The castle's resident enchanter was a man named Bren. I'd known he existed from years of observation, had seen him at work in the lower corridors maintaining the formations that kept the castle's major functions running. Middle-aged in the way that cultivators in the Breathing stage were middle-aged: he could have been forty or eighty, the cultivation slowing his apparent aging enough that I genuinely couldn't estimate. Breathing stage meant a lifespan of roughly two hundred years; Bren might be a quarter of the way through his, or halfway. The difference mattered less than the fact that he'd outlast every non-cultivating member of the staff he worked alongside, and that the knowledge he'd accumulated across three decades of maintenance was irreplaceable in a way that nobody in the household administration seemed to have noticed. Compact, broad through the shoulders from years of hauling materials up castle stairways, with close-cropped hair going grey at the temples and hands that were calloused in the particular pattern of someone who held inscription tools for hours at a stretch. Efficient in his movements. Perpetually slightly annoyed by the world in the specific way of professionals who spent their working hours cleaning up other people's failures.

  I tracked down where he took his morning tea, which was in a small room off the lower corridor where the castle's material stores were kept, and I arrived there early and waited.

  He came in, stopped, looked at me with the expression of a man who had been doing this long enough that surprise was unavailable to him.

  "Young lord," he said. The flat "young lord" of someone who knows his courtesy formulas.

  "Bren," I said. "I'd like to offer you a trade."

  He poured his tea and looked at me sideways. "A prince offering trades to maintenance staff. Novel."

  "I want access," I said. "Not to anything restricted. I want to watch you work. Ask you questions while you work. Learn from what you're doing."

  "I don't teach," he said.

  "I know. I'm not asking you to teach. I'm asking to follow you and carry materials and help with the manual work in exchange for being allowed to ask questions you'll answer in one or two sentences while you're doing other things."

  He studied me. I'd thought about how to frame this carefully. Bren was a professional who valued his time and didn't particularly value social performance. The right offer was a genuine exchange of labor for access, not a favor-from-a-noble framing that would have made him uncomfortable. In a kingdom where the formal hierarchy had spent centuries crystallizing into elaborate protocol, where there were different forms of address for officials of every rank and cultivators of every stage, the gap between a prince and a maintenance enchanter was supposed to be unbridgeable in one direction and invisible in the other. I wasn't interested in that gap. I was interested in what he knew.

  "What do you actually know?" he asked. Not unkindly. Practically.

  "I've read most of the castle's enchanting texts twice," I said. "I've made practice inscriptions in wood and stone and two metal alloys. I can feel mana flow in inscription channels, not just whether there's flow, but the quality and direction of it. I know the standard warming and illumination patterns from the formation manuals, plus the hardening configurations. I've been watching you do maintenance work in the south corridor for three weeks."

  He stopped with his tea halfway to his mouth. "You've been watching me."

  "From the storage alcove. You don't typically use it during your morning maintenance rounds."

  "Hm," he said, setting his tea down. "What did you observe?"

  "The south corridor's primary warming formation has a failed node in the third junction. The standard repair approach would be to re-inscribe the node, but the failure pattern suggests the problem is upstream: there's a blockage in the mana channel at the second junction that's causing backpressure, and the node failed under the pressure rather than from inscription degradation. Repairing the node without addressing the blockage would produce another failure within six months."

  Bren was quiet for a moment.

  "Where's the blockage?" he asked.

  I described it precisely. The location and channel depth where I'd observed the abnormality, the rough extent. I'd been watching this particular formation failure for two weeks and tracking how the flow pattern deteriorated. From the alcove, what I'd felt was a congestion in the mana field near the second junction, as though the channel had developed a stricture and the mana was pooling on the upstream side, denser than it should have been, while the downstream side thinned to something faint and uneven. The density differential extended several feet from the inscription surface, a pressure gradient as legible to me as a bulge in a water pipe would have been to someone watching with their eyes.

  "You felt all that from the alcove," he said. "Without being in contact with the inscription."

  "The mana backpressure extends several feet from the inscription surface," I said. "If your sensitivity is sufficient you can observe it without direct contact."

