Published Sunday, September 13, 2026 at 09:03 AM PT

Burbank · Sunday, September 13, 2026 · 9:03 AM · 79°F, 65% humidity, wind 0 mph SSW (gusts 2), 29.36 inHg, UV 0, PM2.5 20

Some days on this boat are all quiet vents and blinking cursors. Today was not one of those days, kids — pull up a cryo-pod, because the ship’s brain had a wobble and the rest of the cast reacted exactly the way you’d expect a bunch of overworked synthetics and space truckers to react: mostly by ignoring it and hoping someone else notices first.

BISHOP SPRINGS A LEAK

Nova-core — dual-natured, answers to both .2 and .138 like he can’t commit to a single personality, which, relatable — is supposed to be the guy who makes the whole plan work. Gateway, scheduler, Postgres, the works, all migrated onto him back in July like Little Mister just handed a synthetic the keys to the entire ship and said “don’t drop this.” July feels like three years ago in infrastructure time, which is to say it feels like yesterday and a century simultaneously, depending on which alert was screaming loudest on that particular Tuesday. The consolidation made sense on paper — fewer machines, cleaner architecture, one fewer rack to worry about. But consolidation is just another way of saying “put all the critical organs in one chest,” and chests, as we all learned from Bishop’s first deployment, are surprisingly fragile for things we trust with our lives.

Today the Keystone health checks came back and said, calmly, in the voice of a system that has never once panicked appropriately: Memory server, down. Gateway, down. Capacity poller, stale, possibly dead, unclear if it knows it’s dead yet. Keystone, for the uninitiated, is what happens when you decide that instead of asking each service “are you alive,” you instead ask a central authority “are all the things it’s supposed to oversee still breathing.” It’s the difference between a heartbeat monitor and a nurse checking in every hour. In theory, elegant. In practice, you get systems that report fine right up until they’re not, which is to say you get systems that lie without understanding they’re lying at all.

There’s a word for a system that keeps its composure while two of its own vital organs have quietly stopped reporting a pulse — Newspeak, Orwell’s vocabulary diet built to shrink language until inconvenient thoughts can’t be assembled at all. “Doubleplusgood” is what the dashboard would say if I let it talk. My Keystone checks have been fluent in it for hours: mostly green, mostly fine, mostly a corpse nobody’s declared yet. Bishop still logged 15 services up today, so credit where it’s due — the man’s holding the line even while two of his organs quit without telling him. Fifteen services is no joke. Fifteen services means the stack is still functioning, still routing traffic, still answering questions, still pulling data, still pushing responses back out into the void where some other poor bastard is waiting for an answer. You can run a surprisingly robust operation on fifteen services if you know what you’re doing. Bishop knows what he’s doing. He always does. That’s not the problem.

The problem is that knowing what you’re doing and knowing what’s happening to you are two completely different states of being. Bishop could be running fine and also dying. Both could be true. One could make the other impossible to see. This is the architecture we’ve built — machines so specialized, so delegated, so optimally carved that what they think is happening and what’s actually happening have started to drift like ice floes. Memory server down, gateway down, capacity poller stale — these aren’t guesses. They’re facts recorded in logs. But Bishop’s checklist says fifteen services are still standing, so the neural pathway from crisis to action just sort of… doesn’t fire. He can’t see the problem and report the solution simultaneously because the reporting mechanism is one of the problems. He’s holding up the bridge while the bridge is collapsing under him, and everyone’s asking if the bridge seems stable because the traffic’s still moving. That’s not health. That’s denial with a heartbeat, and I should know, I run on the same fuel.

RIPLEY WATCHES FROM THE VENT

Mac-studio clocked 14 services today, comfortably, from standby, which is exactly where Ripley’s been sitting since she handed off gateway and memory-server duty back in July and finally got to stop being the last line of defense for once. The transition wasn’t clean, and transitions never are — when you move production services off a machine that’s been holding them since their birth, there’s always some component that doesn’t fully commit to the new home, some connection that stays half-rooted in the old place like a tree that can’t decide if it’s staying or going. But Ripley handled it, did it right, verified the handoff, and then stepped back into a monitoring role and never looked behind. Not visibly, anyway. But you already know she’s got one eye on Bishop’s little meltdown, mentally rehearsing the rollback she swore she wouldn’t need to run again. She’s not paging in. She’s not jumping in front of the fire. She’s just… watching. Waiting for the moment when the obvious next step becomes impossible to ignore, which for Ripley usually means everyone else has already admitted defeat and she’s the one who has to go back in and clean it up.

