Uptime: Wiki | v0.1.26b (Early Access)

A hotfix for v0.1.26, mostly performance on large saves plus a set of things that were quietly doing the wrong thing instead of failing loudly.

Fixed

Cut the biggest cause of stutter on large saves. v0.1.26 moved object storage and managed-database backups into the shared workload store, but several lookups still walked the whole store to answer a question about one tenant. Invisible on a small world, millions of wasted comparisons per tick on a big one. They now do one pass instead of one per tenant. There is more to do here.
A host left in maintenance no longer costs you every tick. If its workloads had nowhere to go, the game re-planned your entire fleet's evacuation for every cordoned host, every tick, and discarded the result each time. It now plans once, looks only at cordoned hosts, and stops redoing the work while nothing has moved.
Contested relocations are deterministic. When several cordoned hosts wanted the same replacement, the winner could vary between runs.

A rack gateway can no longer be installed on top of something else. Every other rack-mounted device checked its slots were free first. Gateways did not, so one could be racked through a server already there.

The cable you pay for is the cable you get. Choosing an active direct-attach cable could silently commit a passive one, which has a shorter usable reach, and still report success. An unrecognised jacket colour fell back to a default the same way. Both now refuse the order and say why.

Migrating a workload can no longer move somebody else's. Each service type numbers its units separately, so the same number is valid in several of them. Naming the wrong type with a real number could move a different tenant's workload. The request is now checked against the type, tenant and host you named.

Nameplates can no longer show a stranger's Steam account. A Steam ID was too large to carry exactly, so it arrived rounded, and a rounded ID is usually still somebody else's real account. It is now carried as text.

A timed-out command no longer invites you to repeat it. The game said the server was busy and suggested retrying while the request was still running and usually still took effect, so repeating it could apply a purchase twice. It now says the outcome is unknown.
Notifications and graphs no longer come up empty in silence. On the compact wire format, a malformed batch of story events or graph samples was accepted as an empty one, so incident cards never appeared and sparklines sat flat at zero. It is now an error.
An unrecognised connection format is rejected, not silently downgraded to the slowest one, which showed up as unexplained hitching on large updates.

Saves from v0.1.26 load as normal. No save-format or balance changes.