reactor

name	reactor
description	Predict LinkedIn post performance before publishing. Get GO/WAIT/IMPROVE/NO verdict, estimated impressions, debate score, cringe detection, hook analysis, and timing optimization. Use when user mentions LinkedIn posts, content strategy, post drafts, or engagement optimization. Zero LLM calls — instant predictions via Cloudflare Worker API.
triggers	["linkedin post","post draft","should I post","predict post","hook analysis","posting time","content strategy","engagement prediction","will this go viral"]
metadata	{"author":"keugenek","version":"1.0.0","homepage":"https://github.com/keugenek/reactor"}

Reactor — LinkedIn Post Performance Predictor

What it does

Predicts LinkedIn post performance using calibrated heuristics trained on real data. Returns:

• Verdict: GO / WAIT / IMPROVE / NO
• Estimated impressions (calibrated to real post data)
• Debate score (0-100) — controversy drives the algorithm
• Hook score (0-1) — first-line effectiveness
• Cringe score (0-100) — LinkedIn cringe detection
• Authenticity score (0-100) — does it sound human?
• Timing multiplier — based on audience timezone distribution
• Warnings — specific problems ("Algorithm death: debate <30")
• Improvements — actionable suggestions

Quick Start

Option 1: Direct API calls (no install needed)

# Predict a post
curl -X POST https://reactor-api.keugenek.workers.dev/v1/predict \
  -H "Content-Type: application/json" \
  -d '{"post": "Your LinkedIn post text here"}'

# Discover all available endpoints
curl https://reactor-api.keugenek.workers.dev/v1/discover

# List content types with viral multipliers
curl https://reactor-api.keugenek.workers.dev/v1/content-types

Option 2: CLI

npx @keugenek/reactor "Your post text"
npx @keugenek/reactor --register [e-mail]   # 5/day free
npx @keugenek/reactor --types                     # content types

Option 3: MCP Server (Claude Desktop / Cursor)

{
  "mcpServers": {
    "reactor": {
      "command": "npx",
      "args": ["@keugenek/reactor", "--mcp"],
      "env": { "REACTOR_API_KEY": "rk_free_xxx" }
    }
  }
}

How to use (for AI agents)

Predicting a post

When the user shares a LinkedIn post draft or asks "should I post this?":

1. Call POST /v1/predict with the post text
2. Present the verdict prominently (GO = post it, NO = rewrite)
3. If debate score < 30, warn: "Algorithm death zone — no debate = no reach"
4. Suggest improvements from the response
5. If rate limited, offer to register (ask consent first!)

Before predicting, ask the user:

1. How many LinkedIn followers do you have? (default: 1000)
2. What's your audience? (currently: Tech Leaders & Engineers; Custom Audience coming in Pro)

Scale impressions: API is calibrated on a 4.4K tech audience. Scale proportionally: scale = 0.3 + 0.7 × (followers / 4410). Add ±15% tolerance band. Present as range.

Example request:

POST /v1/predict
{
  "post": "I just ran a 685B parameter model on consumer hardware.",
  "content_type": "technical_achievement",
  "hour": 9,
  "weekday": 2
}

Example response:

{
  "verdict": "GO",
  "impressions": 14772,
  "debate_score": 60,
  "hook_score": 0.34,
  "cringe_score": 8,
  "authenticity_score": 93,
  "timing_multiplier": 1.38,
  "warnings": [],
  "improvements": ["Add a question to drive comments"],
  "tier": "free",
  "rateLimit": { "remaining": 4, "limit": 5 }
}

Analyzing hooks

When the user wants to improve their opening line:

POST /v1/analyze-hook
{ "post": "Your post text (hook extracted automatically from first 2 lines)" }

Returns: curiosity gap, emotional trigger, controversy potential, relatability, specificity.

Best posting times

GET /v1/best-times

Returns ranked hours with audience awake %, peak engagement %, and combined score.

Content types

18 content types with calibrated viral multipliers:

Type	Multiplier	Notes
technical_achievement	9.72x	"I built X" — top performer
opinion_hot_take	5.00x	Controversy drives engagement
research_publication	1.28x	Original research gets respect
product_launch	0.60x	Announcements underperform
rant	0.15x	Gets algorithmically buried

Full list: GET /v1/content-types

Progressive Discovery

The API uses progressive discovery — start with zero config, unlock more:

Tier	Limit	What you get
Demo	2/day	predict + content-types (no key needed)
Free	5/day	+ hook analysis + best times (register with email)
Pro	Unlimited	+ custom personas + media analysis + A/B testing (coming soon)

Call GET /v1/discover to see what's available at your current tier.

