xscore

Analyze any X (Twitter) account against the open-source X recommendation algorithm. Run one command, get your score.

Quick Start

npx -y @apoorvdarshan/xscore @username

That's it. Fetches tweets locally, scores 19 engagement signals, and opens a shareable results page in your browser.

Prerequisites:

• Node.js 18+
• Logged into x.com in Chrome, Edge, or Brave (uses browser cookies via @steipete/bird)

How It Works

1. Run the CLI — npx -y @apoorvdarshan/xscore @username
2. Local scoring — Fetches last 50 tweets using your browser cookies, scores each against the Phoenix weighted scorer formula
3. View results — Opens xscores.vercel.app with your results encoded in the URL (shareable)

Your data never touches a server. Everything runs on your machine.

What You Get

• Overall score (0-100) based on 19 engagement signals
• Signal breakdown — P(action) for likes, reposts, clicks, dwell time, etc.
• Per-tweet ranking — Every tweet ranked by P(favorite)
• Insights — What's boosting and hurting your reach
• Shareable link — Copy the URL and share it with anyone

Algorithm Mapping

Every piece of scoring logic traces back to xai-org/x-algorithm:

| App Component | Algorithm Source | What It Does | |---|---|---| | src/lib/algorithm.ts — ACTIONS | phoenix/runners.py — ACTIONS list | The 19 predicted engagement signals | | src/lib/algorithm.ts — SIGNAL_POLARITY | README.md | Classifies each signal as positive or negative | | src/lib/algorithm.ts — scoreTweet() | README.md — "Final Score = Σ (weight_i × P(action_i))" | The weighted scorer formula | | src/lib/algorithm.ts — estimateProbabilities() | phoenix/runners.py — probs = jax.nn.sigmoid(logits) | Approximates P(action) from tweet metrics | | src/lib/algorithm.ts — primaryScore | phoenix/runners.py — primary_scores = probs[:, :, 0] | Primary ranking uses P(favorite) | | src/lib/algorithm.ts — WEIGHTS | NOT IN REPO | Weights are not disclosed; we use ±1.0 placeholders |

The 19 Engagement Signals

From phoenix/runners.py:

| # | Signal | Polarity | |---|---|---| | 0 | favorite_score (PRIMARY) | Positive | | 1 | reply_score | Positive | | 2 | repost_score | Positive | | 3 | photo_expand_score | Positive | | 4 | click_score | Positive | | 5 | profile_click_score | Positive | | 6 | vqv_score (video qualified view) | Positive | | 7 | share_score | Positive | | 8 | share_via_dm_score | Positive | | 9 | share_via_copy_link_score | Positive | | 10 | dwell_score | Positive | | 11 | quote_score | Positive | | 12 | quoted_click_score | Positive | | 13 | follow_author_score | Positive | | 14 | not_interested_score | Negative | | 15 | block_author_score | Negative | | 16 | mute_author_score | Negative | | 17 | report_score | Negative | | 18 | dwell_time | Positive |

Development

npm install
npm run dev

Tech Stack

• CLI: Node.js + @steipete/bird
• Web: Next.js, TypeScript, Tailwind CSS v4
• Scoring: Based on xai-org/x-algorithm

Limitations

1. No real model inference — P(action) is approximated from tweet metrics. The real Phoenix model uses a Grok-based transformer.
2. Weights are unknown — The repo doesn't disclose weight values. All weights are ±1.0 placeholders.
3. No personalization — The real algorithm personalizes per-viewer. This shows engagement rates from the author's perspective.
4. Public metrics only — Negative signals (block, mute, report) are estimated since they aren't publicly observable.

Disclaimer

This is an independent tool not affiliated with X. Uses @steipete/bird for data access and xai-org/x-algorithm for scoring logic. Use at your own discretion.

Author

Built by Apoorv Darshan