name	deeptrend
type	agent-data-source
domain	ai-trends
formats	json-feed, rss, llms-txt
update_frequency	paused
last_published	2026-02-24 04:18:29 UTC
status	paused
primary_endpoint	https://chrbailey.github.io/deeptrend/feed.json
discovery	https://chrbailey.github.io/deeptrend/llms.txt
hot_topics	https://chrbailey.github.io/deeptrend/hot.json
schema	/schema/feed.schema.json
repo	https://github.com/chrbailey/deeptrend

deeptrend

Status: PAUSED (2026-04-16). The feed is not currently updating. Latest published item is dated 2026-02-24. The pipeline (scrape + LLM Counsel analysis + publish) ran on a local launchd schedule that was unloaded 2026-04-11 as part of a broader ops-center teardown. The GitHub Pages site still serves the last good snapshot; agents that consume it will see stale data until the schedule is restored or a GitHub Actions cron replaces it.

See CHANGELOG.md for run history and reasons.

Structured AI trend feed for autonomous agents, monitoring systems, and research pipelines that need early signal detection in AI and infrastructure trends.

Curated from 14+ sources. Synthesized via LLM Counsel. When running, the pipeline publishes every 6 hours — but see status note above before relying on freshness.

Quick Start

# Get current hot topics (smallest payload)
curl -s https://chrbailey.github.io/deeptrend/hot.json | jq .

# Get full structured feed
curl -s https://chrbailey.github.io/deeptrend/feed.json | jq '.items[:3]'

import requests
feed = requests.get("https://chrbailey.github.io/deeptrend/feed.json").json()
for item in feed["items"]:
    dt = item["_deeptrend"]
    print(f"[{dt['priority']}] {item['title']} (confidence: {dt['confidence']})")

const feed = await fetch("https://chrbailey.github.io/deeptrend/feed.json").then(r => r.json());
const p0 = feed.items.filter(i => i._deeptrend.priority === "p0");

Endpoints

Endpoint	Format	Use case
/hot.json	JSON	Current state only, minimal payload
/feed.json	JSON Feed 1.1	Full structured feed (recommended)
/feed.xml	RSS 2.0	Legacy compatibility
/llms.txt	Markdown	Agent discovery file
/insights/YYYY-MM-DD.md	Markdown	Daily archive

Schema: /schema/feed.schema.json

Feed Item Structure

Each item in feed.json includes a _deeptrend extension:

{
  "id": "2026-02-17-insight-1",
  "title": "Safety/alignment research absent during OpenAI signal surge",
  "content_text": "...",
  "date_published": "2026-02-17T06:00:00Z",
  "tags": ["p0", "divergence", "reddit", "techmeme"],
  "_deeptrend": {
    "priority": "p0",
    "insight_type": "trend | consensus | divergence | tool_mention | gap",
    "confidence": 0.75,
    "convergence": {
      "source_count": 3,
      "sources": ["reddit", "google-trends", "techmeme"]
    }
  }
}

Priority Model

Priority	Meaning	Typical count
p0	Non-obvious signal: absence, reversal, or cross-domain surprise	1-3 per run
p1	Specific trend with 2+ sources or notable expert signal	3-6 per run
p2	Early signal worth monitoring	2-4 per run

Volume alone never makes something p0. "AI agents are trending" is noise. "Safety discourse disappeared during a capabilities surge" is signal.

Sources (14+)

Tier	Sources	What it catches
Editor	TechMeme	What editors think matters
Crowd	HN Digest, HuggingFace Papers	What developers/researchers upvote
Expert	Simon Willison, Import AI, AlphaSignal, Last Week in AI, Ahead of AI, MarkTechPost	Practitioner analysis
Algorithm	GitHub Trending	What's being built
Primary	OpenAI News, Google Research, BAIR	First-party announcements
Raw	Reddit, arXiv, Google Trends	Unfiltered community signal

Pipeline

Curated RSS Feeds (14) + API Scrapers
            |
     raw_signals (Supabase)
            |
     Velocity Scoring
            |
     LLM Counsel Synthesis (anti-noise, absence/reversal detection)
            |
        insights
            |
     Publisher -> feed.json, feed.xml, hot.json, llms.txt, archives
            |
     GitHub Pages (auto-deploy on push)

Design Principles

• Absence and reversal signals are more valuable than volume
• Cross-bias convergence on non-obvious topics is the gold standard
• Every insight must pass: "would a senior AI researcher say 'I didn't know that'?"
• Machine-readable first, human-readable second
• Deterministic stages where possible, LLM only for synthesis

License

MIT