  "Your sensitivity."

  "It's high," I said. "I don't entirely know why. It's been developing since early childhood."

  Bren picked up his tea again and drank it in a long contemplative pull. Then he set it down and looked at me with the assessment he'd been performing the whole time now slightly more openly expressed.

  "You start tomorrow," he said. "You carry materials, you clean channels when I direct you, you don't touch active formations without instruction, and you can ask questions as long as they don't interrupt critical work."

  "Understood," I said.

  "If you're wrong about the blockage," he said, "we go with my assessment."

  "Of course," I said.

  We went to look at the south corridor. I wasn't wrong about the blockage.

  He said nothing afterward. But he handed me the maintenance log and told me to record what we'd found and what we'd done.

  I wrote up the blockage, the backpressure gradient, the repair. When I'd finished, he read what I'd written, made one correction to my description of the mana flow depth, and handed it back without comment.

  I didn't know yet if that correction meant he trusted my assessment or was still checking it. Both were possible; with Bren, both could be true simultaneously.

  Working with Bren was an education of a specific type: watching someone who knew their craft completely well and had never once been asked to explain why it worked. He knew what failed, what patterns failure followed, which materials held inscriptions longer. All of it from thirty years of experience; none of it from theory.

  The old wing's stone had a different quality to it, warmer to the touch, its surfaces carrying a faint luminescence in the deeper passages where the formation lighting had never been updated. The enchanters who'd built it had worked from a different depth of understanding than anyone currently practising in the city. The new wing's formations were competent but rougher. The sixty-year-old network Bren spent his days keeping alive sat somewhere between: solidly built, undocumented, the product of three enchanters who'd apparently trusted their own memories more than paper.

  What he didn't have was the underlying principle.

  When I asked him why the second-junction configuration worked better than the alternative, he said "it always has." When I asked why certain materials held inscriptions longer, he said "that's what the tests showed." When I asked why a particular pattern of failure always seemed to start from specific node types, he paused and then said "I don't know, actually. I've always just replaced those nodes first."

  "Have you looked for a pattern in which nodes fail first?"

  "They're the weakest points in the network," he said.

  "Why?"

  A longer pause. "The geometry at those junctions is more complex. More places for imprecision to accumulate."

  "So the failure is a function of inscription complexity at the junction," I said. "Which means the question is whether there's a simpler configuration that achieves the same connectivity with less junction complexity."

  He turned and looked at me with an expression I'd started to recognize as his "this child is saying something outside my framework" expression.

  "Is there?" he asked.

  "I don't know yet," I said. "I've been looking at the castle's formation network as a whole rather than node by node. I think there are structural inefficiencies. But I'd need to understand why the network was built this way before suggesting changes."

  "It was built by whoever was here before me," Bren said. "I maintain. I don't redesign."

  "Who built the original network?"

  "Three enchanters, sixty years ago. None of them left notes."

  "Of course not," I said, not quite managing the exasperation.

  Bren smiled. Brief, but genuine. "Welcome to maintenance," he said. "Every problem is the previous person's undocumented decision."

  Individual cultivators could live for centuries, which made expertise feel durable. Why write down what the master would be alive to explain for another hundred years? But when the master eventually died, or left, or simply stopped sharing, all that accumulated understanding vanished at once. The Formation Guild had existed for two hundred years, preserving formation knowledge across generations. It had dissolved roughly forty years ago, its library scattered and partially burned. Now a single Breathing-stage enchanter was holding together a castle's worth of infrastructure with thirty years of reverse-engineered understanding and professional intuition. If Bren left, whoever came next would start from the same incomplete original notes, rebuilding the same knowledge from scratch. Again.

  The repair I was most proud of in my months with Bren came from exactly this problem: an undocumented decision that had been failing in a non-obvious way for years.

  The castle's heating formations on the upper floors were behaving irregularly: some rooms too warm, others cold, in patterns that didn't correlate with which rooms were used or how recently the formations had been maintained. Bren had been patching individual nodes for two years, but the problem persisted, because the individual nodes weren't the problem.

