A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: "Search Sixteen Years Of API Research From One Call," Kin Lane, October 30, 2026, https://apievangelist.com/2026/10/30/search-sixteen-years-of-api-research-from-one-call/
Marisol had been given the deprecation policy the way these things usually arrive: a one-line message from her manager and a Friday deadline. She was a platform engineer at a company that made routing software, and she had never written a policy in her life, only the code that policies complained about.
She read the post at her desk with a blank document open in the next window. The pitch was simple. Search one word across everything, and see what already exists before writing anything. She was skeptical, because she had been burned by search boxes that returned a thousand things and meant none of them. But the number caught her. The search covered 17,140 records, and the one word deprecation narrowed that to 103. She liked that the response counted results by type before you read any of them. Sixty-eight were posts. Only two were rules.
That second number bothered her a little. Sixteen years of writing, and two things you could actually run against a contract. She wondered whether that meant the field talked about deprecation far more than it enforced it. The post didn't say, and she didn't want to claim it said more than it did. Still, it matched her own company, where plenty had been written down and almost nothing was checked.
She followed the walk the post laid out. The governance filter gave thirteen records, and among them was a guidance entry, a handful of policies, and one of the two rules, called "OpenAPI Operation Deprecated Sunset Info". She read that name twice. A lint rule meant her policy could end in something a pipeline would refuse, and not just a paragraph people nodded at in a meeting. That was the first moment the task stopped feeling like homework.
Then she hit the part she trusted most, which was the author admitting the tool was weaker than it sounded. Searching tools, solutions, standards, papers, conversations and videos together had returned eight results, and some of them were "much looser matches". She had seen that before, a confident list padded with near-misses. The post's advice was to "read the results rather than trusting the count", and she took it as a warning aimed at her, because she was exactly the kind of person who would paste ten links into a document and call it research.
She stayed unconvinced about one thing. A better first draft than nothing was not the same as a good policy, and the post said as much, with "a better first draft than a blank page" being the whole claim. Nobody would be fired for ignoring a search result, and nobody would be saved by it either. The judgment about what her company's customers could stand, how many months of warning, whether anyone would really read a header, was still hers.
She didn't run the curl command. She opened the tool in the client her team already used and tried the narrower search, deprecation limited to rules, policies and guidance. Then she opened the document that had been blank all morning and typed three headings: what we promise, what we check, what we owe people when we break it. Under the second one she pasted the name of the lint rule and the name of the sunset policy, and beside each she wrote a question for herself. Does it fit us? Then she searched the word sunset, because the post had said zeros only meant that one word hadn't matched.