How the Heat Board works

Every rank on this site is computed from public evidence. Nothing on the Heat Board is sponsored, boosted, or for sale — and it never will be.

The Heat score (0–100)

Heat measures live attention: a recency-weighted sum of every news mention of a tool across our sources over the last 7 days. Each mention contributes:

mention score = source weight
              × engagement boost   (1 + log10(1 + points) / 2)
              × recency decay      (half-life: 3 days)
              × story quality      (1 + 0.1 × cluster score)
              × syndication cap    (max 3 full-weight items per story, rest 0.2×)

The per-tool total is then multiplied by a diversity bonus (+15% per extra source kind covering it — a story on HN and Reddit and YouTube is categorically hotter than ten posts in one feed) and normalized 0–100 against the hottest tool in the capture. Snapshots are taken every 30 minutes and kept forever — the sparklines and % changes you see are real history, not estimates.

Source authority weights

Official lab blogs (OpenAI, DeepMind, HF, Google/Microsoft Research, BAIR)	2.5×
Hacker News	2.0×
Techmeme	1.8×
Major outlets (The Verge, TechCrunch, VentureBeat, Wired, Ars Technica) · YouTube · X	1.5×
GitHub trending · Product Hunt	1.3×
Reddit · Bluesky	1.2×
Generic RSS · arXiv	1.0×

Engagement (HN points, Reddit score, Bluesky likes, GitHub stars) is log-scaled so a single viral post can't dominate a ranking.

The 🚀 Trending score

Heat favors big names; Trending finds breakouts. It is a Poisson-style z-score of the last 48 hours of mentions against that tool's own 30-day baseline — so a newcomer going from 0 to 30 mentions outranks a giant idling at 100. A 🚀 badge appears at z ≥ 2 (“unusually hot right now”).

Every tool page has a “Why is this hot” panel listing the exact mentions that drove its score, with their weights. If you can't see the evidence, the rank shouldn't exist.

Adoption metrics

Alongside attention we track objective usage from public registries — Hugging Face downloads & likes, GitHub stars, npm and PyPI weekly downloads, Wikipedia pageviews — captured every 6 hours. Attention and adoption are shown side by side, never blended into one opaque number.

The pledge

Ranking is never for sale. No paid placement, no “featured” slots inside the ranked table, no pay-to-update listings. If sponsorship ever appears on this site, it will be visually separate and labeled — and it will never touch a rank.

Some “Visit site” buttons may use disclosed affiliate links (marked “affiliate”). They fund the portal and have zero influence on Heat scores, ranks, or which tools are listed.

Found a bug in the methodology or a tool we're missing? The catalog is auto-discovered from the news daily and human-reviewed — corrections welcome.

← Back to the Heat Board