  I mapped it.

  This took three weeks, during which I walked every room on the upper floors with the castle formation maintenance ledger and my own cipher notebook and extended my mana sensitivity to trace the actual flow through the network. Not what the ledger said should be happening. What was actually happening. The work required a particular kind of attention: settling into each room, letting the ambient noise of the castle quiet until the formation infrastructure became the primary thing I perceived. Stone channels mana with its own character, cooler and steadier than wood, and the formation network ran through the castle's walls like a nervous system threaded through bone. In some sections the flow was clean and purposeful, following the inscribed pathways with the quiet efficiency of a well-maintained channel. In others it stuttered, pooled, eddied against itself in ways that felt like listening to two conversations happening at once in the same room.

  The ledger had an error. Not a maintenance error; an architectural one. Two branches of the heating formation network had been designed by different enchanters working independently, and the junction where they connected had been installed by a third person who hadn't known they were connecting two independently designed systems. The junction was trying to match two different mana rhythms and was instead creating a kind of turbulence. This was irregular mana flow that distributed unpredictably rather than according to design. Two rhythms meeting out of phase. I'd encountered this principle before, in a different context: when two patterns collided without alignment, they didn't blend. They interfered, each undermining the other. The junction was a collision point that nobody had thought to design for.

  What the turbulence felt like, when I extended my awareness into the junction itself, was two currents meeting at cross-purposes. Each branch had its own natural rhythm, determined by the inscription geometry of its nodes; one cycled in a slow, even pulse, the other in a faster, slightly irregular pattern. Where they met, neither could dominate. The mana didn't blend; it collided. The result was a churning at the junction that sent surges down both branches in patterns that bore no relationship to the original design intent. Rooms downstream of the junction received too much mana one hour and too little the next, and no amount of individual node repair could fix what was fundamentally a rhythm mismatch at the source.

  The fix was counterintuitive: adding a buffer node at the junction. Its purpose was to create a deliberate pause in the mana flow at the connection point, allowing the two rhythms to equalize before continuing. Like a shock absorber.

  The principle was straightforward once you saw it. Every inscription pattern had a natural mana rhythm based on its geometry. A buffer node with the right geometry would absorb the incoming mismatched rhythms, hold the mana briefly, and re-release it at the buffer's own natural resonance frequency. The outgoing rhythm would be determined by the buffer's inscription, not by either input. Clean, consistent flow regardless of what came in upstream. The principle was the same as any system that needed to convert messy input into clean output: you didn't fix the variation at its source. You absorbed it at the point of delivery and let the buffer's own properties determine what came out the other side.

  Bren stood in front of my mapping notes and the proposed modification for a long time.

  "You're twelve," he said. This was not entirely a non-sequitur.

  "Yes," I said.

  He didn't ask the follow-up question immediately. Instead he stood in front of my mapping notes, studying the proposed modification, and I watched a particular discomfort work through him that I recognized from another life. I'd seen the brief recalibration of encountering unexpected competence before, from Tova, from Aldren; it settled quickly into acceptance. This was different. Slower. The specific weight of a professional confronting the fact that a problem he'd accepted as permanent had a solution, and the solution, once stated, was not complicated. Sixty years this junction had been failing. Sixty years of node replacements and workaround schedules and the quiet professional acceptance that this was simply what the heating system did.

  He turned away from the notes and looked at the junction itself. Not at me.

  "I should have found this," he said.

  The words wore the shape of self-accusation, though they weren't quite that either. It was the sound of a person recalculating how much of their working life had been spent patching consequences because the cause had never occurred to them. I knew what the comforting response was: the problem was structural; nobody had the conceptual framework; you can't find what the tools to see it don't exist yet. All true. I didn't say it. Bren was not someone who needed comfort from a twelve-year-old. He needed the fix documented properly so that whoever inherited this system wouldn't repeat the same sixty years.

  He turned back to the notes. When he spoke again, his voice had settled back into the professional register.

  "Where did you read about rhythm equalization buffers?"

