Trend Physics Pulse

Two AI-ecosystem trend signals you can't get from any single public API, shaped as flat Apify Dataset rows. $0.10 per 1000 rows.

{
  "_slice": "repos",
  "repo": "xai-org/x-algorithm",
  "anomaly_signal": "🔴 Star bomb (100x spike)",
  "composite_score": 1745.91,
  "velocity_per_day_7d": 124.43,
  "total_stars": 5791
}

☝️ One row from one run. Surfacing that without this Actor takes weeks of GitHub crawling plus your own anomaly logic. A default run gives you ~40 rows like it (across accelerating GitHub repos and a scored AI-tool registry) in under 2 seconds.

Two more slices — PyPI download velocity and topic-entropy / narrative-collapse — are in active development. See Roadmap at the bottom.

---

Try it in 30 seconds

Easiest path: click ▶ Try for free above — Apify runs the Actor in your browser with no setup.

For everyone else, grab an API token from console.apify.com → Integrations and run any of these from a terminal:

export APIFY_TOKEN=apify_api_...

1. Today's top accelerating repos

curl -sX POST "https://api.apify.com/v2/acts/brilliant_hemlock~trend-physics-pulse/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"slices":["repos"],"limitPerSlice":50}' \
| jq -r '
    map(select(.composite_score != null))
    | sort_by(-.composite_score) | .[0:5][] |
    "\(.composite_score | floor)\t\(.repo)\t\(.anomaly_signal)"
  '

Expected output (one row = one accelerating repo, sorted by momentum):

1745   xai-org/x-algorithm                    🔴 Star bomb (100x spike)
 817   anthropics/claude-plugins-official    🟢 Normal variance
 115   HKUDS/CLI-Anything                     🟢 Normal variance
 100   microsoft/ai-agents-for-beginners      🟡 Suspicious spike (20x)
  88   langchain-ai/langgraph                 🟢 Normal variance

2. Just the star-bombs — anything growing unnaturally fast today

curl -sX POST "https://api.apify.com/v2/acts/brilliant_hemlock~trend-physics-pulse/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"slices":["repos"],"limitPerSlice":200}' \
| jq '[.[] | select(.anomaly_signal // "" | startswith("🔴"))]
       | .[] | {repo, anomaly_signal, total_stars, velocity_per_day_7d}'

3. Top 5 useful AI-coding tools, formatted for a Slack digest

curl -sX POST "https://api.apify.com/v2/acts/brilliant_hemlock~trend-physics-pulse/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"slices":["tools"],"limitPerSlice":50,"toolVerdict":"useful","toolCategory":"ai_coding"}' \
| jq -r '
    map(select(.avg_score != null))
    | sort_by(-.avg_score) | .[0:5][] |
    "*\(.canonical_name)* (score \(.avg_score)) — \(.one_liner)"
  '

Each call runs the Actor once on Apify's runners and returns the dataset rows in a single HTTP response — no polling, no dataset-URL juggling. Pipe straight into jq, csvkit, a Slack webhook, or a cron job.

---

The two slices

Every row lands in the same Apify Dataset, tagged with _slice for filtering.

repos — accelerating GitHub repos

Backed by a multi-million-repo stargazer corpus with multi-year history. The Actor never calls GitHub itself — it requests precomputed signals from our backend. The repo cohort rotates daily, so each run surfaces a fresh set of accelerators.

{
  "_slice": "repos",
  "repo": "anthropics/claude-plugins-official",
  "snapshot_day": "2026-05-23",
  "total_stars": 5729,
  "velocity_per_day_7d": 88.57,
  "acceleration_per_day_squared": 79.86,
  "anomaly_signal": "🟢 Normal variance",
  "composite_score": 817.38
}

Notable fields:

• composite_score — momentum score that weights acceleration against repo size. A fast-growing small repo can outrank a slow giant. Sort by this for the day's top movers.
• anomaly_signal — 🟢 Normal variance / 🟡 Suspicious spike / 🔴 Star bomb. Derived from each repo's historical star series, not just on-the-day spikes, so an old organic burst is distinguishable from a fresh coordinated one. Exact wording comes from the upstream materialised view — see METHODOLOGY.md for the burst-ratio thresholds.
• velocity_per_day_7d, acceleration_per_day_squared — first- and second-derivative momentum; week-over-week diff is built in.

tools — scored AI-tool registry

Curated registry refreshed by an internal LLM-rubric pipeline on a daily curation cycle. Each tool is scored 1–10 on novelty / utility / maturity / hype-vs-substance, with a written verdict (useful / promising / noise) and justification. Sources include GitHub Trending, HN, Reddit, and a handful of newsletters.

{
  "_slice": "tools",
  "canonical_name": "TanStack Router",
  "category": "ai_infra",
  "verdict": "useful",
  "avg_score": 8.25,
  "novelty": 6, "utility": 9, "maturity": 9, "hype_vs_substance": 9,
  "one_liner": "A fully type-safe router and full-stack framework for building web applications with React and other frameworks.",
  "repo_url": "https://github.com/TanStack/router"
}

---

What you'd use this for

• Daily "what's moving" digest. Sort repos by composite_score for the rotating top movers across the corpus.
• Spot 100x star bombs the day they happen. Filter repos where anomaly_signal starts with 🔴 — typically days before HN or X notice.
• Weekly AI-tool Slack digest. Filter tools by verdict: useful plus a category, push to your channel.
• VC / scout pipeline. Sort repos by composite_score and review small-but-accelerating projects worth a meeting.
• Retire your own crawler. GitHub polling + anomaly re-derivation + LLM tool scoring, replaced with one daily Actor run.

---

Input

{
  "slices": ["repos", "tools"],
  "limitPerSlice": 20,
  "repoMinStars": 1000,
  "toolVerdict": "useful",
  "toolCategory": ""
}

Field	Default	Range / values	Purpose
slices	both	repos / tools	Uncheck to skip a slice and save spend.
limitPerSlice	20	1–200	Default keeps first-time runs cheap; raise to 200 for full daily dumps.
repoMinStars	1000	≥0	Drops low-noise repos.
toolVerdict	useful	any / useful / promising / noise	Filter tools by overall verdict.
toolCategory	—	e.g. llm_framework, ai_coding	Empty = all categories.

---

Pricing & limits

• $0.10 per 1000 rows delivered ($0.0001/row, Pay-Per-Event). A default ~40-row run ≈ $0.004; a maxed-out ~400-row run (limitPerSlice: 200) ≈ $0.04. Set Maximum cost per run in your run options to cap spend.
• 60 requests/minute per Apify customer (you won't hit it in normal use).
• Backend caches each filtered slice for 5 minutes per query.
• Freshness: repos refreshed within 24h (see snapshot_day); tools per daily curation cycle (see scored_at).

What this is not

• Not a real-time API — refresh cadence above.
• Not a historical-query endpoint — time-series queries are planned for a higher-tier SKU.
• Not a GitHub or PyPI proxy — the corpus is curated and the signals are computed, not raw pass-through.

---

Roadmap

Two more slices are wired into the backend but not yet exposed in this Actor's input — data pipeline work is still in progress:

• packages — PyPI download velocity per tracked LLM framework, with rolling average and spike-ratio signals.
• entropy — weekly Shannon-entropy analysis flagging AI subcategories that have concentrated onto a single leader repo.

They'll be added to slices once the underlying collectors are production-ready. METHODOLOGY.md tracks the current exposure status.