Unknown Chains

In one minute:

By mid-July 2026, the WHO said 80% of new Ebola cases in eastern DRC came from transmission chains it could not trace.

If only one case in five is already linked to a known contact, a contact list catches a small slice of the outbreak, and every untraced case that spreads is one nobody is watching.

Surveillance is the lever that shrinks the untraced share, and the untraced share is what keeps the chain of new infections running.

1 July 2026: the quiet count

Weather forecasters have a habit a narrator could learn from. They do not wait for the storm to announce itself. They watch the pressure fall, the wind turn, the clouds build in the wrong direction.

On the first day of July, the Ebola outbreak in the Democratic Republic of the Congo was already a story. By the second week the headline numbers had climbed past the marks that make front pages. But the number that mattered most was not the death toll. It was a quieter question: where were the new cases coming from?

What was visible

A contact-tracing system works like a net. You pull in a case, find who it came from, list who it touched, and watch them. The net holds when new cases come from people already on the list. It tears when they do not.

Through the first half of July, the signal was in the shape of the outbreak, not in its size. A forecaster reading the reports side by side would have seen a net that was not holding. Cases were arriving from chains that no one had mapped, and that fact decides how much a faster lab or a bigger team can do.

Borders reacted before the map existed. Governments do not wait for a full picture when a disease crosses a line. The United States said it would block DRC citizens from commercial flights, and Canada barred foreign nationals who had visited the DRC in the prior 21 days. Those are the moves of a system that is still looking for the edges of the problem.

What happened

Two other signals arrived in the same weeks, and they pointed in different directions. A vaccine for the Bundibugyo strain, not the more familiar Zaire strain, entered its first human trial. And in neighbouring Uganda, the last patient was discharged, which started a 42-day countdown to being declared free.

The table below holds the numbers from the July record. Every figure has its source.

Table sources: NPR, Reuters, AP, Medical Xpress, Reuters, Reuters.

Read the table by its middle column. The 80% figure is the one that changes how the other rows should be read. A rising death toll tells you the outbreak is serious. An 80% untraced share tells you the net is torn, and that the rest of the numbers will keep arriving faster than the tracing can catch them.

The math

Start with a plain question: how often does a new case already sit on a list someone is watching?

Step 1. The share already linked to a known chain is the complement of the untraced share.

known-chain share = 1 − untraced share = 1 − 0.80 = 0.20

So, illustratively, of every 10 new cases, 8 come from unknown chains and only 2 were already on a contact list.

Step 2. Each untraced case seeds some further cases before it is detected. Call that average R. The source does not give R for this outbreak, so the value below is a teaching number, not an estimate. Let R = 2. Seeding from untraced cases is then:

effective R (untraced) = untraced share × R = 0.80 × 2 = 1.6

Under this illustration, each new case passes on about 1.6 further infections through chains nobody is tracing.

Step 3. Surveillance is the lever. Suppose better tracing moved the untraced share down to 0.50. The same arithmetic gives:

effective R (untraced) = 0.50 × 2 = 1.0

The seeding that goes unchecked falls from 1.6 to 1.0, a drop of (1.6 − 1.0) / 1.6 = 37.5%. The illustration uses one simple model, and the real epidemiology will be messier. The point holds anyway: the way to shrink effective R is to move cases from the untraced column to the traced column, and the only way to do that is to watch.

The Pawanax lens

For a clinic, the traced case is the one that walks in and gets asked the right questions at the door. The intake form, the travel question, the note on who else is sick at home: that is surveillance at the smallest possible scale. A clinic that asks well is a clinic whose cases belong to a chain someone can follow.

For families, the question is simpler. Many people will describe a fever to a phone before they describe it to a receptionist. A WhatsApp-native triage layer, the kind Pawanax is building, can carry those questions into homes and send the answers to the people who can act on them. That is the practical reading of the 80% figure: distribution decides who gets help.

For health systems in Kenya and East Africa, the lesson is that a border ban buys time but does not map a chain. Keeping the net intact takes a steady flow of data from clinics, labs and households, and it takes it before the outbreak is loud enough to make the news.

The next 90 days

Two registered forecasts sit on this chapter. Both were published on 16 October 2026 and will be scored.

• F6: 80%. The WHO Bundibugyo Ebola emergency declaration is still in force on 31 January 2027.
• F7: 20%. Kenya confirms at least one locally acquired, non-imported Bundibugyo case by 31 January 2027.

Both resolve on 31 January 2027, a few weeks after this 90-day window closes on 14 January. We will score them on the day they resolve and publish the result either way.

The scoring rule is the Brier score, the mean of (forecast − outcome) squared. For F6, if the emergency is still in force the score is (0.80 − 1)² = 0.04. If it has been lifted, the score is 0.80² = 0.64. For F7, if a local case is confirmed the score is (0.20 − 1)² = 0.64. If none is confirmed, it is 0.20² = 0.04. A forecaster who says 50% on everything scores 0.25, so a skilled forecaster must land clearly below that. If the score comes in above 0.25, we will say so.

Part 7 of The Signal Ledger. Read the series: [The Signal Ledger](/blog/signal-ledger-2026-read-backwards).

Cover image: One Health 130420-F-MV374-825.jpg — Capt. Jason Ostrich · Public domain. Photographed 20 April 2013 in Kakute, Uganda, during a site survey of village homes. The image shows a household health assessment team, not a current Ebola contact tracing operation or a 2026 outbreak. Resized for web display.