How this outage index is built, and where it stops.

Why one outage can appear four times

The rules apply to downstream carriers too, not only to the carrier whose equipment actually failed. ACMA says so plainly: information about the same underlying outage may appear in several registers, each reflecting that carrier's own view of the damage.

So a single backhoe through a fibre can surface as four entries with four reference numbers, four slightly different start times and four different counts of affected services. Read one register and you see a fault. Read four and you see the same fault four times.

We join them. Two entries from different carriers are treated as one event when their outage windows overlap and either the geography overlaps, or one names the other as the carrier responsible for resolving it. That second signal is the strong one, because it is an explicit statement of who owned the fault. Entries that merely happen to fall on the same day in different states stay separate.

Within a single carrier we also collapse progressive updates. Some registers publish a row per update to the same incident; we keep the most complete, resolved entry and record which rows it superseded.

Judgement calls we had to make

Real registers are messier than the standard imagines. These are the decisions that affect the numbers:

Duration is written as hours.minutes, not decimal hours. A value of 3.9 means three hours and nine minutes, not three and nine tenths of an hour. Reading it as an ordinary decimal inflates that outage by 45 minutes. We confirmed the reading against entries that state a start time, an end time and a duration: the three only agree under the hours.minutes interpretation.

Some technologies were being counted twice. One register lists services as “NBN broadband services connected by means of FTTC; FTTN; FTTP”, which is one description spread across a semicolon-separated list. Split on the semicolons alone, the same technology lands in two different buckets depending on where it appeared in the list. We fold every variant back to the technology itself.

Blanks are not zeroes. Where a carrier cannot see how many services were affected, the standard lets it say so in words instead of giving a number. We treat that as unknown and leave it out of totals, rather than counting it as none.

We repair obvious typos and say so. One register published a date as 28/07/026. We read that as 2026 and flag the repair rather than dropping the entry.