A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: "Telecom Network APIs: The Shape Of The Answer," Kin Lane, October 27, 2026, https://apievangelist.com/2026/10/27/telecom-the-shape-of-the-answer/
Priya worked in fraud operations at a mid-sized bank, and she read the post on a Tuesday with a cold coffee and a half-finished ticket open in the next window. The ticket was about whether a transfer above a certain size should trigger a SIM swap check before release. Her team had been arguing about it for two weeks, mostly about the window.
The line that caught her was the one about the check request. It carries a phoneNumber and a maxAge in hours, "an integer from 1 to 2400 with a default of 240." She had assumed the window was something her team would invent. It turned out to be a number someone had already bounded. That should have comforted her, and partly it did.
Then she read the part about replay, and her comfort turned into something closer to unease. The post says that if a model decides a transfer warrants a check with a maxAge of 72, a bank needs to show a regulator or a customer why, and "get the same decision when it replays the same input next month." She thought about the last audit. Someone had asked her to reproduce a decision from August, and she had spent a day assembling logs that only approximately agreed with each other. Nobody had ever promised her sameness. She had only ever promised an explanation.
She did not agree with all of it. The post is confident that a constrained decoder makes certain mistakes impossible, and she wanted to see that on her own data before she believed it. A model that can only emit a valid boolean can still emit the wrong one. Valid and correct were different problems, and the post, to its credit, seemed to know that. Its argument was that the hard problem moves to agreeing on the shape, not that the shape solves everything.
She also bristled a little at the section on operators. She knew which carriers her bank depended on, and the post said the operators "sit lower" in the cohort and publish the least around their contracts. It was uncomfortable to read that about the companies whose answers her fraud rules would eventually wait on. She could not tell from the post whether hers were among the ones with a published SIM swap contract, and the post did not claim to say. She noted that as her own question, not something to blame on the author.
What she did next was small. She closed the post, went back to the ticket, and stopped arguing about the window. Instead she added one comment asking the team to find out whether each carrier they used capped maxAge below 2400, since the post says an operator may do that and answer with a 400 OUT_OF_RANGE when a request goes past it. If the cap differed by carrier, the window was not really a policy choice yet. It was a per-carrier constraint, and they should write it down that way.
She assigned the comment to herself, set a reminder for Thursday, and finally drank the coffee, which was terrible.