Methodology
PG Finds began as a 2024 pilot and has expanded to resources discovered in discussions from 2023 through 2026. It is a curated, model-assisted selection, not a complete map of PostgreSQL knowledge.
Read the story behind the project: From Hacker News threads to PG Finds ↗.
The process
We use the open-index/hacker-news dataset, pinned to revision c54ced5973b95d80cd934b8921f44ec8d87c3cd1, to find PostgreSQL discussions and links. The ingested files contain 4,587,333 records for 2023, 3,734,176 for 2024, 3,886,492 for 2025, and 2,943,118 for partial 2026: 15,151,119 records in total. File coverage does not establish completeness of the source mirror.
The 2026 dataset used here ends on August 21 at 23:59:48 UTC, partway through August. A separate Algolia Hacker News search supplements it from August 22 through November 7, 2026 at 22:51:07 UTC (exclusive). It yielded 2,270 filtered keyword or domain matches; this supplement is not a complete raw archive.
The expansion screens ranked shortlists: up to 300 resources per added archive year and 200 from the recent supplement, or 927 distinct resources after deduplication. We checked 320 source URLs and submitted 273 complete extracted texts for review. This is a bounded editorial pass, not a review of every link in the archive.
We look for concrete PostgreSQL techniques, explanations, research, or useful tools. Thin announcements, generic marketing, duplicates, inaccessible pages, and sources needing further verification are held back. Newer material receives priority, while useful older resources remain eligible.
Each resource links to its original page and the HN discussions where it appeared. A separate model-assisted audit checks a sample of selections and exclusions, plus dated project-status evidence. The selection is not independent technical validation or a completed human editorial review.
An HN mention is a discovery date, not necessarily a publication date. Where a reported publication date conflicts with that history, the catalogue flags it instead of treating it as established fact. Link-check and project-status dates are shown separately; pages may have changed since.
Project status notes are dated snapshots based on first-party sources such as repositories, documentation, and release notes. Activity or a release does not establish production suitability. Check current documentation and suitability for your own environment.
The selection is not comprehensive, including for bugs and security. A working URL does not establish that a project is maintained or production-ready.
Thanks to Hacker News and its contributors for the discussions. This independent project is not affiliated with Hacker News. The source dataset is distributed under ODC Attribution License 1.0; see its dataset card for details. The catalogue links to original resources and does not republish full articles.
The original pilot’s mining CLI made no paid LLM API calls. The expansion used DeepSeek Flash: logged calls cost approximately $0.49 at the applicable off-peak rates. This is a partial API estimate: interrupted requests without returned session IDs and Luna/Terra session usage are excluded.