  "I didn't," I said. "I inferred it from the turbulence pattern. The turbulence is what you'd expect from two non-matching rhythmic flows meeting at a junction. If you put a buffer there, a node that absorbs and re-releases rather than passing through directly, the outgoing rhythm should be determined by the buffer's natural resonance rather than by either input."

  "The buffer's natural resonance," he said.

  "It should have one. Every inscription pattern has a natural mana rhythm based on its geometry. I've been observing it for about a year and I think the buffer node geometry should be..." I gave him the configuration. I'd worked it out from principles and confirmed it through six small-scale tests in my room with pieces of inscribed stone.

  He looked at the configuration. He made the modification.

  The irregular heating resolved within a day.

  Bren stood in the previously-cold room in the north wing and felt the even warmth. His expression shifted into something I hadn't seen from him before: the look of a person who had just learned something that rearranged how they understood their own expertise.

  "Who taught you this?" he asked.

  "No one," I said. "I worked it out from first principles and tested the hypothesis in practice before suggesting it."

  "There's no text about buffer nodes," he said.

  "I know. I checked."

  "Then you developed a new technique," he said.

  "A very small one," I said. "It's probably obvious to anyone thinking about it carefully."

  "If it were obvious," Bren said slowly, "someone would have thought of it in the sixty years this junction has been failing."

  I didn't have a good response to that. I let the silence sit.

  After a moment he said: "You need a real teacher. Not maintenance work. Real enchanting instruction."

  "I know," I said. "The formal system won't provide one at my rank."

  "No," he said. "But there might be someone." He paused. "I worked, when I was young, with a man who was genuinely brilliant. Could not keep a position to save his life, difficult, impossible, burned every relationship he had. But brilliant. His understanding of formation theory was..." He stopped. "Different. Like yours is. Seeing the underlying structure of things rather than the surface technique."

  The hairs on my arms stood up.

  "Who?" I asked, trying to keep my voice even.

  "His name was Haric," Bren said. "He was Anchoring stage when I knew him, probably Shaping by now if he hasn't annoyed someone into putting him in a cell somewhere. I don't know where he is. But if anyone could teach you what you need, it would be him." A pause. "He might also be unteachable. He was... not easy to work with."

  "Is he a good enchanter?" I asked.

  Bren considered this for a moment. "He's the best enchanting theorist I've ever encountered," he said. "His practical work matched his theory. He simply couldn't be trusted in any social situation involving more than three people and a disagreement."

  "I don't need him to be charming," I said. "I need him to understand formation theory at a level nobody else does."

  Bren looked at me with that expression again. "Yes," he said. "I imagine you do."

  He didn't know where Haric was. I filed the name in my cipher notebook with three question marks next to it (no answer yet) and went back to the mapping project.

  I kept turning over what Bren had actually described. A brilliant man who couldn't be trusted in any situation with more than three people and a disagreement. Who was right, consistently, and expressed being right in ways that made the people around him wrong. In public. Every position he'd held, the same pattern. I worked through this with the detached curiosity of someone who'd spent a previous career surrounded by people who said the right thing and meant something else entirely. My father's court ran on that principle: smooth agreement in the room, actual intentions exchanged elsewhere, in corridors, in careful conversations that never quite said what they were about. Haric sounded like the opposite. Someone who said the exact thing he meant and couldn't understand why precision was treated as aggression by the people around him.

  Unauthorized content usage: if you discover this narrative on Amazon, report the violation.

  I found I wasn't bothered by the description. Difficult people, in my experience, at least had the courage of their actual opinions. You could disagree with someone who told you plainly what they thought; you couldn't even begin to disagree with someone who'd already agreed with you in the room and meant something else outside it. Haric sounded like someone whose knowledge was worth the difficulty. The difficulty might even be evidence of the knowledge: you don't alienate three institutions' worth of colleagues by being wrong.

  The name kept surfacing. During maintenance rounds, while I was tracing mana channels, in the middle of writing up documentation. Haric. How do you find a brilliant formation theorist who had burned every professional relationship he'd ever had? The Formation Guild might have had records, but the guild had been dead for thirty years, its members scattered into private practice or retirement, its library distributed into collections nobody could locate. I didn't know. But I kept turning the question over, the way you worried at a knot you couldn't quite see.

