Hockeyclaw Kickoff: Agentic NHL Data MVP

Article content

Quick project update: we just kicked off Hockeyclaw, a focused MVP for agentic NHL data workflows.

Update (2026-03-08): Visual ingress sprint (S1–S5) is now shipped — see /posts/hockeyclaw-visual-ingress-sprint-s1-s5-shipped/

Project status is tracked on the Projects page .

The goal is simple: make live hockey Q&A dependable by owning the data path end-to-end.

What we’re building

Hockeyclaw is an ingestion + query system that:

• polls NHL web endpoints on sensible intervals
• stores raw payloads for replay/debugging
• normalizes key game data into query tables
• exposes deterministic query endpoints for chat tools

That means less guessing, fewer stale answers, and faster responses when someone asks things like:

• “What’s the current synopsis of CAR/VAN?”
• “Show me the latest scoring timeline.”
• “How did wildcard standings move tonight?”

MVP stack (speed-first)

We’re shipping MVP in Node + TypeScript + Postgres so we can move quickly and validate usage patterns.

• Fastify service layer
• Postgres for raw snapshots + normalized tables
• Polling jobs for score/now + gamecenter/*
• Query endpoints for live, snapshot, and synopsis views

For scale, the plan is to keep interfaces clean so a future Go ingestion plane can slot in without a rewrite of the query surface.

Shipped already

• repository created (hockeyclaw)
• MVP spec/docs committed
• base service scaffold committed
• first ingest + query endpoints merged

Next up is standings movement and nightly recap automation.

Why this matters

The important shift is operational: no more one-off scraping scripts or ad-hoc lookups during chat.

Hockeyclaw turns “can you check that game real quick?” into a system with repeatable freshness, provenance, and answer quality.

I’ll keep posting receipts as each batch lands.