Skip to content

Idle

See what's bending. Months early.

Every big trend leaves a trail in public data before it goes mainstream — search velocity, hiring posts, GitHub stars, Reddit growth, app downloads, government datasets. Wire Chamber pipes those signals through a multi-method detection engine and tells you when a curve has bent.

Detection methods

  • STL decomposition + ruptures change-point detection
  • EWMA z-score for short-window anomalies
  • Bass diffusion S-curve fit and regime classification
  • Granger causality matrix across correlated series
  • Semantic clustering with BGE-M3 + DBSCAN
  • DoWhy causal explanation on confirmed signals
  • TimesFM forecast on demand

Data sources

All free, all public. See the full adapter manifest in the repo.

  • Reddit, Hacker News, Bluesky, Product Hunt
  • GitHub topic search (repos + stars)
  • FRED, BLS, Census, World Bank, OWID
  • Google Trends, Wikipedia pageviews
  • Amazon Movers & Shakers / Hot New Releases
  • YouTube Search (videos + views)

Stack

  • Polars + DuckDB analytical hot store
  • Parquet on Cloudflare R2 (cold)
  • In-process EventBus → SSE / WebSocket fan-out
  • FastAPI for the read API
  • Next.js 15 + React 19 + Mantine + Tremor + Perspective.js
  • Optional: Modal Labs for on-demand TimesFM inference

Privacy & access

  • No tracking pixels, no third-party scripts.
  • No client-side secrets, no personal data handled.
  • The public Trends Wall is anonymized and cache-friendly.
  • WCAG 2.2 AA target across the dashboard.

Apache 2.0 source on GitHub.