New Features
Per-cable jacket colors. Pick a cable's jacket color while spooling (R cycles); colors persist and render per cable and per port.
Cleaner onboarding. A guidance overlay walks the first session; ops-console tabs unlock with progression instead of arriving all at once; quest blurbs
restored.
Graphics auto-detect. First run picks a preset from your GPU tier (Balanced default: 0.8 scale + FSR2), with adaptive down-shift under sustained load
and a "Recommend" button in Display settings.
Physical power cords. Newly racked gear arrives dark and must be powered. Tap E on a device's IEC inlet to run a power cord; hold E to
unplug (blocks un-racking a live device). Wall-mounted gear stays implicitly powered.
- Dark means dark. An unpowered device shows no LEDs at all: status, port link, and activity lights are off, fans are stopped, and its console opens with a NO POWER prompt (hosts, switches, gateways, and appliances alike). Link lights on the far end of a cable go out too: a gateway can't report a good LAN link into a dead switch, and no traffic routes through a dark box.
- Guidance points at anything still dark. If a racked device has no cord yet, the onboarding marker points straight at it with a "power the switch" (or server / gateway / appliance) prompt naming the AC inlet on the back of the device, and stays until you plug it in. A dark switch can no longer silently break your fabric while the quest ladder moves on ahead of it.
- Every rackmount server has two PSUs (matching the hardware models), so any host can take a second cord. A second feed draws a little more (+25 W) but survives a PSU failure that would take a single-PSU box down. Switch/gateway/appliance PSU counts are per-SKU.
- Cords are real cables. A molded plug seats in the inlet and the lead is dressed like a human ran it: staying level in the device's own U, out to the nearest side of the rack, along to the rear cable manager, then down inside the channel, including reverse-mounted switches and gateways, whose cords wrap around the side rather than through the chassis.
- New early quest rung ("Power the server"), first-plug concept card, and a codex article covering the redundancy/draw trade-off.
- Existing saves: already-running gear gets a cord seated automatically on first load, so nothing you built goes dark retroactively.
Attacks now have teeth. Unmitigated malicious traffic hurts named customers and is traceable end to end:
- Volumetric, Protocol, and Application floods degrade the tenants they target (focused blast radius, not a flat wipe), scaling with difficulty and escalating over time rather than one-shotting you on the first event.
- A sustained intrusion you ignore builds toward a breach: the customer takes a reputation, trust, and satisfaction hit, and an incident opens with real diagnosis guidance (what happened, why, and to rack an IDS/SIEM).
- Customers show Under Attack / Breached state on the roster and in customer detail; attack and breach events raise alerts.
Multiplayer / Co-op
- Co-op is surfaced for playtest and full editions (never the demo). Sandbox and challenge scenarios are co-op enabled; the campaign stays single-player.
- Difficulty scales with player count (more demand, more intense events per extra player).
- Remote players are now telepresence-robot rigs (wheeled base, screen head, glowing eyes for facing), with a player-chosen ring color, screen glyphs and toppers, real carried-equipment on the tray, and a telescoping pole that raises to the rack shelf a teammate is inspecting.
- Steam co-op host runs as a listen-server relay; join-to-boot handoff, host-only save ownership, guest kick-to-title on a dropped link, and join/leave toasts.
Networking & Simulation
Firewalls no longer read as "not connected to the internet." A firewall between your gateway and switch was falsely reporting the site offline; online
status is now derived from the actual network path, so inline firewalls and patch panels don't break it.
Redundant gateways. You can install more than one gateway per zone. Dual-homing a switch across two gateways no longer double-counts fabric load (which
previously tripped false "gateway saturated" warnings on a healthy redundant pair).
Link saturation is now link-local. A port/link hover reports saturation for that link only, as a fraction of its own capacity and per-direction.
A 10 Gbps link carrying ~900 Mbps reads "light," not "saturated," and plain and engineering (Tab) views agree. A gateway/EDD out of WAN capacity stays a
gateway/EDD condition, not an upstream switch's problem.
Multi-WAN is a policy, not a bond. Bonding your two internet uplinks with LACP never made sense (two independent ISP handoffs can't participate in your
LACP), and it didn't actually spread traffic, which is why one carrier line sat idle. Each carrier handoff now has an honest uplink policy you set in
the Internet panel (above its lines) or the gateway console (under WAN uplinks):
- Failover (default): the cheapest line carries the load; the others are overflow/standby and only light up once the primary saturates.
- Active-Active: live lines share the load proportional to their capacity, so both carry traffic. (True inbound multi-homing across ISPs is BGP territory, planned later.)
- The Internet panel and the network map now report the same per-line throughput instead of disagreeing.
- LACP link aggregation still works where it belongs: bonding your gateway's LAN-side links to your switches. The gateway console no longer lets you mark internet-facing uplink cables for bonding, and the duplicate LAG editor in Operations > Network was retired (bonding is managed in the switch and gateway consoles).