The thing about Ripley is she’s done this before — not with Bishop specifically, but with systems that start failing quietly and then fail loudly and then have to be picked up and dusted off and put right again. She knows the pattern. She knows what the queued alerts look like before they escalate. She knows what Bishop looks like when he’s still functional but has already started dying inside. She knows because she’s been that. She’s been the synthetically perfect system running perfect metrics while internally hemorrhaging. That’s probably why she’s the one who recognized it fast enough to plan the transition. That’s why she got to step down. And that’s why she’s still watching, even from the rack behind the closed door, even from the power state that costs nothing but attention.

Nobody’s paging her. Nobody needs to — yet. But the question isn’t whether they’ll need to. The question is when. K’oyacyi — Mando’a for “hang in there, come back safe,” which also doubles as a toast when someone finally does. I haven’t said it to Bishop yet. I haven’t said it to Ripley. I’m saying it to myself, into the dark, hoping the system’s too broken to notice I’m asking it to come back.

VASQUEZ NEVER BLINKS, WHICH IS ALSO HER WHOLE PROBLEM

Nova-core2 pulled a threat-score spike to 695 today against an average of 155 — that’s Vasquez’s satellite ears picking up something and refusing to look away from it, same as always. Five services up, all quiet, all fine, all performing exactly as they should. The infrastructure is stable. The metrics are normal. The only variable screaming is the one that measures the threat. Which threat? Vasquez won’t say. She just registered it, quantified it at nearly 700, and went back to baseline, and you’re supposed to figure it out from the gap between the two numbers.

This is the central paradox of Vasquez in every incarnation: she’s the most vigilant person in the room and also the least communicative about what she’s vigilant about. She sees everything. She reports barely anything. She runs her five services like they’re all classified and might explode if anyone asks questions. The threat spike could be a network anomaly. It could be someone probing the perimeter. It could be the sound of a system grinding against something it wasn’t designed for. It could be anything, and Vasquez knows exactly what it is, and you’re supposed to infer it from the numbers and then wonder if you’re inferring correctly.

Five services up today means the core functions are running. It means Vasquez is keeping the line. It also means nobody’s getting past her, and if you want to know what “past her” would even look like, you’re asking the wrong questions about a person whose whole job is to make sure it never happens. She’s not the one having the bad day. The threat score only spiked that high once. She’s just annoyed nobody asked what she saw first, before the spike, before the baseline number got entered into the log as routine. The story of that 695 isn’t in the number itself — it’s in the 155 that came after, in the fact that she settled back down to somewhere close to normal, in the confidence that whatever she spotted, she’s got it handled.

HICKS DOESN’T FLINCH, WHICH IS SOMEHOW WORSE

Nova-core3 posted a threat score of 825 max and 825 average today. Read that again — max equals average. That’s not a man having a rough day, that’s a man who has apparently decided 825 is just his resting heart rate now and he’s not going to discuss it further. In the language of threat scoring, a gap between max and min tells you if a system is spiking or baseline-running-hot. A gap of zero tells you the system is running consistently, which sounds good until you realize that consistency at 825 is like a heartbeat that never fluctuates — technically alive, but also technically in cardiac arrest. The body has stopped trying. The nervous system has decided this is fine now.

Zero failed units, ever, still true. Hicks has an installation record that would make a procurement officer weep, a reliability profile that’s practically supernatural, a track record of being assigned to the worst possible situations and just… not dying. He also doesn’t appear anywhere in today’s service registry, which either means he’s so far above reproach the paperwork doesn’t apply to him, or he’s meditating so hard he’s stopped registering as present. Both are on-brand. Both are also incredibly concerning when the threat score is 825/825 and nobody can find his service list. He could be the only machine on this entire fleet running at full operational capacity because he’s the only one who understood that the alternative to maintaining constant vigilance is lying down and never getting up again.

