BeyondCX_Insights/docs/CHANGELOG.md

CHANGELOG

Registro de cambios del proyecto CXInsights

---

[2.1.0] - 2026-01-19 - Streamlit Dashboard + Blueprint Compliance

Added

Streamlit Dashboard (dashboard/)

• Main Application (app.py)

  • 8 navigation sections: Overview, Outcomes, Poor CX, FCR, Churn, Agent, Call Explorer, Export
  • Automatic batch detection and selection
  • Beyond brand header with logo and metadata

• Brand Configuration (config.py)

  • Beyond Brand Identity colors: Black #000000, Blue #6D84E3, Grey #B1B1B0
  • Custom CSS with Outfit font (Google Fonts)
  • McKinsey-style chart formatting

• Data Loading (data_loader.py)

  • load_batch_data() - Load summary and individual analyses
  • calculate_kpis() - Dashboard KPI calculations
  • aggregate_drivers() - Cross-analysis driver aggregation

• Visualization Components (components.py)

  • render_kpi_cards() - Metric cards with deltas
  • render_outcome_chart() - Horizontal bar chart
  • render_driver_analysis() - Driver frequency bars
  • render_fcr_analysis() - FCR 4-category view per blueprint
  • render_churn_risk_analysis() - Binary churn risk view per blueprint
  • render_agent_performance() - Talento Para Replicar / Oportunidades de Mejora
  • render_rca_sankey() - RCA flow: Driver → Outcome → Churn Risk
  • render_driver_correlation_heatmap() - Jaccard similarity co-occurrence
  • render_driver_outcome_heatmap() - Driver distribution by outcome
  • render_outcome_deep_dive() - Root causes, correlation, duration analysis
  • render_call_explorer() - Filterable call detail table

• Export Functionality (exports.py)

  • create_excel_export() - 8-sheet workbook with all analysis data
  • create_executive_summary_html() - Branded HTML report
  • create_json_export() - Raw JSON data package

• Theme Configuration (.streamlit/config.toml)

  • Light theme base
  • Brand primary color
  • Port 8510 configuration

Changed

Blueprint Terminology Compliance

• FCR Analysis: Updated from simple distribution to 4 categories

  • Primera Llamada Sin Riesgo de Fuga
  • Primera Llamada Con Riesgo de Fuga
  • Rellamada Sin Riesgo de Fuga
  • Rellamada Con Riesgo de Fuga

• Churn Risk Analysis: Updated to binary categories

  • Sin Riesgo de Fuga (LOW, MEDIUM)
  • En Riesgo de Fuga (AT_RISK, HIGH)

• Agent Performance: Updated labels to blueprint terminology

  • Talento Para Replicar (positive skills)
  • Oportunidades de Mejora (improvement areas)

FCR Rate Calculation Fix

• Before: FIRST_CALL + RESOLVED counted as success
• After: Only FIRST_CALL counts as FCR success (per blueprint)
• Updated in data_loader.py and exports.py

Fixed

• Plotly colorbar titlefont deprecated property → title.font
• Streamlit dark theme issue → Light theme in config.toml
• Port conflicts → Using port 8510

---

[2.0.0] - 2026-01-19 - Blueprint Alignment

Added

New Analysis Dimensions (High Priority Gaps)

• FCR Detection Module

  • FCRStatus enum: FIRST_CALL, REPEAT_CALL, UNKNOWN
  • fcr_status field in CallAnalysis
  • fcr_failure_drivers field for tracking repeat call causes

• Churn Risk Classification

  • ChurnRisk enum: NO_RISK, AT_RISK, UNKNOWN
  • churn_risk field in CallAnalysis
  • churn_risk_drivers field for churn indicators

• Agent Skill Assessment

  • AgentClassification enum: GOOD_PERFORMER, NEEDS_IMPROVEMENT, MIXED, UNKNOWN
  • AgentSkillIndicator model with skill_code, skill_type, evidence, coaching recommendations
  • agent_positive_skills and agent_improvement_areas fields in CallAnalysis

