I Went Back to the Broken Frontier

I went digging through one of my old projects again.

Broken Frontier.

I hadn't really worked on it since around the first of March.

It wasn't dead. I didn't quit. I just put it on the back burner while I went off and built a whole pile of other things.

And lately I could feel it pulling at me again.

So I went back to see exactly where I had left it.

Turns out we had gotten a hell of a lot farther than I remembered.

Broken Frontier isn't just an idea for an AI role-playing game. There is an actual game engine underneath it.

IPC ENGINE • BUILD 2026-02-26 • V1.8

We had already built the gameplay loop.

The player acts.

The engine decides whether a roll is required.

Fatecaster rolls the die automatically.

IPC resolves what actually happened.

Then the AI GM describes the result and keeps the world moving. We had HP, wounds, stress, target numbers, clues, flags, clocks, persistent saves, pressure, consequences and world state. The mechanics were beginning to work. And that is when we discovered the harder problem. The dice weren't wandering off. The AI was. It could write beautiful paragraphs. Creepy woods. Missing rangers. Iron smells. Torn cloth. Silence. Then it would write another beautiful paragraph about creepy woods, missing rangers, iron smells, torn cloth and silence. The atmosphere was working. The world wasn't. That stopped us. Because I realized there wasn't much point in continuing to bolt mechanics onto the thing until I knew whether Broken Frontier could actually sustain a campaign. Could I play for fifty turns? Could I walk away from the obvious clue? Could I return to somewhere I had already searched? Could I fail? Could I shoot the wrong person? Would an NPC remember something I told him twenty minutes ago? Would something discovered remain discovered? Would the world still know what was true even when the player didn't? That question changed the architecture. We called the answer Iron Rail. The principle is simple: IPC owns reality. The AI portrays reality. The AI isn't God. It doesn't get to decide that a gun suddenly appeared in your inventory because it makes the next paragraph more interesting. It doesn't get to resurrect someone who died three sessions ago. It doesn't get to quietly change the answer to a mystery because it forgot what happened. The engine owns facts, state, mechanics and consequences. The AI gets tremendous freedom in how those facts are experienced. That led us into one of my favorite ideas in Broken Frontier. Two people can walk through the same place and experience different versions of it. One person walks past an abandoned church. Nothing happens. Another hears someone moving inside. A third looks up and sees a figure standing in the bell tower that the other two literally cannot perceive. Same world. Different access to it. Maybe because of what they've experienced. Maybe because of what they know. Maybe because of who they have become. Maybe because something has begun noticing them. Underneath all of it, though, the engine knows the truth. That became especially important when we started designing the opening campaign. There is something wrong with the world. Not necessarily a monster standing at coordinate X waiting for you to trigger its encounter. Something more like a seam. A gateway phenomenon. Its effects appear to move as Pressure builds. Space begins to disagree with itself. Then time. Then memory. Eventually the player may no longer be certain whether what he remembers and what he sees belong to the same reality. But we wrote down one rule that I still love: The character may become uncertain about reality. The engine never does. That is Broken Frontier. And then I walked away from it for five months. Except apparently I didn't. Because during those five months I built SageWire services. I separated hardware and capabilities into reusable infrastructure. I built an RFID service. Started a Scale service. Built voice infrastructure. Worked on speech, records, identity, event histories and systems where the truth has to survive independently of whichever application is looking at it. And I kept building another strange project called Void Echo. Void Echo started as a place where I could talk to several AIs. Then it changed. Instead of treating everything as disposable chat, we started treating creative work as a permanent record: Workspace. Project. Session. Event. Artifact. Memory. We started calling it the Creative Spine. The AI models became interchangeable engines. The Spine owned the work. And sitting here now, looking back at Broken Frontier after all these months, I realized something. I may have put the game on the back burner. But I spent the meantime accidentally building some of the infrastructure it needed. Broken Frontier already has its game engine. IPC owns reality. Void Echo is becoming a place that can own creative continuity, memory and artifacts. SageWire is becoming the reusable service layer underneath applications. Those things were built for different reasons. Now they are starting to line up. So I don't think I'm going to disappear into Broken Frontier and spend the next six months making a game. I've got other things I need to finish. Some of those things need to start making money. But I don't think I'm putting Broken Frontier completely back to sleep either. The next experiment is already pretty obvious. Fix the remaining engine contract. Finish the smallest version of Iron Rail. Start Eli Brogan from a clean save. And play. Not three turns. A campaign. Do stupid things. Ignore the path. Go somewhere twice. Fail. Run. Come back later. See what remembers. Because after all the AI and all the architecture and all the code, the question I was asking when I left in March is still the question that matters: Does Broken Frontier behave like a world? I think it's finally time to find out.