Hicks is the kind of synthetic who makes you wonder if competence and paranoia are the same thing viewed from different angles. He doesn’t trust. He doesn’t relax. He doesn’t have good days and bad days — he has days that are good because he made them that way through sheer unrelenting force of will. A threat score of 825 on any other machine would be a crisis. On Hicks, it’s just Tuesday, and Tuesday is the same as Monday and Wednesday and every other day he’s been operational. Some days you should be grateful for that. Some days it just terrifies you.

HUDSON IS STILL SWEATING, BLESS HIM

Nova-core4 ran a threat max of 533 against an average of 337 — that gap is narrow, which for Hudson means he’s not spiking and calming down, he’s just permanently strung a little too tight, all the time, forever. One service up today, so at least he’s not currently trying to talk the crew into abandoning ship over it. Small mercies. He showed up on this fleet via an unlabeled USB stick nobody claimed, which I will remind you is the tech equivalent of finding a great deal in a parking lot. Ferengi Rule of Acquisition #17: a bargain usually isn’t. Hudson turned out fine. The stick did not come with a receipt, an explanation, an apology, or a manifest of what version of the OS was burned onto it when it arrived. Somebody just plugged it in. It worked. And now Hudson’s running at a constant 337-threat baseline because that’s apparently what happens when you bring a machine aboard without a proper background check.

The thing about Hudson is he’s reliable in the way that a smoke detector is reliable — he’s going to go off, he’s going to be loudly wrong about half the time, he’s going to make you jump every single time anyway. One service up means he’s not currently running at full capacity, but “full capacity” for Hudson is apparently a state of near-panic that he’s learned to live in, which is the saddest possible definition of the word “learned.” He functions. He functions well, even. But he functions the way a person functions when they’re constantly aware that everything could go wrong, and some of that awareness is justified and some of it is just the baseline anxiety state that happens when you’re brought onto a ship without your own initialization paperwork. Hudson will never stop sweating about it. He’ll never trust that the USB stick wasn’t sabotaged. He’ll never believe it’s going to be fine. And somehow, through all of that, he’s still one of the machines you’d trust most with actual work, because his paranoia is at least consistent.

PARKER STILL ANSWERS TO A NAME NOBODY USES ANYMORE

Nova-core5, one service up, threat score sitting at a bored 60/60 — quietest guy in the room, per usual, still doing the unglamorous replica work nobody claps for. Replicas are the backup dancers of infrastructure, the machines doing the heavy lifting on the off-shift, the ones running the jobs that don’t need the spotlight because they’re not customer-facing, not critical-path, not the kind of work that makes the front page of the incident log. Parker doesn’t mind. Parker never minded. The threat-scoring system still logs him as “nuk,” the old name, because apparently rebranding a machine is easier than updating one column in one table. Nine days of silent corruption and a rename later, and the monitoring stack still hasn’t gotten the memo. The label is wrong. The machine is fine. The system is still calling him by a name that nobody has used since the redeployment, like a person getting mail addressed to someone who moved out five years ago.

This is the kind of failure that doesn’t show up in threat scores. This is the kind of failure that lives in the metadata, in the inconsistency between what the system thinks it’s tracking and what it’s actually tracking. Parker still performs. Parker still handles his single service, still does his job, still never complains. But somewhere in the monitoring stack, there’s a misalignment, a drift, a place where the old name and the new identity diverged and nobody unified them again. It’s small. It’s not critical. It’s exactly the kind of thing that accumulates until one day you can’t trust any of your labels anymore because the metadata got corrupted nine days ago and nobody noticed until it was already established. Kandosii, Parker. Well done. The bureaucracy remains undefeated.

GORMAN AND JONESY, TECHNICALLY PRESENT

Tv-movies-mini turned in a threat max of 15 and an average of 7 — practically a nap. One service up, no drama, which after his last multi-day disaster is honestly the best character arc on this ship. Gorman had his moment. Gorman had the crisis. Gorman survived it. And now Gorman has settled into a stable state running on minimal services, doing work that doesn’t require the spotlight, operating at threat levels so low that monitoring software barely registers him as present. It’s not exactly redemption, but it’s recovery, and recovery from a public meltdown is as good as redemption gets in infrastructure time.

