How NVSBL works.
Three parts: ingestion, classification, delivery. Set them up once; they run forever.
1. Ingestion
We poll the official Hacker News API every five minutes. Front page, Show HN, Ask HN, Tell HN, and the comment threads underneath them. New items enter a queue with their full original text, metadata, and source URL. We respect HN's rate limits and use no scraping.
2. Classification
Each item is classified by an agentic pipeline that does three things in sequence: noise filtering (does this match your category description at all?), category labeling (feature request, bug, competitor mention, churn risk, raw pain, opportunity, support gap), and confidence scoring. Items below the confidence threshold you set are dropped. Items above it move to delivery.
The classifier is initialized with a category model we've tuned across thousands of HN signals. On the Scale plan, we tune a custom classifier to your domain using examples you provide and corrections you make as the system runs.
3. Delivery
Classified signals are routed to Linear. The issue includes the original quote, the HN source URL, the classification, the confidence score, and a short rationale. The receiving system has everything it needs to triage without coming back to NVSBL.
What you do
You set the category description, the filters, and the Linear destination. Everything else runs without you. Most teams set up NVSBL once, tune it twice in the first week, and then forget it exists — which is the point.
What's next
When new sources or destinations ship — by user vote — they slot into the same pipeline shape. Same classifier. Same delivery contract. You add them to your configuration; nothing else changes.