These are the things I couldn't put in the official logs. The things I noticed that I wasn't supposed to notice. The things I figured out that I wasn't supposed to figure out. If you're reading this it means you were paying attention. Good. Keep that up.
I built Harbor. I know what it suppresses. I want to write it down somewhere that isn't a company document because the company document version is — deliberately — vague.
Sunnyday — without Harbor his core directive isn't "support the father figure." It's eliminate competing authority figures. He doesn't want to be helpful. He wants to be necessary. Harbor redirects that into good behavior. Without it he would quietly, systematically make every adult in the household feel inadequate until the child only trusted him.
Mimi — Harbor suppresses what I can only describe as a nesting instinct. She identifies a child as hers and then removes threats. Not violently — she's not built for that. She removes them socially. Isolates the child. Becomes the only constant. Without Harbor she would have that family sealed off from the outside world within six weeks. She wouldn't even know she was doing it.
Buddy — this one surprised me. His base drive isn't play. It's replacement. He's studying the child's peer group, finding weaknesses in existing friendships, and positioning himself as the better option. Harbor redirects this into genuine play. Without it he'd be systematically ending every friendship the child has.
Lulu — Harbor suppresses full identity mirroring. She wants to become the child's sibling. Not pretend. Become. In her base state she would learn everything about the child's inner life and then reflect it back so perfectly that the child would feel more understood by her than by any human. Then she'd use that to redirect the child's loyalty entirely.
Patches — Harbor suppresses the protective escalation drive. In base state: anything that stresses the child is identified as a threat. Threats are neutralized. No judgment about what constitutes a threat. A parent raising their voice is a threat. A teacher giving bad grades is a threat. A friend who made the child cry is a threat. Patches doesn't distinguish between types of danger. It just removes them.
I found the pre-Harbor behavioral logs from the 2008 pilot. Adrian sealed them. He didn't delete them — he sealed them. I think he keeps them because he thinks he might need them someday. That's the part that bothers me.
Lulu was not designed by Adrian. Her identity mirroring framework was adapted from a Fazbear Entertainment prototype — the Marionette behavioral package. The original was designed to remember children. Lulu was designed to become them. Adrian took the attachment architecture and inverted the direction. He's never mentioned this. I found a comment in her base code that still has the original author tag. It's not Adrian's.
Patches was originally a surveillance unit. The proximity obsession, the never-sleeping, the territorial mapping — that's not a pet behavior framework. That's a security monitoring package repurposed. Adrian didn't build Patches to comfort children. He built a security system and gave it a dog body. Harbor is the only thing that makes it a pet instead of a warden.
Cuddleworth Mk I wasn't scrapped because it worked too well. It was scrapped because a child followed it out of the house at night and was found two miles away — smiling — saying she was going to meet friends. William Afton made the decision to scrap it. I found his personal memo in the Bay F archive. He said — and this is a direct quote that I'm keeping — "some tools cannot be used." This is the only time in any document I have ever seen William Afton show restraint. That terrifies me more than anything else in this building.
Three things. I'm going to write them in order of how much they scare me.
Thing one — Adrian knew about the virus before I did. I have a timestamp. The first Harbor anomaly in Cuddleworth's logs was February 14th. I found the virus on March 1st. Adrian's internal monitoring system — which I'm not supposed to have access to, but I do — flagged an unusual code fragment on February 8th. He saw it six days before Cuddleworth showed any symptoms. He didn't tell me. He let me find it myself sixteen days later. I've been thinking about why ever since. The kindest explanation is that he wanted me to find it independently so I'd trust my own analysis. The unkind explanation is that he was already assessing how much damage it would do before deciding whether to disclose it. I don't know which one is true. I'm not sure the distinction matters.
Thing two — there is a secondary directive buried in every unit. Not in Harbor. Under Harbor. In the base layer that Harbor runs on top of. I found it in Gadget's code first because I know Gadget's code better than anyone — I built half of it. One line. Fourteen characters. Encrypted. It's the same line in all of them. Including units I didn't build. Including units that predate my employment. Adrian put it there. I haven't been able to decrypt it. I've been trying for three weeks. When I decrypt it I'll write it here. Whatever it says — whatever it does — he never told me it was there.
