v0.1.47 (Early Access)

A slow connection no longer gets you kicked out of a shared game, and joining one downloads less than half what it did. The audit recorder has its own console, with real logs in it.

Added

• Four more languages: Greek, Italian, Russian and Swedish. Pick yours in Settings, or let the game follow your system language. Coverage matches the other eight, engineer wording behind Tab included.
• The audit recorder has its own console. Walk to a racked SIEM and press E on its management port. Five tabs: the log, ingest, storage, reporting machines and ports. Search it or filter to alerts.
• The log lines are real logs now. Failed sign-ins, port scans, firewall denies, blocked exploits, filling disks, certificate expiries. Engineer mode shows the raw line, plain mode the same fact in words.
• The recorder shows how full its log drives are. The console gives capacity, use, fill rate and how far back it can go. Full drives drop the oldest logs, and the readout says so.

Fixed

• Equipment no longer vanishes from a rack when you look at it. Inspecting a machine stopped every other machine being sent to you, so racks emptied out and gear you had just installed disappeared.
• Panels stop showing things that are no longer there. A prospect counted but not listed, a growth request already resolved when you approve it, and network cards missing from hosts that have them.
• Joining a friend's game no longer drops you partway in. A large world took long enough to arrive that the host counted you as gone and disconnected you while you were still loading it.
• Joining shows you it is working. The loading screen now stays up until the world has actually arrived, with a running count, instead of dropping you into a datacentre that is still filling in.
• Object storage, Kubernetes, CDN and load balancers no longer unprovision themselves. Two workload records could share one name, so a serverless container took over a shard, worker node or proxy belonging to another customer.
• Storage placement runs once an hour, not dozens of times. Every save load gave each object-storage customer another copy of the hourly maintenance pass, and the copies never went away.
• Object-storage-only customers remember their isolation segment. Assigning one to a segment worked until you reloaded, and then they were back on the default.
• CDN, object storage and load balancer request rates read as 0.0. Their traffic was counted in one lump a minute while everything else was measured continuously, so the lists showed what was left between lumps.
• CDN and storage tenants no longer flicker as if they were down. The uptime strip read that same lumpy figure, so a tenant being served normally alternated green and grey four times a second.
• Racked equipment no longer disappears at middle distance. Machines vanished from their racks between roughly eight and twelve metres away, and came back as you walked closer or further off.
• Large datacentres draw faster. Equipment models carried a great deal of hidden internal geometry that no camera could ever see. A room packed with full racks now costs far less to render.
• Another player's actions show up straight away. Buying, racking, cabling and bonding took up to two seconds to reach everyone else, so the room saw the world as it was before.
• A slow connection no longer gets you kicked. The host used to drop anyone trailing the shared world by twenty seconds. You now get fewer, larger updates and catch up on your own.
• Notifications no longer go missing when several arrive at once. Anything queued behind the visible ones could be dropped instead of shown as space freed up.
• The player list shows bandwidth and lag, not just ping. Each row carries what the host spends on that player and how far behind they are. The card warns when the world itself runs slow.
• A slow join finishes instead of failing. When a joining player fell behind, the host resent the whole world, making it worse. It now waits for one update to land before preparing the next.
• The connection warning fires again, and far less often. It had stopped being able to fire at all. A player running slightly under full speed is normal and no longer reported as struggling.
• A player who leaves is cleaned up even if they played first. One who dropped out mid-session could keep their roster seat and hold an item forever.
• The network map no longer reports zero ports for distant switches. A switch outside your current room read as zero at no speed. It now shows a dash, because the count is not known there.
• The read-only API returned one row per list. Every collection endpoint reported a single entry whatever the page size asked for, so anything reading hosts, incidents or service planes over HTTP saw almost nothing.
• The audit recorder listed power sockets as network ports. Its Interfaces tab now shows network ports only, and a cabled port names what it is plugged into.
• The audit recorder's console port did nothing. Press E on a device's port and its console opens. The recorder's ports are named differently, so there was no way into the box at all.
• Committed CPU reads against what a host may sell, not its raw cores. An overcommitted machine used to report a share of its raw cores, so a full one showed figures like 329%.
• The allocation bar turns red only when a host is genuinely out of room. It used to sit full and red on every overcommitted machine, whatever headroom was left.
• A host past its allowance shows as over its limit, not simply full. Lower the overcommit dial below what a host has already sold and the bar says so until its workloads move.
• Dedicating hosts explains a refusal in a sentence. Refusing on a large fleet used to list every affected workload by name; it now names the tenants and counts the rest.
• Cables no longer shoot off to a point on the floor. A cable running to equipment that had just been removed was drawn to the centre of the world.
• An interrupted save no longer hides itself from the Load list. A crash while a save's details were being written left a damaged entry that never recovered.

Changed

• Comms loads a thread when you open it. Every conversation used to send its last fifty messages on every update. The list shows only the newest line, which is all the others carry now.
• Tenant detail loads when you open a tenant. The customer list used to carry every tenant's full per-service breakdown on every update, though it only ever showed a summary.
• Joining a game downloads less than half what it did. The world you receive on joining is now compressed, on top of the two changes above.
• Autosaves no longer take a screenshot. Grabbing one stalled the game briefly every five minutes, so autosave rows use a placeholder while manual saves keep their picture.

Notes

• Saves work in both directions with v0.1.46. The save format did not change. The recorder's retained log window is not saved: it refills within a few minutes of loading.
• Load once to heal. Existing saves have their duplicated maintenance passes and stale workload records cleaned up on load. Anything a container overwrote is re-placed on the next hourly pass.
• Multiplayer needs everyone on v0.1.47. The world update format changed for the comms work above, so a v0.1.46 client and a v0.1.47 host will not talk to each other.
• The customer API gains three fields. detail_omitted, provisioned_total and requested_total. Scraped reads still carry the full per-service breakdown, so nothing you already read changes.
• The multiplayer API reports host tick health. connection_health gains tick_delivery_rate: how much of its intended speed the shared world is actually running at, or absent when not yet known.
• Empty and withheld are different answers now. A genuinely empty list comes back empty; one an update left out is absent entirely. Readers that treated missing as empty should check for absence.
• A connection-health field was renamed and its meaning changed. pushes_dropped is now pushes_skipped, counting updates the host declined to build rather than ones it discarded. Read it as a gentler signal.
• --max-hold-secs now measures silence, not lag. A dedicated server drops a client that has sent nothing for that long. A client that is merely behind is never dropped for it.