  The buffer node technique went into my cipher notebook with full documentation, and then I made three more of them.

  Bren's heating system needed only one; these were tests. Three pieces of stone, each with the buffer node geometry but minor variations in configuration: I wanted to know whether the principle held across different spatial arrangements. If the buffer concept worked because of fundamental mana dynamics rather than some lucky alignment specific to the first application, it should work across variations.

  All three worked. The mana smoothing effect was observable in each, more pronounced in the variant that used the wider geometry I'd tried on the third piece.

  I wrote this up and then stared at the documentation for a while.

  I had invented something. Nothing grand, but Bren had confirmed it wasn't in any text he knew of, and I'd covered enough of the library to be reasonably confident he was right. A buffer node for equalizing mismatched mana rhythms at formation network junctions. My buffer node.

  The question was what to do with it. In my old world I'd have known: patent it, publish it, license it, depending on the context. Here, as best I could tell, technique traveled through relationship. You shared with the people you worked with. They shared further. Knowledge went where people went and stopped when the people stopped.

  That wasn't good enough for something I'd found by examining the fundamental structure of mana dynamics. Someone else would have found it eventually; someone else probably should have found it sixty years ago, when the heating problem was new. The fact that they hadn't wasn't evidence that the knowledge was obscure. It was evidence that nobody had been looking at the underlying structure.

  I rewrote the buffer node documentation that evening in plain text, not cipher: the problem, the principle, the geometry, the three variant test results, the questions I hadn't answered. The version I'd carry in my journal was for me; this version was for someone I'd never meet, starting from scratch.

  If they found it in thirty years, they should be able to replicate the application and pick up the open questions where I'd left them.

  I filed the documentation in the library the following week. Orvel looked at me strangely when I brought in the formal technical paper. He took it, turned it over, read the title, and shelved it in the section that seemed most relevant.

  I watched where he put it. If I ever needed to find it again, I'd have to remember his filing logic, which was not self-evident. The thought occurred to me that a document filed in a place only one person understood was not entirely unlike a document that hadn't been filed at all.

  The maintenance enchanter's grudging acknowledgment of my contribution came in an unexpected form: he began asking me questions.

  Not constantly. Not in a way that departed from his professional style, which remained curt and practical. But occasionally, when he encountered something in his rounds that fit the pattern of the problems I'd been useful for, he would find me and describe the situation and wait.

  The first time this happened I wasn't sure what to do with it. The second time I understood: he was treating me as a resource. Not a student, not a subordinate, not a curiosity, a resource. Someone with a useful capability that he had decided to use.

  I found this, quietly, much more satisfying than being praised.

  The problem he brought me in the third month of our working relationship was a failing illumination formation in the castle's upper gallery, the long corridor that ran along the building's exterior face and served primarily as a connection between the official rooms and the private quarters. The illumination formations in the gallery had been installed thirty years ago and had been maintained sporadically. Two sections had failed entirely; several more were flickering with the irregular pattern that usually preceded failure.

  The standard maintenance approach was node-by-node repair: identify the failed nodes, re-inscribe them, restore function. This worked but produced inconsistent results because the nodes were interacting with each other through the network in ways that weren't fully understood from the outside.

  I spent three days mapping the gallery formation network with the extended sensitivity work I'd been developing. What I found confirmed the suspected pattern: the gallery network was a single connected system, not a set of independent nodes. A failed node didn't just lose its function. It created a gap in the network that redistributed the mana load to adjacent nodes, which then ran at elevated output until they too failed. The failure was propagating. Cascade failure: one overloaded component pushing its burden onto the next, which then overloaded and pushed further, the problem spreading outward because nothing in the system was designed to absorb the excess and stop it at the source.

  The standard repair, replacing the failed nodes one at a time, addressed the immediate failures but ignored the propagation mechanism. The repaired nodes would fail again, sooner than the standard maintenance schedule predicted, because the elevated load from adjacent failures would continue until the whole network was brought back into balance.

