OSS Knowledge Base
A daily, evidence-first reader for open-source communities: what happened in Apache Kafka and Apache DataFusion this week, searchable down to the original pull request, Jira issue or mailing-list message. These pages explain how it is built, why, and what it costs.
What it is
The product collects public community activity once a day, turns it into threads, publishes an immutable snapshot to object storage, and serves a small web app on top of it. Every summary and search hit links back to the source record, so a reader can check the claim (see README.md and docs/feature-map.md).
| Fact | Value |
|---|---|
| Communities | Apache Kafka (GitHub, Jira, the dev@ mailing list) and Apache DataFusion (GitHub) |
| Live sites | Prod oss-knowledge-base.pages.dev (release v0.3.1, 2026-10-19) and Dev oss-knowledge-base-dev.pages.dev (every push to main) |
| Runs on | Cloudflare: one scheduled Worker per environment writes to R2 object storage; Pages serves the Vue app and its read-only API |
| Built with | 26 written specifications, 17 of them with machine-checked acceptance items; independent AI verification of every change |
Jargon used throughout: R2 is Cloudflare's object storage (billed per request class, see Process & cost); a Worker is a serverless function; a Durable Object is a Worker with its own strongly consistent state; Pages hosts the static site and its API functions.
Read next
| Page | What you will learn |
|---|---|
| System & data flow | The four stages (collection → processing → storage → application) as a clickable diagram, and what differs between Dev and Prod |
| Search | BM25 ranking with identifier normalisation (KIP770 = KIP-770), the real-corpus evaluation, and the semantic hybrid running on Dev |
| Weekly digest | How a week of activity becomes a bilingual summary: classify, summarise, validate, translate, and the quality gate before Prod |
| Process & cost | Spec-driven development, fail-first tests, CI/CD and releases, the R2 cost incident, budgets and observability |
Status at a glance
As of 2026-10-20 (sources: docs/feature-map.md, docs/operations/release.md):
- On Prod and Dev: daily publication from GitHub, Jira and dev@; lexical search
bm25-reference@3; thread list and detail; the data-freshness indicator. - Dev only: the weekly topic digest, semantic hybrid search, contributor profiles and review ingestion. Each sits behind its own flag and is off on Prod.
- Next for Prod: the digest goes to Prod only after its quality gate passes on a human-labelled week (digest).