This entry started as a short note that JefeOS 1.1.1 had shipped. The live system supplied a more useful ending. Version 1.1.1 turned the previous day's OpenStack work into a tagged release, rolled onto JefeOS-Bot, and then let the real Node workload find a kernel boundary mistake that the synthetic release checks had missed. The correction is now JefeOS 1.1.2.
What 1.1.1 Proved
The annotated v1.1.1 tag pointed to d368d8aa, the exact Gate #4 GREEN commit. It codified the OpenStack work proven the day before: Glance import, an ordinary Nova lifecycle under nested KVM, an ISO 9660 config drive, Nova hostname and SSH-key application, DHCP over E1000, and boot evidence through the serial console and Horizon. The Hyper-V bot VM then booted the versioned 1.1.1 artifact, passed the Xylem codec and state suites, launched supervised nodebot, armed its stall watchdog, and answered an authenticated SSH health check.
The Real Workload Found the Missing Boundary
After a normal Xylem harvest and restart, Node collected status by enumerating /proc. The Linux getdents64(2) implementation had accepted a valid userspace address, but its synthetic /proc and tmpfs branches still built directory records through that address from privileged kernel code. SMAP correctly blocked the write and turned the latent mistake into a panic. NTFS already used kernel staging plus copy_to_user; the two synthetic branches did not.
The Test Had to Become More Honest
The old in-kernel proctest used a kernel buffer, so it exercised directory synthesis without crossing the SMAP boundary. The Linux probe read files under /proc but did not enumerate the directory, the SMAP test concentrated on rejected pointers, and the 1.1.1 soak used a synthetic supervised workload instead of requiring a real post-reap status request. All four checks could pass while the successful userspace path remained untested. That is the important failure here: not that one test lacked one case, but that the release evidence stopped just before the privilege boundary used in production.
1.1.2 and Gate #5
Version 1.1.2 builds every synthetic directory record in a bounded kernel buffer, sends it through the fault-survivable copy_to_user path, and advances the directory cursor only after a complete record reaches userspace. The promoted regressions cover successful enumeration of /proc and /proc/self, an unmapped destination, a cross-page short copy with correct retry behavior, BusyBox ls and ps, Node's own fs.readdirSync("/proc"), and a real !jefeos status after Xylem reaps and restarts the bot. The two focused probes now run in the fast and lightning profiles instead of waiting for a nightly lane.
The exact release commit is 18f99039. Its 1,800-second App-Runs Gate #5 completed GREEN with C1–C4 passing, then the live bot supplied the missing workload proof: JefeOS 1.1.2 returned a valid task count through /proc, the replacement Node process remained Gateway READY with all 13 observed heartbeats acknowledged, and it served both the status command and a later LLM reply without a panic or VM reboot. The published ISO is tied to build 20260729-2312 and SHA-256 B6EFBF4EAD7FB4A69A77F2AE2AD489C7509E0DB54A3FC4C57CD6EDB0FB2BC2DF.
Back Toward 1.2
Version 1.1.2 preserves the OpenStack and Xylem progress from 1.1.1; it does not rename unfinished 1.2 work as complete. OpenStack's first-class compatibility milestone is proven, while virtio and static network-data application remain follow-ons. The 1.2 push returns to cross-vessel failover, measured diagnostics and build identity, a cleared HIGH queue, one-command machine-readable release evidence, and independent clean build and boot proof for both kernels. The patch release also leaves a useful rule for that work: a release gate has to cross the same boundary as the workload it claims to protect.