Twelve months at 99.995%: a year inside Hall A

Hall A has now run for twelve months with customers on the floor. Over that year, measured from the customer side of the network, the hall was available 99.995% of the time. That is about 26 minutes of downtime, spread across two events.

The two incidents

The first was a firmware fault in a top-of-rack switch in November that took one row offline for 17 minutes. The second was a planned grid transfer in April that ran nine minutes longer than its window. Both have public post-mortems, and both changed how we work: switch firmware now rolls out one row per day, and grid transfers run on the generator side first.

What the weather did

Polar night lasted from early December to early January. Outside air averaged −8 °C over the winter, with a low of −24 °C in February. The intake louvres iced twice, and both times the recirculation loop cleared them before inlet temperatures moved.

Summer was the bigger test. We saw eleven days above 20 °C outside, and the hall stayed inside its inlet target on all of them without a compressor.

What we would do again

Write everything down, walk the aisle every morning and keep the hall boring. The most useful tool of the year was a shared notebook in which every technician logs what they touched and why. Next year that habit moves to Kiruna with the first N-02 team.