Reports · update

Malformed names removed from the census

15 October 2026 · Dirišwago
Figures as of 2026-10-15 15:48 SAST; they will not change.

On 15 October 2026 Palo removed 4,434 entries from the census that were not the name of one website. From the next analysis on, the number of sites found and the number of dead sites are each about 4,400 lower. Nothing about the South African web changed: these entries were never websites.

What they were

Before October 2026, one of Palo's sources was the certificate transparency logs, the public record of the security certificates that websites use. A certificate lists every name it covers, and sometimes a contact e-mail address. The old discovery code stored some of those lists whole, as a single "domain". The census held entries like these:

Stored as one domain What it holds
ru.ac.za, twice, on two lines the same name twice
[email protected], with two more names on further lines an e-mail address and two names
*.example.co.za a wildcard, which covers many names and is not itself a site
.nisis.gov.za a name with a stray leading dot

Some held a ligature (the single character "fi" in place of "fi") or an invisible zero-width character. That code was replaced in October 2026, and nothing has stored such entries since; the existing ones stayed until now.

How they were counted

An entry holding several names cannot be visited, so Palo recorded most of them as dead.

Counted as Entries
dead 4,414
active 17
blocked 2
unavailable 1
total 4,434

The few counted as active or blocked were ones Palo could still reach, for example an address in the form [email protected], which leads to example.co.za. In those cases the same site was likely counted twice.

What was done

Reading the figures