Search lowers the bar
Search now keeps posts that only partly touch your topic instead of demanding a confident match. Expect more results, and a few looser ones.

You type AI coding into the search box and press return. Instead of a tight handful of posts that answer the question head-on, you get the conversation around it. You see someone's week with a coding agent, a skeptic's opinion, a joke about token bills. The closest matches still come first.
Credit paulinevos, who lowered the bar the filter uses from 50 percent to 30 percent and rewrote what it asks the model to keep. The filter now asks the model to keep related discussions, personal experiences, opinions, jokes and livestream updates, even when they cover only part of the topic or use different words. The nearest posts are scored one by one in a single pass, and anything at 30 percent or above stays. Anything at 29.9 percent is gone. The search dialog asks for five posts, the full page twenty, and you can request up to a hundred.
One catch. The judging happens after the nearest posts are picked, so the lower bar only keeps more of that first pool. It never reaches for a post sitting just outside it. Ask for twenty and you can still walk away with fewer, with no second page to make up the difference.
Key files · #370
- app/
Services/ OpenAI.php - app/
Queries/ CommitSearchQuery.php - app/
Http/ CommitSearchController.phpControllers/ - app/
Http/ ListCommitSearchRequest.phpRequests/ - tests/
Unit/ CommitSearchQueryTest.phpQueries/ - resources/
js/ commit-search.tslib/ - resources/
js/ search-dialog.tsxcomponents/ - resources/
js/ index.tsxpages/ search/