v0.1.18 (Early Access)

Fixes

World & interaction

• A and D now step to the uplink rack. While you're inspecting a rack, A and D slide you to the rack next to it in the same row. In the data center the pre-built uplink rack sat slightly out of line, so it got skipped and you had to leave rack view and walk over. It's squared up in the row now, like the racks you place yourself, so A and D reach it directly. Its uplink hardware is still yours to unrack whenever you want. The cabinet itself stays put as a room fixture.
• Renamed racks show their name in the walk-up prompt. Rename a rack and the prompt reads "Enter 'DB Rack'" instead of "Open Rack 3 · 12U", so you can tell your racks apart without opening each one. Racks you haven't renamed read the same as before.
• Pressing Enter to confirm a name no longer traps the keyboard. You'd type a name into a rename or create field (a rack, a device, a private network, a host pool) and press Enter, and the text box kept hold of the keyboard afterward. Q, E and the rest did nothing until you clicked away, so stepping back or cancelling felt broken. Enter now hands the keyboard back the moment it confirms.
• Wall displays are up in the basement and the data center. NOC screens now hang in the basement (one) and the colo (two). Each one is a live channel that updates on its own, so there's nothing to open. Walk up and press E to move it to the next channel: Security, Finance, Network, Alerts, Capacity. Rooms with more than one screen boot them onto different channels, so you're not looking at the same page twice. The screens are read-only, so cycling them can't break anything. They're room fixtures for now; you can't buy or move them yet.
• Server management ports name the machine on hover. Every server has two console jacks on its face: the in-band USB management port and the out-of-band BMC port. Hovering either one opens that machine's console. The hover used to just say "BMC port", which was easy to mix up in a full rack. It now leads with the server's name, like "srv-01 · BMC port", and updates the moment you rename the server.

Business & customers

• The eviction fee is capped, and it can't bankrupt you. Firing a tenant charges a termination fee on top of the reputation hit. It's a slice of their monthly bill, and for a reserved tenant it also refunds part of the sum they paid up front at signing. That refund used to be the whole upfront payment, which on a big reserved contract could be a five-figure surprise with no ceiling. It's now capped at one month of that tenant's revenue. The base penalty came down too, to about half what it was and lighter the lower the tenant's tier, and the total is capped at your available cash so an eviction can draw your balance down but never below zero. The Evict button still prints the exact figure in its confirm step ("Confirm · -$1,240 + rep") before you commit.
• Network-isolation tenants pay the premium across their whole bill. Some customers won't share the network. A tenant that wants its own private network pays 2x the resource rate; one that wants that private network on dedicated, single-tenant hardware pays 4x. That premium now applies to every service on the account, including an isolated database, object store or load balancer. It used to hit only their virtual machines. The ask shows up front on the prospect card and on the customer's detail page before you sign them.
• Growth requests show the real cost and fit against the tenant's own hosts, and the list folds up. When a customer asks for more machines, the request now spells out the compute behind the ask, so you see "needs +8.0 vCPU · +16.0 GB RAM" next to the "+4 VMs" instead of a machine count alone. The "can you fit it?" verdict is measured where those machines actually land: a tenant on its own private network with dedicated hardware is checked against that segment's spare capacity, not your whole fleet. The card names the segment ("on Acme Net · L3VNI 5, 3 dedicated hosts"), reads "Fits on Acme Net's dedicated hosts" or "Over Acme Net's dedicated hosts, add hosts to the segment", and its free-capacity line switches from Fleet free to Segment free. Tenants on the shared fleet read against the fleet as before. The whole pile of pending requests is also collapsible now, with a running count in the header, so a busy inbox no longer scrolls the customers page forever.

Reliability

• A single-site garage or basement can't black out for no reason anymore. In a small single-site setup, the whole place could sometimes go dark at once: every machine and switch offline together, nothing on screen to explain it, and the network reading broken end to end. That can't happen at those tiers now, and any save already stuck in that state recovers when you load it back in.

Fleet & capacity

• Capacity counts only the machines that can actually serve. A server you'd racked but left switched off, or never cabled up, still counted as usable capacity: its cores, memory, storage and GPU padded your headroom even though it couldn't run a thing. A machine now adds to your capacity only once it's powered on and on the network. That applies everywhere the number shows: the Capacity lens in the ops console, the "hosts online" count on the cash card and the wallboard, and the "can you serve this customer?" check on a prospect's card, including the GPU check for a GPU tenant. A dark or unplugged box still takes up its rack space; it just stops counting as working.
• Object storage no longer raises a "can't be provisioned" you can't clear. A tenant's object store could be placed and serving fine, every file stored and uptime at 100%, and still show a permanent "Object storage can't be provisioned · No host capacity · add compute" incident. On a single-rack site it fired whenever the durability setting wanted copies spread across more racks than the site has, so "add compute" pointed at nothing you could actually do. That's fixed. The alert now shows only when a tenant is genuinely fenced off from racks it could otherwise use and your fleet has enough of them to spread across, which you can act on by freeing a rack or widening the tenant's reach. Saves already showing the bad alert clear it when you load back in.