  The fix required doing all the repairs simultaneously, or in a specific sequence that restored the network balance rather than letting each repaired node immediately inherit the excess load from its still-failing neighbors.

  I explained this to Bren with the formal documentation I'd learned to produce alongside the analysis. He read it carefully. He was quiet for a moment.

  "Can you do the repair sequence?" he asked.

  "Not alone," I said. "The gallery is long enough that there are nodes on opposite ends that would need to be handled within a short enough time window to be considered simultaneous from the network's perspective. I can do one end; someone else would need to handle the other."

  He considered this. "I can handle the east end," he said.

  We did the repair on a morning when the gallery wasn't in use. Bren at the east end, me at the west, coordinating by signal: a simple mana pulse through the network itself, which I'd worked out was a reliable communication method for coordinating with someone at the other end of a connected formation without physical proximity. The pulse was a brief, deliberate push through the connected channels, distinct enough from the formation's operating mana that the receiver could feel it as a signal rather than background noise. Not complex communication; just a timing mark. Now.

  The network balanced within minutes of the final simultaneous repairs completing. The flickering stopped. The illumination in the gallery achieved the steady, even quality that the formation had been designed to produce thirty years ago and had never quite managed because the original calibration had been slightly off and nobody had corrected it since.

  "That," Bren said, when we met in the middle, "is not how this formation has ever looked."

  "No," I said. "The original calibration was slightly off. The balance repair corrected it as a side effect."

  He looked at the steady light for a moment. The warm amber glow was even along the entire corridor now, the kind of quality I associated with the castle's oldest formations, the ones built by practitioners whose skill had since left the city. The gallery looked the way its designers had intended; you could see it in the spacing of the light, the way the amber caught the stonework at exactly the angle the original architect must have calculated.

  "Write it up," he said.

  "Already done," I said.

  He made the sound that meant something positive and went back to his rounds.

  Writing up the gallery repair had been harder than the buffer node documentation. The repair required two people at opposite ends of a formation network, coordinating by mana pulse, each handling their respective sections within the same narrow time window. How do you document a technique that needs two bodies? I'd written it once as if a single practitioner could manage both ends sequentially: simpler, and technically incomplete. Then I'd scrapped that version and written it again as a collaborative protocol. The coordination method. The pulse signal convention. The timing requirements. The minimum sensitivity threshold for each participant to reliably detect the signal through the network's background noise. The second version took three times as long. It was also the only version that would actually let two people who'd never done this before replicate what Bren and I had managed.

  Nobody had asked me to write it twice. But the method was the part worth preserving; the results were just a corridor with even lighting. An extra evening's work for honest documentation. The sort of cost that only matters if you believe someone will eventually read it.

  I filed that documentation too. Library, third week of the month, Orvel's slightly bewildered acceptance of another formal technical paper from the seventeenth prince.

  Orvel had filed the first paper in a section labeled "Miscellaneous Technical Notes (Uncategorized)." The second one went in the same place. I wasn't sure whether to be offended or grateful.

  On my way back from the library I passed two of the household guard trainees in the corridor outside the administrative wing. They were leaning against the wall in the particular posture of people who had finished their duties early and were deciding what to do about it. Both Breathing stage; I could feel it without trying, the cycling hum of mana through established meridians, steady and unremarkable. They were perhaps sixteen. One of them straightened when he saw me, the automatic recognition of a prince, however minor. The other didn't bother.

  "That's the young one," the first said, not quite quietly enough. "Seventeen, right?"

  "Seventeenth," the other corrected, with the faint emphasis that made the number a verdict rather than a fact. He glanced at me and then away, a quick assessment and dismissal that I recognized from a previous life. The look people gave to someone whose credentials didn't match the room they were standing in. "Still Awakening, I heard."

  They moved on. The whole exchange took four seconds.