• Enhanced RCALabel Structure

  • DriverOrigin enum: AGENT, CUSTOMER, COMPANY, PROCESS, UNKNOWN
  • origin field in RCALabel for responsibility attribution
  • corrective_action field for actionable recommendations
  • replicable_practice field for positive behaviors to replicate

New Taxonomy Categories

• churn_risk drivers: PRICE_DISSATISFACTION, SERVICE_QUALITY_ISSUES, REPEATED_PROBLEMS, COMPETITOR_MENTION, CONTRACT_ISSUES, BILLING_PROBLEMS
• fcr_failure drivers: INCOMPLETE_RESOLUTION, PENDING_ACTION_REQUIRED, MISSING_INFORMATION, UNCLEAR_NEXT_STEPS, SYSTEM_LIMITATIONS, PROMISED_CALLBACK
• agent_skills.positive: EFFECTIVE_CLOSING, GOOD_RAPPORT, OBJECTION_MASTERY, PRODUCT_KNOWLEDGE, ACTIVE_LISTENING, EMPATHY_SHOWN, SOLUTION_ORIENTED, CLEAR_COMMUNICATION
• agent_skills.improvement_needed: POOR_CLOSING, MISSED_OPPORTUNITIES, OBJECTION_FAILURES, KNOWLEDGE_GAPS, PASSIVE_LISTENING, LOW_EMPATHY, PROBLEM_FOCUSED, UNCLEAR_COMMUNICATION

New Files

• config/prompts/call_analysis/v2.0/system.txt - System prompt for v2.0
• config/prompts/call_analysis/v2.0/user.txt - User prompt with all taxonomy sections
• config/prompts/call_analysis/v2.0/schema.json - JSON schema for v2.0 response
• docs/GAP_ANALYSIS.md - Comprehensive gap analysis vs BeyondCX blueprints

Changed

Models (src/models/call_analysis.py)

• Added 4 new enums: FCRStatus, ChurnRisk, AgentClassification, DriverOrigin
• Extended RCALabel with origin, corrective_action, replicable_practice fields
• Added AgentSkillIndicator model
• Extended CallAnalysis with 7 new fields

Inference (src/inference/)

• prompt_manager.py: Added TaxonomyTexts dataclass, updated load_taxonomy_for_prompt() to return all sections
• analyzer.py: Updated to parse all v2.0 fields, added _parse_agent_skills() method
• Default prompt_version changed from v1.0 to v2.0

Aggregation (src/aggregation/)

• models.py: Added DriverCategory type, extended RCATree and BatchAggregation with v2.0 fields
• statistics.py: Updated calculate_frequencies() to return dict with 6 categories, added FCR/churn/agent metrics to calculate_outcome_rates()

Configuration

• config/prompts/versions.yaml: Changed active version from v1.0 to v2.0
• config/rca_taxonomy.yaml: Added 3 new top-level sections

Tests Updated

• tests/unit/test_inference.py: Updated active version assertion to v2.0
• tests/unit/test_aggregation.py: Updated sample_analyses fixture with v2.0 fields, updated frequency tests

Documentation

• Updated docs/PROJECT_CONTEXT.md with v2.0 status
• Updated docs/TODO.md with completed and new tasks
• Created docs/CHANGELOG.md (this file)

---

[1.0.0] - 2026-01-19 - MVP Complete

Added

• Complete pipeline: transcription → features → inference → aggregation → exports
• CP1-CP8 checkpoints completed
• AssemblyAI transcription with diarization
• GPT-4o-mini inference with JSON strict mode
• Transcript compression (>60% token reduction)
• RCA tree building with severity scoring
• Export formats: JSON, Excel, PDF/HTML
• CLI interface with resume support
• Comprehensive test suite

Features

• CallAnalysis model with observed vs inferred separation
• RCALabel with mandatory evidence_spans[]
• Versioned prompts system
• Checkpoint/resume mechanism
• Batch processing with rate limiting

---

Última actualización: 2026-01-19