strip. Anything you place there looks perfect in the editor and vanishes when it posts.

The first Monday draft put the poll sticker’s bottom option under the reply bar, and the series badge entirely inside it. Both read fine while editing.

The fix is a quick field test: in the editor, the “Add a caption…” prompt sits roughly where the reply bar will land. If a sticker touches that line, it’s in the danger zone. Now every Story draft is screenshotted and measured against it before it goes out, and the drafts that got it wrong are kept alongside the ones that got it right — a fix nobody photographed is a fix nobody can check next week.