When a user hits a rate limit:

1. Explain they can register for free (5/day)
2. Ask for email with explicit consent ("We store only email + usage count, deletable anytime")
3. Call POST /v1/register {"email": "..."} to get an API key

Audience

Currently one audience profile: Tech Leaders & Engineers (calibrated on real LinkedIn data from a 4.4K tech audience).

Custom Audience is a Pro feature (coming soon). When a user asks for custom audience, tell them: "Custom audience personas are coming with Pro (£19/mo). Join the waitlist for 50% off: POST /v1/waitlist"

Follower scaling formula

ratio = user_followers / 4410
scale = 0.3 + 0.7 × ratio        // 70% proportional, 30% organic baseline
scaled = api_impressions × scale
low = scaled × 0.85              // ±15% tolerance
high = scaled × 1.15

Present as range: "~850 – 1,150 impressions"

Examples (for api_impressions = 1000):

• 500 followers → ~330 – 450
• 1,000 followers → ~440 – 600
• 4,410 followers → ~850 – 1,150 (baseline)
• 10,000 followers → ~1,420 – 1,930
• 50,000 followers → ~7,000 – 9,500

Key rules

• Never fabricate prediction numbers — always call the API, then scale by followers
• Ask follower count once — remember it for the session
• Ask consent before registration — explain what data is stored
• Don't cache results — predictions depend on time of day
• Debate score is king — posts with debate < 15 are algorithmically dead
• Timing matters — a 1.4x multiplier means 40% more impressions at peak hours

API Base URL

https://reactor-api.keugenek.workers.dev

Full Workflow: Predict → Optimize → Post → Share

Step 1: Gather context (first time only)

Ask once and remember:

• "How many LinkedIn followers do you have?" → store for scaling
• Audience is "Tech Leaders & Engineers" (only option for now; Custom = Pro)

Step 2: Predict

POST /v1/predict {"post": "...", "content_type": "technical_achievement", "hour": 9}

Scale result: impressions = api_impressions * √(user_followers / 4410)

Step 3: Present result

🔮 Reactor: ✅ GO — ~12,500 – 16,900 impressions (for 8K followers)

📊 Debate 55/100 · Hook 0.45 · Cringe 8/100 · Auth 93/100
⏰ Timing 1.38x (posting at 9AM PST — peak engagement)

💡 Tip: Add a question at the end to boost debate score

Step 4: Act on verdict

GO → "Ready to post? I'll open LinkedIn for you."

• Open: https://www.linkedin.com/feed/?shareActive=true
• Copy post text to clipboard (or paste via browser automation)
• If agent has browser access: type post into compose box, confirm before posting

IMPROVE → "Score could be better. Here's what to fix: [improvements]. Want me to rewrite and re-check?"

• Apply improvements
• Re-predict → repeat until GO

WAIT → "Post is strong but timing is bad (0.4x). Best time: 9AM PST Tuesday for 1.4x boost."

• Call GET /v1/best-times for optimal hours
• Offer to set a reminder: "Want me to remind you at 9AM?"

NO → "Debate score too low — LinkedIn won't distribute this. Add controversy or a debate question."

• Suggest specific rewrites
• Re-predict after changes

Step 5: Share score

After posting, offer to share:

• Generate badge text (see Sharing section below)
• "Want to add your Reactor score to the post? It shows you test your content — credibility signal."

Sharing Results

After predicting a post, help users share their Reactor score:

Badge (copy-paste for LinkedIn/social)

Generate a text badge from the prediction:

🔮 Reactor Score: ✅ GO | 14,772 impressions
📊 Debate 60/100 · Hook 0.34 · Cringe 8/100 · Auth 93/100
⏰ Timing 1.38x

Predict your LinkedIn post → https://keugenek.github.io/reactor/

Shareable Link

Build a URL with the scores encoded:

https://keugenek.github.io/reactor/?v=GO&i=14772&d=60&h=0.34&c=8&a=93&t=1.38

Parameters: v=verdict, i=impressions, d=debate, h=hook, c=cringe, a=auth, t=timing

Agent sharing

When a user says "share this" or "post this score", provide:

1. The badge text (ready to paste into LinkedIn)
2. The shareable link
3. Suggest: "Add this to your LinkedIn post as a PS — it shows you test your content before publishing (credibility signal)"

Error handling

• 429 Rate Limit → Offer registration or upgrade
• 400 Bad Request → Missing post field
• 401 Unauthorized → Invalid API key
• 500 Server Error → Try again in a few seconds