A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: "Advertising: The Shape Of The Answer," Kin Lane, October 28, 2026, https://apievangelist.com/2026/10/28/advertising-the-shape-of-the-answer/
Marisol works in ad operations at a mid-sized agency. She read the post on a Wednesday afternoon, between a billing reconciliation and a call she was dreading. She had expected a piece about models and JSON. She hadn't expected it to open with her own industry.
The part about the planner's brief felt familiar. A planner writes something like CTV, US only, 30 second video, no gambling adjacency, private deal with this publisher. Then someone like Marisol turns it into fields a platform will accept. She had done that by hand more times than she could count. The post says the machine needs the impression, the video constraints, the blocked categories, the deal id and the floor. She agreed with all of it, and the agreement made her a little uneasy. Her job was starting to look like a shape problem too.
She also bristled. The post treats a model sitting in the bidding path as a given, and she wasn't sure she wanted one there. When it argues that a buyer asking why an impression was classified as news and not gambling can't be answered with "the model said so that time", she felt the point land. But she thought the opposite remedy was just as plausible. You could keep the model out of the path and let people write the rules. The post doesn't dwell on that option. It treats determinism as the fix, and she wasn't convinced it was the only one.
What held her attention was the section on what the catalog holds. OpenRTB is tagged by 14 providers, and only five of them sit in the cohort the author tracks. The post says the TripleLift profile records buyers posting bid requests to a single endpoint, with "no JSON Schema for it" covering the supported subset, which is documented as prose tables. Marisol had spent a week last spring reading exactly that kind of table, matching a column of field names against what her team's buying platform actually sent. She hadn't known it was a pattern. She had assumed her exchange was just sloppy.
Then she reached the line she copied into her notes: "Everyone implements a subset of OpenRTB and describes the subset in their own words". She thought about the two exchanges that disagreed on how a deal floor was expressed. Both had been right by their own documents.
The last paragraph said that "Advertising did the hard part years ago when it agreed on the bid request." She didn't find that comforting. It sounded to her like a compliment with a bill attached. If the agreement is real but unpublished, then every team like hers is quietly rebuilding it from tables.
She didn't change anything big. She closed the post, opened the spreadsheet where her team tracks each exchange integration, and added a column. For each exchange it would say one of two things: whether they publish their OpenRTB subset as a machine-readable schema, or whether it lives only in prose. She filled in the first row, the exchange she knew best, and wrote prose. The other rows stayed empty until she could check each one properly. Then she went to her call.