About

This is a comprehensive user behavior analytics and success prediction canvas for the Zerve platform. It analyzes 409K+ events from 4,774 users to identify the behavioral patterns, event sequences, and feature workflows that distinguish successful from unsuccessful users.

**Architecture Overview:**
The canvas flows in two main pathways: (1) **Data Foundation Path** — column auditing, success metrics derivation, and user-level feature engineering; (2) **Analytics & Prediction Path** — behavioral sequence analysis, n-gram lift scoring, workflow simulation, and actionable intelligence generation.

**Key Analytical Outputs:**
- **Success Metrics:** Derives user success via retention (≥2 active weeks), depth engagement (≥3 distinct feature types), and compute usage (credits consumed)
- **Behavioral Sequences:** Identifies top n-gram patterns (bigrams/trigrams) that predict success with lift ratios ranging from 7× to 272M× over unsuccessful users
- **Workflow Stages:** Maps users across five progression stages (Explorer → Builder → Runner → Iterator → Producer) with event-based flags
- **Predictive Scoring:** Implements a 5-component weighted success score (retention 25%, execution 25%, iteration 20%, progression 15%, diversity 15%) normalized 0-100
- **Event Transitions:** Builds Markov transition matrices showing P(next_event | current_event) differential between successful vs unsuccessful users
- **Simulation Engine:** Deploys a rule-based + probabilistic scoring system that can evaluate arbitrary event sequences in real-time and output stage, risk level, and next-best-action recommendations

**Visualization Suite:**
Six behavioral charts (top events, transitions, funnel, feature adoption, comparison ratios, distinguishing events), four sequence analysis heatmaps (n-gram lift, transition matrices, early predictive bigrams), and a Sankey flow diagram showing user progression through workflow stages.

**Final Outputs:**
- `user_intelligence_table` (4,774 rows × 15 cols) — each user with success_score, stage, risk_level, and all feature flags
- `simulation_engine_config.json` — serialized weights, keywords, n-gram lift table (200 sequences), early bigrams, and NBA recommendation map
- 10 presentation stats: total users, successful %ile, stage distribution, top flag lift, feature adoption gaps, funnel dropoff, event ratios, early-warning AUC, exclusive bigrams, and creation friction
- Callable Python functions: `score_sequence(event_list)` and `compare_to_top_users(event_list)` for real-time evaluation