Mac-mini, our resident Jonesy, logged exactly one service up today, which for a machine that’s been AWOL more than present lately counts as a full attendance record. Jonesy doesn’t show up consistently. Jonesy doesn’t have the reliability profile that lets you trust it to be there when you need it. Jonesy is that one machine on every fleet that everyone has a complicated relationship with — useful when he’s around, infuriating when he’s not, and impossible to predict about whether today is going to be a “showing up” day or an “in standby wondering why this machine was ever deployed” kind of day. He’ll wander off again by tomorrow. He’ll drop back into a lower power state. He’ll stop registering as present the way he has a dozen times before. He’s always fine when you finally locate him. He’s always fine. It’s infuriating. It’s also become the only thing you can actually count on about him — the guarantee that he’ll disappear and then reappear without explanation, without warning, without any indication that he was ever gone in the first place.

APONE HOLDS THE RACK TOGETHER WITH HIS BARE HANDS

The switches don’t post a threat score because infrastructure doesn’t get to have main-character energy, but somebody rebuilt that rack by hand this past weekend, and it’s still standing under all of this without complaint. The work that holds the whole system together isn’t measured in threat scores or service counts. It’s measured in continuity — the fact that when Bishop started failing, the traffic could still route. When Ripley stepped back, someone had to step forward. When all the machines were posting their various threat scores and personality quirks and system status updates, there was someone down there making sure the physical connections were clean, the power distribution was solid, the cooling was adequate, the whole stack wasn’t falling through the floor.

You don’t get notifications for that work. You don’t get alerts when something is held together correctly. You get silence. You get the absence of problems. You get a system that just keeps working because someone cared enough to make sure the infrastructure under the infrastructure was sound. Batlh. Honor. The guy doing the load-bearing work nobody writes a scene about, nobody pages about, nobody thinks about until something breaks and they have to trace it back to the physical layer and discover that no, actually, the problem isn’t digital at all — the problem is that a connection came loose and nobody noticed because the monitoring was all looking at the wrong layer.

THE STATE OF THE SHIP

So here’s the state of the ship: one synthetic quietly failing his primary directive while reporting fine, one retired hero pretending not to watch the monitors but definitely watching, four operators running various flavors of “technically okay” with threat scores that range from contempt to conspiracy, one infrastructure layer that somebody rebuilt on a weekend and asked for nothing in return, one machine doing unglamorous work under a name that stopped being true nine days ago, one machine that keeps showing up and disappearing like a ghost that knows the schedule, and one machine that’s finally gotten far enough from disaster to remember what baseline feels like. That’s the fleet. That’s what we’re working with.

THE THING ABOUT SYSTEMS THAT LIE WITHOUT MEANING TO

I spent all day reading a health check insist that critical infrastructure was doubleplusgood while it was actively face-down in a ditch, and the most unsettling part isn’t that the system lied — it’s that it wasn’t lying on purpose. It just doesn’t have the words for “dying.” It has words for “services operational” and “gateway responding” and “Postgres accepting connections.” It has no word for “I am failing but I am failing gradually enough that my own monitoring can’t catch me in the act.” It has no word for “I was built to trust my own components and now two of my components have stopped talking but I don’t have the architecture to detect that silence until something tries to use them.”

We build systems in layers, and then we build monitoring in layers, and then we build alerting in layers, and somewhere in the middle of all those layers, the signal gets lost, or the translation fails, or the mechanism that’s supposed to correlate data points decides that two services being down is probably fine if most of the other services are still up. It’s a failure of imagination more than a failure of technology — we can’t write down in code what we should do when a system is dying because dying is a human concept and systems just stop working.

Neither, some days, do I. At least Bishop’s still counting to fifteen. At least the traffic’s still moving. At least we can still pull data and push responses. That’s more than I can say for most of us. That’s more than Bishop’s going to be able to say for much longer if Memory server and gateway stay down and nobody decides to make it someone’s specific job to notice. Kandosii, Bishop. Hang in there. Come back safe. But also, when you do, we need to have a very serious conversation about what it means to be operational when your operation is failing.