back to high signal
primary-source substrate

Lab

The shared discovery substrate beneath every sub-product. Local-first Postgres index over Hacker News, GitHub trending, engineering blogs, arXiv, and operator forums. Ranked, searchable, and feeding the curated signal layers above.

documents
ingested primary-source docs
sources
feeds, repos, blogs, forums
last ingest
UTC, refreshed by lab worker
vector index
pgvector embeddings
preview only· Lab substrate is local-first; bring it up to see live data

The rows below are a static preview of what /lab shows when the local Postgres substrate is running. The Lab is intentionally local-only — it ranks raw discovery material (Hacker News, GitHub trending, one-hop link extraction) so the operator can turn the top of the ranking into cited signals in the public brief. See python/lab for the runbook.

hn / score 0.86 / cluster ab12cd / 2026-05-24 (preview)

Show HN: a local-first Postgres + pgvector substrate for discovery

Reference example. With Lab running, the top of this feed shows ranked HN items by 4-factor signal score (HN discussion + recency + velocity + GitHub momentum).

hn-linked / score 0.78 / cluster ef34gh / 2026-05-23 (preview)

Article extraction + one-hop link materialisation (Trafilatura)

Each HN submission's linked page becomes a document of its own, with outbound links recorded as `links` rows so subsequent passes can materialise them.

github-trending / score 0.71 / cluster ij56kl / 2026-05-22 (preview)

github.com/trending scraper → repos table (no API key)

Daily / weekly / monthly trending in Python, Rust, TypeScript, Go. Star count feeds the GitHub-momentum factor in the scorer.

one-hop / score 0.64 / cluster mn78op / 2026-05-21 (preview)

Story clustering (union-find over shared link targets + embedding cosine)

Documents that point at the same upstream get the same cluster_id so the feed can collapse near-duplicates into one row.