Paste the LinkedIn post you want scored.

[
  {
    "question": "To score your post against real data, I need your LinkedIn history. How should I get it?",
    "header": "Data source",
    "multiSelect": false,
    "options": [
      {"label": "Scrape my posts", "description": "Pull my last 100 posts from LinkedIn via Apify. Takes 1 to 2 minutes, costs about $0.50."},
      {"label": "Use Charlie Hills data", "description": "Score against Charlie Hills benchmarks (1,872 avg engagement, 500 posts analysed). Good fallback."},
      {"label": "Skip data scoring", "description": "Score against generic best practices only. Less accurate but instant."}
    ]
  }
]

1. Ask for their LinkedIn username
2. Call Apify actor apimaestro/linkedin-profile-posts with input: { "username": "[their-username]", "total_posts": 100 }
3. Download results (do NOT use the fields parameter, it strips engagement data)
4. Save as [username]-all-posts.json in the project
5. Proceed to analysis

1. Calculate engagement score for every post: total_reactions + (comments x 3)
2. Identify the top 10% of posts by engagement score
3. From those top posts, extract:
   • Hook types that appear most often (contrarian, number-led, bold claim, personal story, question, news)
   • Average post length (word count)
   • Format distribution (text only, image, carousel, video)
   • CTA patterns (newsletter mention, comment gate, repost ask, question, none)
   • Topic clusters that over-index on engagement
   • Sentence rhythm (average sentence length, paragraph breaks per post)
4. Also note the bottom 10% patterns to identify what fails

• Does it use a hook type that historically performs for this author?
• Is it specific with a number, name, or concrete detail?
• Would it stop a scroll based on what actually stops scrolls in their data?
• Score 8+ only if the hook type matches a pattern in their top 10%

• Does the post match tone, rhythm, sentence length from voice.md?
• Does it violate any rule in voice.md's absence patterns section (what the voice never does)?
• Does the sentence length match the average from their top performers? If no voice files: score against the patterns extracted from their post data.

• Do their best posts teach, give steps, share data, or tell stories?
• Does this draft match that value pattern?
• Is the takeaway specific enough that someone would save or share it?
• Compare word count to their top 10% average. Flag if way over or under.

• What format (text, image, carousel) gets the most engagement for them?
• Does the draft's structure match the line break and paragraph rhythm of top posts?
• Is the post scannable on mobile?
• Does the CTA match patterns from their best performers?

• Did the user actually write this or does it read like unedited AI output?
• Would this post blend naturally into their feed based on their posting history?
• Are there any red flags: banned words listed in voice.md's absence patterns, generic phrases, corporate tone?
• Is it the right length compared to their top performers?

Want me to rewrite the weakest section using patterns from your top posts, or ship it?

• Always try to use real data before falling back to generic advice.
• Every score and every fix must reference specific data points, not subjective opinions.
• Never score higher than 8 unless the draft genuinely matches top 10% patterns.
• Be honest. A generous scorer is useless.
• If data is stale (14+ days old), suggest a refresh before scoring.
• Inform the user before running an Apify scrape (costs money).
• Never use em dashes in any output.
• British English throughout.
• Keep the scorecard compact. It needs to look good on a big screen at events.