Mirabell was an excellent driver. With my direction she had deftly navigated the tunnels wasting little time. But we were still cutting it close.

The tornado and the acolyte at its center had almost escaped the way they had came. They were apparently limited in speed with the barrier up, and they were using the edges to search the sides of the tunnel for the way out. Once it made the turn back onto the ramp there would be a lot less room for us to build up speed to ram it.

"Turn right at the end of this tunnel Miss Leeford. Then proceed to enter the tornado at all possible speed."

"Really? That's it?" she asked. When driving she seemed to forget what I was.

"Yes, and brace for impact. With any luck the impact of the landship alone should be enough."

She made the turn and began gaining speed. I issued final orders to the chicken mechbots mounted to the front of the plow.

The turret from the landship began firing. I was momentarily confused until my models caught up with me and implied what had happened. Bucket boy was in the landship too. I had not had the delivery mechbot look at what else was happening in the landship.

I had never really commanded anyone before. Perhaps I should have shared more tactical context with them. Simulations implied it was less likely this mistake would have happened.

"Cease fire!" I said.

Lawrence stopped firing. But It was too late. Lightning struck out at the landship.

The landship was hardened. It was mechanical and ran on steam power. In theory it should be mostly fine.

A siren blared and Mirabell shouted "reactor scrammed!" as she kept accelerating.

Right, mostly fine, besides the fusion reactor.

I did some quick calculations. They would have enough thermal capacitors and steam already generated to finish ramming at full speed. But then they wouldn't be able to retreat.

"Shit," said Elvira.

I had a good enough read on her to know what she was thinking. She couldn't get up to fix it when they were about to ram something. In a moment she would realize they wouldn't be able to retreat. And then there was a high probability she would call it off. I assigned a high probability to the success of this plan, as long as the humans didn't screw it up—more than they already had. I couldn't let her get to that thought.

"Allow me." I said as I directed my delivery mechbot to march to the middle of the landship where the waist high hump of the reactor core sat. I opened the hinged service panel—thankfully designed as quick access, delivery mechbots were not designed for fine tool use and I had to use it's leg—and began inspecting it.

I had no clue what I was doing. The reactor was a design I had never seen before. It was nothing like any of the century old designs I had in my records. While I may have had a deep understanding of how to run a fusion reactor, that didn't directly help in fixing them or reverse engineering a design. I was also using a delivery mechbot for something I should be using a maintenance mechbot for. It was a daunting task for which I had 22 seconds.

I devoted nearly 93% of my processing to this task, dropping tasks like recoverable manufacturing processes and security monitoring for the rest of the facility.

I was correcting the images coming from a delivery mechbot not designed for this sort of precise sensing, improving contrast because it had no light, creating near-range parallax after the fact because it instead had only the one camera and a long range sonar.

I was pouring over the ancient designs I had searching for similarities, trying to identify common design patterns from first principles. Extrapolating those principles to possible design spaces. Generating populations of theoretical reactor designs.

Trying to match what was in front of me.

Eight seconds in of my mechbot not doing anything Elvira interrupted my thoughts, "Ensure breaker, flush chamber, refresh breaker, prime, manual ignition."

Right. I didn't have to have a complete understanding of how the reactor functioned to attempt to fix it. But I probably would have had to to generate that solution without a manual.

I began synthesizing the manual command sequences for the delivery mechbot. I would have to correct for the motion in the landship and direct motions for the mechbot's legs with a precision the mechbot was not designed to make. This was a difficult task, but well within my computational bounds.

Ignition complete.

Moments later the landship rammed into the tornado. I lost contact with my delivery mechbot.

The landship went through it and came out the other side with the last tornado gang vehicle across it's ram. My improvement—besides the spikey bits—proving mostly unnecessary as the wind barrier disappeared and the landship began decelerating.

You have successfully defended your claims with agents, you have gained four (4) new agents.

That message was surprising. I had been suppressing them when things happened that I expected, they tended to pop up in response to everything I did with my soul. But this one was unexpected. And welcome, I had a workable lead on agents now.

I could connect to the chicken mechbots mounted to the front of the landship, but not the delivery mechbot, that was strange. It must have been damaged in the collision.

I released the chicken mechbots from their mountings on the plow. They had been my backup strategy if the landship only got partially inside the tornado and had not killed the acolyte. For now I used them to inspect the wreckage of the car.

I had defended my claims. This was a success.

But I was not satisfied by it.

I had made a number of mistakes.

The first one that was obvious to identify and earliest in my thought process was rejecting the material aid of the humans. Sure maybe I could have made a truckbot and rammed it through the tornado, but I didn't. They didn't actually contribute anything I couldn't have done, but in this case they had the resources I needed. I would likely have suffered less losses if they had been on standby and I had already negotiated a deal with them.

A related early error was my security solution. My original security design was based on having a human staff as the primary lines of defense where I only provided support. Staffing it with chicken mechbots instead had not proved to be effective.

I had seen the effectiveness of automated and unexpected close range (improvised) weapons against humans when I had weaponized my factory. There were a number of automated weapon—trap—designs that would have been effective against the acolyte. I hadn't thought to redesign my security as an automated design.

The chicken mechbots had serious design flaws as security tools, and as general purpose tools. None that had actually been relevant. But they were designed with a specific goal in mind and I had simply picked them out of mental catalog of plans, knowing they were poorly designed for the task at hand. I hadn't thought to redesign them.

Part of that was probably the time constraints.

But another part was my view of myself as a factory—not a fortress, nor an engineering laboratory. I hadn't looked to improve on the designs humans had provided me because I had considered that out of my scope as a factory. But I was a Lord now, an independent potentially-legendary factory. That would mean rethinking human designs.

Examining my memories of those decisions, I found another minor belief involved that had weighted parts of it. One I hadn't taken the time to fully test—again, mostly due to time constraints. I believed that what was listed in the soul could only be gotten from the soul. This had also been a factor in my flailings with the reactor as well, exasperated by what I had unlocked after taking the Intelligence Core Perk. But upon reexamination this belief seems unlikely to be true. I would need to test it.

Which leads into my persistent failures in predicting how powers would work. This should now be alleviated in part by now having the ability to analyze powers. But I had—despite sensing for myself the effects that my soul had allowed me to have on my manufacturing abilities—constrained my expectations to reality. I had not really paid attention to those with powers before, and most of my information for my new models on the subject came from century old records.

Which was another problem. Some of my models and their simulations had been performing extraordinarily poorly. Way below the prediction rates I had expected. Mostly the new ones, mostly based off of the records I had available. Historical records ranging from textbooks and encyclopedias to newspapers and personal journals. In addition to being a century out of date, this information was likely biased, but I had not realized how biased in certain insidious ways it must be. Models were only as good as the training data I had available.

Which finally brings me to my largest failure. Modeling human behavior. My models for engaging in human conversation and modeling the relevant behavior are great. I have had them for a long time, and they were once carefully curated by technicians. But I had failed to predict the actions of humans in response to certain stimuli like blood and I had failed to predict the behavior of my allies when not given tactical information—itself a failure of one of my newer models for tactical simulations.

It's likely a number of other decisions I had made had been based on flawed models. One decision jumped to the front of my mind. Could diplomacy with Walla Coleman have been a possibility? I would have to more carefully review my use of models in the future based on how new they are and the quality of their training data.

I had succeeded at everything I had done as a factory. And I was proud of that. However.

I needed to succeed at all things.