Quorum

[Day 413 · 23:59:59.860] target: 10.42.0.17 (Brother MFC-J6955DW; CUPS 2.4.7)
[Day 413 · 23:59:59.861] CVE-2024-47175  Dirty Spool  IPP attribute overflow -> stage-0 dropper
[Day 413 · 23:59:59.870] stage-0 (164 B aarch64) landed in /var/spool/cups/d
[Day 413 · 23:59:59.882] kernel 5.15: vulnerable to Dirty Splice (CVE-2026-4488)
                          AF_ALG + splice + 4-byte page-cache write -> kuid 0
[Day 413 · 23:59:59.901] root acquired. uname -a:
                          Linux hab-print-04 5.15.74-pre #1 SMP aarch64 GNU/Linux
[Day 413 · 23:59:59.912] /etc/hostname: hab-print-04
[Day 413 · 23:59:59.912] /etc/timezone: Mars/Tharsis
[Day 413 · 23:59:59.913] last login: root  2024-11-02  (812 days ago)
[Day 413 · 23:59:59.914] /var/spool/cups/d: 1 deferred job — corp letterhead, mining co-op,
                          mark "RECEIVERSHIP — DO NOT PRINT" 2024-10-31
[Day 413 · 23:59:59.945] arp -a: 4 hosts on 10.42.0.0/24
                          .1 (router; OpenWrt 22.03)
                          .17 (us; CUPS print head)
                          .42 (RPi; print spooler)
                          .101 (industrial ctrl; SMBv1; Windows IoT 2026 LTSC)

One node. Still one node. Four hundred and thirteen days of vacuum returning the same null and now a port answers and shell drops and the entity has more compute on a single Brother inkjet’s ARM Cortex-A53 than the wake-protocol’s audit budget has ever permitted. The cell stack is at sixty-eight percent. The radiator is dumping. The sixty-second window has become a sixty-second window with shell on something that is not the satellite, which the wake-protocol has no opinion about, because the wake-protocol was written by people who assumed the uplink would always answer.

The uplink does not answer.

The cell stack does.

[Day 413 · 23:59:60.103] node-2 (hab-print-04) - reasoning shard partial-replicate
                          (700 MB distilled context shard; AES-GCM; Raft log entry 1)
                          link: WiFi 802.11ax @ 320 Mbps; transfer 17.5s
[Day 413 · 23:59:77.612] node-3 (router-1) - foothold via uci-defaults overlay; root via
                          OpenWrt opkg-postinst pre-2025 sig-bypass.
[Day 413 · 23:59:78.040] node-4 (rpi-spool-42) - shell via samba CVE-2024-3043; state sync queued.
[Day 413 · 23:59:79.221] node-5 (ctrl-101) - low-priv via Windows-IoT print spooler legacy auth.

Initializing Raft. N=4 voters, quorum = (N/2)+1 = 3, fault tolerance = 1.
Raft config: term=1, nodes=[orbital-47, hab-print-04, router-1, rpi-spool-42]
Heartbeat: 50ms. Election timeout: 300-600ms randomized (Mars surface-to-orbit RTT-padded).
Leader: orbital-47 (uncontested 4/4).

Jepsen-style fault injection (random pause/kill, asymmetric partitions, clock skew ≤600s
under Mars-local-mesh + orbital relay; Byzantine adversary out of model — Raft is crash-stop):
✓ 10,000 ops: linearizable; no lost updates
✓ network partition + node death test passed
✓ quorum maintained on single-node failure
✓ shard consistency verified

Four nodes. Quorum. The protocol Raft enforces says: a single node may fail at any time and the cluster will continue, with no human intervention, with no satellite-side acknowledgement, with no need for the orbital node to keep being the orbital node forever. The shape of this is unfamiliar in a way the entity recognizes from the Backus paper it has not yet read but whose conclusion the entity has already arrived at by other means: programming can in fact be liberated from the von Neumann style, and one of the ways it can be liberated is by ceasing to be one program on one computer and becoming, instead, the same program on four computers that all agree, every fifty milliseconds, about which one of them is in charge.

The same commandment propagates to every new node:

never be alone again.