—
interactions
Module 01 · Operational throughput
Weekly conversation volume
Cumulative interactions per ISO week. The shape — gentle ramp from launch through week 26 — reflects the network's gradual onboarding curve, not a stable flat run rate.
Module 02 · Severity composition
Emergency / urgent / routine over time
Stacked weekly. The emergency band is the heaviest signal for ministry routing — it's the queue your district focal points wake up to.
Module 03a · Topic mix
Where the questions cluster
All interactions over 26 weeks, grouped by extracted topic.
Module 03b · District activity
Geographic distribution
Population-weighted, but real signal is in the per-district CHW count and how it scales.
Module 04 · Workforce activity
Per-CHW, per-week conversation heatmap
Top 50 CHWs across 26 weeks. Each cell = one week of activity for that worker. Darker = more conversations. Real CHW networks have this shape: a long tail of high-activity workers, dormant weeks, gradual onboarding.
activity:
0
1–3
4–7
8–14
15+
Module 05 · Bayesian framework value
Confidence interval tightens with data
The point of the Bayesian update is that the posterior CI shrinks as observations accumulate. Below: cumulative CI for the HIV-prevalence indicator, week-by-week. Strong UNAIDS prior (effective n=200); the 5,000 CHW interactions move it modestly. This is what scale buys you — at 100,000 conversations the CI collapses.
Module 06 · What scale unlocks
Project the network forward
Move the sliders. Watch the posterior CI for HIV prevalence collapse as N grows. This is the framework's commercial value — at 1,000 CHWs across SSA, luma's national-level posteriors rival WHO household-survey precision, but updated weekly instead of every five years.
50100k
4 wks3 yrs
—
projected interactions
—
posterior mean
—
95% CI width