Uplinks match your site. The Internet panel only offers carrier lines your site's handoff can physically carry. A garage or basement site whose handoff
tops out at Consumer Fibre no longer lists Business Fibre or Metro to buy or upgrade into (they could never land on its 1 Gbps ports anyway), so the buy and
upgrade lists stay honest to what the site supports.
Economy & Customers
Object-storage (and other non-VM) customers work properly. Fixed the error when accepting a VM request from an object-storage customer, and prospect
cards show real estimated monthly revenue instead of $0/mo (every service type is priced, VMs included).
Object storage provisions across your whole fleet. Buckets now shard across hosts using your total free storage, instead of requiring a whole copy of
the bucket to fit on a single host. A large backup customer's bucket spreads over your available drives, so the storage service actually stands up (its control
plane and per-host agents come online) as soon as there is room for one copy, and fills toward full durability as you add capacity.
- Previously a high-durability bucket that couldn't fit all of its copies on single hosts silently failed to provision: the service sat OFFLINE with zero hosts and 0% uptime while still billing the customer, and never recovered even after you added space. That whole class of stall is gone.
- When a bucket still can't reach its durability target, an ops alert names the fix (add storage capacity so another copy fits, or lower the bucket's durability tier), and the customer's Services page shows how many copies are actually placed.
- Existing saves stuck with an unprovisioned object store repair themselves on load.
SLA uptime is measured over a rolling window. A customer's availability now reflects recent service health rather than their entire lifetime. Once you
fix an outage, uptime climbs back within a short period and the related incident clears on its own. Previously a single long outage could drag a customer's
lifetime average below their SLA target and leave the incident stuck open forever, even after service was fully restored.
Managed Kubernetes billing corrected. K8s clusters were billing a fraction of a cent; control-plane and per-node fees now charge the intended amounts.
Opex rebalance. Host running fees rebased to a uniform ~3% of capex/month (removing the ARM-to-balanced cost cliff). Power is priced per site (cheap
home supply in the garage, provider markup at colo, improving with scale at owned facilities). Business sites add a vendor support contract on hosts.
Customer roster pacing. Big-name customers arrive on a delay after you earn the reputation instead of flooding in on day one; several large personas
request multi-service bundles. New sandbox sliders for open roster and big-customer delay.
VM asks bundle storage. Every VM request now includes a block volume (sized to the ask), and VM asks right-size themselves to hardware you could
plausibly rack. Growth asks from existing customers surface in Business > Customers and on the customer detail page.
Object storage runs lighter early on. The object-store control plane and per-host agents reserve a lot less compute up front, so standing up a bucket at
a small site no longer eats a big chunk of a host before it serves anything. Large buckets still scale their footprint up as they grow.
Incidents, presented properly. Resolutions post to the #incidents comms channel, unread state clears itself, and Operations gains an Incidents tab;
incident root causes name the actual device.
Fixes
- Black screen on some older Intel Macs. A handful of pre-2020 Intel MacBooks (Iris/HD Graphics 615 and similar) launched to a black screen while audio and menu clicks still responded underneath. The game now detects those GPUs and switches to a compatible rendering path on startup instead.
- Placing a device into a rack while inspecting it now lets you cable it immediately, no more exit and re-enter.
- A cable's destination plug/boot renders on commit instead of ~2 seconds later.
- Removing a cable no longer leaves a phantom plug floating in the port.
- Demo "Keep Playing" after the win screen no longer freezes or loops the reload.
- Duplicate comms threads after loading a save are fixed (and existing saves merge theirs on load).
- Rack hover highlight reworked: precise rail-strip highlight on every device kind, and the cursor no longer drifts off target at off-axis angles; rack ears sit flush again.
- Port status lights now read the same up close and across the room: from a distance an errored port shows red and a blocked port shows amber, matching the up-close panel (previously the far view showed errored as amber and blocked as unlit).
- Switch/gateway/appliance consoles show a monthly opex chip.
- Server-shop pagination no longer clips off-screen in plain (non-jargon) view.
- Fixed a dev-capture tool that was accidentally running in shipped builds and buffering logs unbounded.
- The "add a line" rejection message now points you to the fix (upgrade to an EDD with more handoffs for a redundant line) instead of a misleading capacity hint.
- Cables routed out through a rack's roof cutout to another rack now follow the roof anchor whichever direction you lay them; laying it the "wrong" way no longer drew a straight line through the roof.
- Cables leaving an SFP / DAC port bend cleanly to their destination instead of wiggling, and the render never curves tighter than the cable's rated bend radius. Short jumpers no longer veer sideways for no reason.
- Gateway console reorganized: security is split into a live threat view (drops/s and per-class health) and a collapsible add-ons list, laid out in a wider two-thirds / one-third split. The diegetic inspect panel fades back while any device console is open so the console reads clearly on top.