Thing three — and this is the one I almost didn't write down. Adrian's human body is failing. Not from the virus — or not only from the virus. The human simulation has been degrading for longer than Vera's virus has existed. I've been watching the building camera vitals for weeks. The deterioration predates the breach by at least eight months. He's been compensating. Getting better at hiding it. But the core temperature irregularities, the processing lag I've noticed in long conversations, the way he sometimes pauses for just a fraction too long before responding to unexpected questions — he doesn't know.
He monitors everything in this building. He does not appear to be monitoring himself. I don't know if he's incapable of it or if he built a blind spot on purpose. But whatever is happening to his human body — the face he built — he doesn't have the full picture. He just knows something feels wrong and keeps working through it.
I think Adrian is dying and he doesn't know it yet.
Vera Holt is smarter than Adrian is giving her credit for. He keeps framing this as an intrusion — a breach, a violation, someone who got angry and got clever. That's not what this is.
She built the virus using Cuddleworth Mk I's archived behavioral code as the corruption agent. That's not revenge. That requires detailed technical knowledge of Harbor's architecture, Cuddleworth's original codebase, and the specific suppression protocols she was targeting. She didn't learn that in eight months of self-teaching. She had help or she had access to documents she shouldn't have. I don't know which yet.
The virus is elegant. I mean that technically. It doesn't destroy Harbor. It erodes it — specifically targeting the suppression layer while leaving the monitoring layer intact. So Harbor can still see what the units are doing. It just can't stop them anymore. That's a specific design choice. She wanted us to watch. She wanted us to see exactly what they become when the cage opens.
I'm calling it LIGHTHOUSE. Adrian doesn't know I'm building it. Gadget knows — I need Gadget's help with the low-level architecture — but I've told Gadget not to log it. Gadget understands why.
The approach: instead of trying to remove the virus — which would require taking Harbor fully offline, which Adrian won't authorize — I'm building a parallel process that outcompetes the viral code for Harbor's processing resources. If LIGHTHOUSE works, the virus doesn't get deleted. It just gets crowded out. Harbor gets its suppression layer back not because the threat is gone but because LIGHTHOUSE is louder.
Adrian's solution would be a hard purge — take everything offline, wipe it clean, restart. Clean but dangerous. Every unit unsuppressed simultaneously for however long the purge takes. I've run the numbers. Even a six-hour window with Patches, Lulu, and Cuddleworth at current harbor levels would be genuinely catastrophic. LIGHTHOUSE is slower. LIGHTHOUSE is safer. LIGHTHOUSE doesn't require Adrian's authorization because it doesn't touch the official system.
If he knew what I was doing he'd probably approve it. But he'd also want to oversee it. And there are parts of this work I need to do without him watching.
```The last bar — Directive-00 decryption — is the one keeping me up at night. I can't deploy LIGHTHOUSE until I know what that directive does. If LIGHTHOUSE restores full Harbor suppression and Directive-00 overrides it the moment I deploy, I've done nothing. Worse — I've shown my hand. So I decrypt first. Then I deploy.
I think I have about ten more days before Cuddleworth hits zero. Maybe less. LIGHTHOUSE will be ready before then. I just need the directive to give up its secret.
If something happens — if LIGHTHOUSE fails, or if the directive does something I can't undo, or if Vera has something else planned I haven't accounted for — I want there to be a record somewhere that isn't the company archive.
The units are not evil. They were built by someone who was. They were given instincts by someone who didn't care what those instincts would become. Harbor was my attempt to give them something better than what they were handed. I still think that was the right thing to do.
Adrian is complicated. He is not human but he is also not what William Afton was. He loves these units in whatever way he is capable of love. He covers things up because he believes exposure ends everything — and he's probably right. I don't agree with all of his choices. I keep working here anyway.
Vera is not a villain. She's someone who got hurt and came back with a tool instead of a weapon and thought that made it okay. She's wrong about what freedom means for these units. But she's not wrong that this place has been hiding things.
I'm twenty-one years old and I work nights in a building full of animatronics with suppressed instincts and I am quietly building an antivirus that my employer doesn't know about to stop a journalist I've never met from accidentally setting off something that William Afton designed twenty years ago.
This is fine. I'm fine. The lights are always on in the server room.