DESIGN ID: D02

[U0]
Analytical Workflow: Vastopolis Outbreak Investigation

[U1]
Stage 1 - Data Preparation
- Clean & geocode microblog data; remove duplicates/bots; parse timestamps to consistent timezone.
- Join population statistics to the 13 zones (spatial join using zone boundaries).
- Align weather data (wind speed/direction) by day for later correlation.
- Checkpoint: Visual scatter of all geolocated posts on the map to confirm coverage and spot GPS errors (e.g., points outside the metro area).

[U2]
Stage 2 - Text Filtering & Symptom Classification
- Apply keyword/NLP filtering (flu, fever, cough, sick, vomiting, etc.) to isolate illness-related posts.
- Optionally cluster symptom terms to distinguish respiratory vs. gastrointestinal language (clue for airborne vs. waterborne).
- Checkpoint: Analyst manually samples classified vs. unclassified posts to verify precision/recall of the filter.

[U3]
Stage 3 - Spatiotemporal Exploration
- Build a time-animated density map (heatmap or hexbin) of illness-related posts by hour/day.
- Identify the earliest cluster(s) in time - candidate outbreak origin.
- Overlay hospitals, landmarks, water bodies, highways to see proximity of early cases.
- Checkpoint: Domain expert inspects animation to flag plausible index location(s) (e.g., near a reservoir, transit hub, or stadium).

[U4]
Stage 4 - Population Normalization
- Compute illness-report rate per capita per zone per day (using population stats), not just raw counts.
- Rank zones by normalized incidence to correct for population density bias.
- Checkpoint: Compare raw vs. normalized hotspot maps - resolve discrepancies with analyst judgment.

[U5]
Stage 5 - Transmission-Pathway Hypothesis Testing

| Hypothesis | Expected Pattern | Data to Check |
| Waterborne (point source) | Cases cluster around a specific water body/supply; sudden onset near source, then diffuse | Overlay cases on water infrastructure; check GI symptoms |
| Airborne (wind-driven) | Case density shifts downwind over time; direction matches wind vector | Cross-correlate daily wind direction with spatial shift of new cases |
| Person-to-person | Gradual spatial spread from index case along transit/commute routes; growth follows population density/mobility, not wind | Check spread along highways/transit corridors, contact-network-like diffusion |

- Visualize case centroid movement per day vs. wind vector per day (overlay arrows).
- Visualize spread rate vs. distance from highways/transit to test person-to-person diffusion.
- Checkpoint: Analyst determines which hypothesis best fits observed spatial-temporal signature; may combine (e.g., point-source origin + secondary person-to-person spread).

[U6]
Stage 6 - Source Localization
- Trace back the earliest normalized cluster and cross-reference with map landmarks (e.g., convention center, water treatment plant, market).
- Validate using text content of earliest posts (mentions of specific location/event).

[U7]
Stage 7 - Synthesis & Reporting
- Combine spatial, temporal, and textual evidence into a narrative: likely origin point, transmission mode, and current spread trajectory.
- Provide actionable map for emergency response (zones needing resources, quarantine perimeter suggestions based on wind/mobility spread model).

[U0]
Tools suited for this: GIS/mapping (QGIS, Kepler.gl, ArcGIS), text mining (Python/NLTK or spaCy), time-series/animation (Tableau, D3, or Python+Plotly), all with iterative human review at each checkpoint above.