📚 Algorithm Internals & Architecture
Skill References
Algorithm Internals & Architecture
Algorithm Signals (by weight)
| Signal |
Weight |
Generated by |
| Extended dwell (>2 min on thread) |
11.0x |
Scroll + read pause |
| Profile click |
4.0x |
Influencer / own profile visits |
| Reply / comment |
1.0-27.0x |
Comment on posts |
| Bookmark |
~1.0x |
Bookmark quality content |
| Like |
0.5x |
Like niche posts |
| Follow |
High (graph) |
Follow niche accounts |
| Search |
High (intent) |
Search niche keywords |
Phase Model (fresh account → thought leader)
| Phase |
Days |
Focus |
| Seeding |
1-7 |
Pure consumption. Search + scroll + like + follow. |
| Engagement |
8-30 |
Replies, quote tweets, early engagement on viral posts. |
| Authority |
31-90 |
Original threads, community building, Spaces. |
| Maintenance |
90+ |
Consistency, content creation, community stewardship. |
24/7 Architecture
┌─────────────────────────────────────────────────┐
│ CLI: xactions persona run │
├─────────────────────────────────────────────────┤
│ Persona Engine (personaEngine.js) │
│ ├─ Niche presets (6 built-in + custom) │
│ ├─ Activity patterns (sleep/wake schedules) │
│ ├─ Engagement strategies (limits & ratios) │
│ ├─ Session planner (shuffled activity queues) │
│ └─ State persistence (~/.xactions/personas/) │
├─────────────────────────────────────────────────┤
│ Algorithm Builder (algorithmBuilder.js) │
│ ├─ Puppeteer + stealth (headless Chrome) │
│ ├─ Human-like behavior simulation │
│ │ ├─ Variable scroll speeds │
│ │ ├─ Reading pauses (2-8s) │
│ │ ├─ Mouse movement with jitter │
│ │ ├─ Typing with per-char delay (30-100ms) │
│ │ ├─ Coffee breaks (30s-2min) │
│ │ └─ Session gaps (10-120min) │
│ ├─ Activity executor │
│ │ ├─ search (top + latest tabs) │
│ │ ├─ browse_home (For You timeline) │
│ │ ├─ like (on-topic tweets) │
│ │ ├─ follow (from search & target accounts) │
│ │ ├─ comment (LLM-generated in persona voice)│
│ │ ├─ create_post (LLM original content) │
│ │ ├─ profile_visit (targets & discovered) │
│ │ ├─ check_own_profile │
│ │ └─ check_notifications │
│ └─ Graceful shutdown (SIGINT/SIGTERM/abort) │
├─────────────────────────────────────────────────┤
│ LLM Brain (OpenRouter) │
│ ├─ Comment generation (persona-voiced) │
│ ├─ Post creation (niche-relevant) │
│ ├─ Reply threading │
│ └─ Any model: Gemini, GPT-4o, Claude, Llama │
└─────────────────────────────────────────────────┘
Cost Estimate
| Component |
Monthly Cost |
| VPS |
$5-10 (any VPS with Chrome support) |
| LLM API |
$3-10 (depends on model and volume) |
| Total |
~$8-20/month for 24/7 operation |
Research Papers
Full technical deep dives:
docs/research/algorithm-cultivation.md — X's recommendation algorithm internals (SimClusters, TwHIN, scoring formula), signal taxonomy, human behavior simulation, detection avoidancedocs/research/llm-powered-thought-leader.md — LLM-powered architecture design