A portfolio-grade Streamlit analytics application built from the real Excel survey files in this repository.

This improved final version is stronger on the points that matter in a serious portfolio review:

• cleaner decision flow with filtered-vs-baseline benchmark context
• better UX utility through active filter summaries and export of the filtered analytical layer
• broader behavioral reading with access-style mix analysis by generation and value tier
• safer exploration flow with explicit empty-state handling and a corrected missingness chart
• stronger maintainability through clearer sidebar logic, reusable metrics helpers, and additional tests

The application analyzes digital cultural consumption in France through four decision lenses:

• audience structure
• engagement intensity
• monetization potential
• risk-heavy access behavior

It combines a curated analytical export with broader raw barometer modules to produce a dashboard that is more useful than a basic descriptive survey app, while staying technically honest about the integration limits.

Compared with the previous delivered zip, this version adds practical improvements that are actually useful:

• active filter summary in the sidebar instead of making the user infer the state mentally
• download button for the filtered analytics dataset for handoff and validation
• baseline comparison table in the executive overview to avoid reading the filtered slice in isolation
• segment opportunity table to surface conversion headroom directly
• generation × access-style share chart in the behavior section
• value-tier × access-style share chart in the monetization section
• better audience deep-dive with filtered-share vs full-population-share context
• correct missingness visualization in the survey explorer
• empty-filter safety so the dashboard fails gracefully instead of silently producing nonsense
• cleanup of the distributable zip with no cached test artifacts included

Raw inputs used by the project:

• data/raw/2024-barometre-consommation.xlsx
• data/raw/2024-datamap.xlsx
• data/processed/data.xlsx

Generated helper artifacts:

• data/processed/raw_module_inventory.csv
• data/processed/datamap_module_excerpt.csv
• data/processed/analytics_dataset_generated.csv

data.xlsx is kept as the primary interpretable layer.

That is the right architectural decision here because it preserves a business-readable subset of the survey and avoids pretending that the project fully reconstructs the entire raw questionnaire semantics.

The raw barometer is used to derive broader feature blocks from selected modules such as:

• paid vs free content matrix
• device diversity
• illicit access channels
• illicit methods breadth
• stream-ripping behavior
• legal-offer advantages
• legal-consumption motivations

The project builds explainable composite signals such as:

• engagement_score
• monetization_score
• risk_behavior_score
• content_depth_score
• raw_risk_intensity_score
• legal_value_perception_score

The final layer turns those signals into interpretable consumer segments:

• Premium Legal Users
• Hybrid Monetizers
• Heavy Free Users
• Risk-Oriented Free Users
• Casual Value Seekers
• Low Engagement Users

This is more defensible than decorative ML on top of a survey structure that does not justify it.

• KPI board with quantiles, monetization, engagement, and risk signals
• filtered view vs full population benchmark table
• segment opportunity table for conversion headroom
• spending distribution and engagement progression by spending quintile

• spending dispersion by generation
• generation → family stage → segment composition
• urbanity vs content breadth
• profession concentration
• audience deep-dive with filtered vs full-population share context

• access style mix
• legality self-positioning
• segment-level risk radar
• VPN / DNS / cracked-app behavioral intensity
• generation × access-style share

• conversion funnel
• segment × value-tier structure
• engagement distribution by value tier
• value-tier × access-style share
• spending percentiles by segment

• content-category breadth
• device diversity
• illicit-method breadth
• legal value perception
• content breadth vs spending

• segment distribution
• generation × segment mix
• average score profile by segment
• risk vs engagement scatter
• segment summary and opportunity prioritization

• analytical missingness
• datamap variable search
• text-label search
• raw-column explorer with preview and distribution

• feature engineering
• segment logic
• limitations
• pipeline summary

.
├── app.py
├── data/
│   ├── raw/
│   └── processed/
├── scripts/
│   └── __init__.py
├── src/
│   └── dashboard/
│       ├── sections/
│       └── utils/
├── tests/
└── .github/workflows/

pip install -e ".[dev]"
streamlit run app.py

Alternative entry point:

dcca-dashboard

To export the generated analytics table:

python scripts/build_analytics_dataset.py

ruff check .
pytest -q

The project uses conservative row-order alignment between the curated analytical file and the raw barometer because the shared business key is not exposed in the processed export.

That is acceptable for a portfolio project as long as it is stated clearly. It is not equivalent to a production-grade keyed integration.