The analytics wing smelled like cold air and active display screens, a specific combination that Kai associated immediately with the deep-night hours of very serious work.
He’d passed the building a dozen times without going in.
It required a badge scan even to enter the study floor, and the lights in the corridor were calibrated for screen visibility rather than human comfort, which said something clear about the priorities of whoever had designed the space.
Most of the terminals sat empty.
One near the back glowed with the specific brightness of several screens running simultaneously, and the person in front of them was so still that for a moment Kai read him as part of the furniture.
“Noah Grimm?”
The boy’s eyes didn’t leave the center screen.
“You are loitering in a restricted study floor without a registered research pass,” he said.
The tone was not unfriendly, but it was also not the tone of someone who considered unfriendliness a meaningful distinction from neutrality.
“State your business or leave. Either outcome is acceptable. Idling is inefficient.”
“Rex told me you’d been looking at my lab logs,” Kai said.
“Thought I’d come find out why.”
Now Noah looked at him.
Pale eyes, direct, the kind of direct that assessed without performing assessment — not unfriendly, but without the social wrapper of pretending the assessment wasn’t happening.
He looked at Kai the way Kai looked at machines, which was to say, thoroughly and without embarrassment.
“You took longer to come find me than I expected,” Noah said.
“I would have come to you within another two days if you hadn’t.”
“You would’ve walked up to someone you don’t know to explain why you’d been looking at their private data?”
“It’s not private. It’s shared lab infrastructure. Anything running through the Academy’s diagnostic relay is observable to any student with network access and sufficient pattern recognition to know what to do with it.” He turned back to his screens, apparently considering the matter of how Kai had arrived here settled.
“Your Core’s adaptive response patterns have appeared in my data feed seventeen times across the past eight days. I have flagged each instance. Would you like to know what they have in common?”
Kai pulled up the chair beside him and sat down.
“Yes.”
Noah’s posture didn’t change — he’d been the same calibrated stillness since Kai walked in — but something in his focus shifted slightly, the quality of concentration adjusting toward shared problem rather than solitary analysis. You are reading on g.com, the official source.
He pulled a display to the left screen without looking away from the center one, three-dimensional stress mapping layering over tactical overlay data in a configuration Kai recognized, distantly, as his own lab telemetry.
“Your Core performs predictive correction,” Noah said.
“Not reactive correction. Standard adaptive architecture in modern Frames receives damage input, processes damage data, then executes compensatory adjustment. The response lag is measurable — typically 0.3 to 0.7 seconds depending on processor speed and algorithm complexity. Your Core executes compensatory adjustment before the input has completed registering.” He pointed at a data cluster on the left screen.
“Every instance. Consistent across different stress types, different impact vectors, different operational scenarios. Your intelligence array is not responding to events. She is anticipating them.”
Kai looked at the data.
He’d known something unusual was happening in ARIA’s correction behavior — had sensed it from the way her adjustments felt, that quality of arriving before he’d quite understood they were needed.
He hadn’t had the analytical language for it.
“ARIA,” he said, to the unit on his belt.
“Were you going to tell me about this?”
“I was waiting for you to notice,” ARIA said, audible through the unit’s speaker in the quiet of the analytics wing, “or for someone with the appropriate analytical tools to bring it to your attention. I am pleased Mr. Grimm has done so. It saves us some time.” A pause.
“Mr. Grimm. Your analysis is accurate. I would add a refinement: the prediction is not perfect, and the accuracy degrades over novel input types. I perform considerably better predicting behavior I have observed before than behavior I am encountering for the first time. This is, I would note, a meaningful tactical limitation as well as an asset.”
Noah absorbed this without the reaction most people had to ARIA speaking for the first time — a slight blink, a recalibration, and then immediate, effortless engagement.
“You self-monitor your predictive accuracy,” he said.
“Continuously.”
“And you report the monitoring data to the candidate?”
“When relevant. I try not to flood his attention with continuous status updates during operational situations. Timing the delivery of tactical information is itself a tactical decision.”
“Hm.” Noah’s eyes had gone to a private middle distance for three seconds, the look of someone integrating information.
Then he was back.
“The accuracy degradation over novel input is consistent with Bayesian prediction architecture — you’re updating probability models in real time, which means early-match prediction is weaker than late-match, and opponents you’ve never encountered take more time to model than familiar combat styles.”
“Correct. Mr. Grimm, may I ask how you are familiar with Bayesian prediction architecture? It is not standard curriculum at the first-year level.”
“It’s not standard curriculum at any level in applied Frame science,” Noah said, without apparent pride.
“I read extensively outside the curriculum. Most of the curriculum is designed for students who don’t already know the material it’s attempting to introduce.” He turned to Kai. This chapter is updated first at novelping.com.
“I would like to run a full simulation series against your Core’s telemetry data. Not publicly — I do not want to flag this data for faculty review before I understand it properly myself. I am asking for your voluntary access.”
“Why are you doing this?” Kai asked.
Not suspiciously — genuinely curious.
“You spent eight days analyzing data on someone you’d never met. Why bother?”
Noah looked at him for a moment, and something in his expression moved — a very small shift, the kind that on a more expressive face would have been a tell.
“Most data I encounter here is predictable,” he said.
“Every Frame system I’ve analyzed in the candidate field performs within known parameters. The highest-rated incoming students are outliers by a few standard deviations, but they are outliers from a known distribution. Your Core is not performing within any known distribution I have access to.” He paused.
“I find genuine anomalies interesting. I do not encounter them often. When I do, I prefer to understand them rather than simply note their existence.”
“I appreciate that approach,” ARIA said.
“I find I am most comfortable around people who prefer understanding to simple categorization. It is, in my experience, a better starting point for both of us.”
Noah looked at the unit again, and this time the almost-expression that crossed his face was unmistakably the expression of someone deciding they’ve met something worth paying attention to.
“Full access,” he said to Kai.
“In exchange for which I will share every analysis I produce, including the pieces that don’t favor your tactical situation. No filtering.”
“No filtering,” Kai agreed.
“I would add,” ARIA said, “that the reverse transparency applies as well, Mr. Grimm. If I observe tactical patterns in your analytical approach that I consider strategically significant, I will note them to you directly rather than simply feeding information to my Builder while you work. Is that acceptable?”
Noah considered it for approximately one second.
“Yes. That is more efficient than one-directional information flow.”
“Good.” The word carried something that was almost, but not quite, warmth — more like satisfaction, which was ARIA’s version of the same thing.
“Then I suggest we begin with the match history from the Class Cup scrimmage data. There are six incoming candidates whose tactical styles I’ve had insufficient time to model. If you can help me correct for that, we will be considerably better prepared for the bracket.”
Noah was already pulling up a new display.
“I have been compiling that analysis for eleven days,” he said.
“Excellent,” ARIA said.
“I suspected as much.”