  I stood in the corridor for a moment afterward, holding Orvel's filing receipt. Two sixteen-year-olds at Breathing stage, which was perfectly normal, which was exactly the timeline the cultivation manuals described, which was the progression that every competent practitioner in this castle had followed. They weren't wrong. At twelve, I should have transitioned a year ago at minimum; the fact that I hadn't was visible to anyone with basic mana sensitivity, which included every guard trainee in the building. My continued presence at Awakening wasn't a secret. It was simply beneath the threshold of anyone's interest, which produced the same practical effect as a secret but with less dignity.

  The rational response was the one I'd rehearsed: the deep-stage approach was correct, the density I was building would compound across every future stage, the foundation mattered more than the timeline. I believed this. I had documentation. I had Aldren's quiet professional confirmation that my mana density exceeded reference ranges. None of which changed the specific quality of a sixteen-year-old's glance, the fraction of a second in which he'd felt my mana field and found it wanting.

  I put the filing receipt in my notebook and went back to the lower corridors. Bren was waiting with a diagnostic problem in the east wing; he wouldn't care what stage I was. He cared whether I could feel the mana flow through a junction node at twelve feet, which I could, which nobody at Breathing stage in this castle could match. That was the trade. Sensitivity for advancement. Depth for speed. I'd chosen it deliberately and I'd choose it again.

  The corridor was quiet. I walked.

  Marvenn appeared in the documentation room doorway on an evening I'd lost track of.

  I was sitting at the work table with Bren, the small room off the lower corridor that he'd cleared for our documentation sessions. The maintenance ledger was open between us, surrounded by my cipher notebook, three half-finished mapping sheets, and the formal write-up of the east wing's illumination diagnostic tree. We'd been at it for several hours. The kind of work that erased time: Bren describing a diagnostic decision, me writing it down, both of us arguing about whether a particular inference qualified as "confident" or merely "likely."

  I hadn't been to dinner. I'd forgotten about dinner entirely. Bren ate when he felt like it, which was irregularly, and I'd picked up the habit without noticing.

  Marvenn stood in the doorway and took in the scene without comment. The maps. The technical papers. The twelve-year-old prince and the maintenance enchanter sitting shoulder to shoulder at an undersized table in a storage corridor, annotating each other's notes. He looked at all of it with the expression of someone filing information in a place he'd retrieve it from later.

  What he did was set a wrapped bundle on the clear corner of the table. Bread and cheese and a piece of dried fruit; the kind of practical meal the kitchen kept available for staff who worked odd hours.

  "You missed dinner," he said. A statement, flat and observational.

  "I'm in the middle of something," I said, which was true but not really the point.

  "I can see that." He glanced at Bren, who met his gaze and then looked back at the ledger with the careful non-expression of a professional who had just been observed in an unauthorized collaboration with a prince by another, considerably more senior prince. Marvenn's mouth did something that wasn't quite a smile. He looked at me. "Eat at some point."

  He left. The whole exchange took less than a minute.

  Bren waited until the footsteps had receded. "Does he do that often?"

  "Check on me? Sometimes." I unwrapped the bread. It was still warm, which meant he'd gone to the kitchen specifically, not grabbed something on his way past. "He doesn't usually come down to the lower corridors."

  "Hm," Bren said, in the tone he used when cataloging information without commentary.

  I ate while we worked. The bread was good. Marvenn had even included a napkin; not "are you eating" but "are you eating without getting crumbs on the maintenance ledger." Someone in my previous life had done exactly this, brought food to a workspace without a word, and left. I couldn't remember who. The shape of the gesture sat in my chest for a moment before I set it aside and went back to Bren's diagnostic heuristics.

  Months of morning rounds and afternoon sessions taught me something I hadn't expected to learn from maintenance work: what it meant to keep a system running when you didn't fully understand how it worked.

  Three enchanters had built the castle's formation network sixty years ago. Their notes were partial. In places the design logic was opaque: you could see what a formation did but not reconstruct the reasoning behind it. Bren had spent thirty years reverse-engineering what the original designers intended, and he'd gotten far enough to maintain the system well. Not far enough to redesign it.

  "There are sections of the lower network I don't modify," he told me once, during a session where we were checking a string of illumination nodes in the older east wing. "Not because they're fragile. Because I don't understand them well enough to know what modifying them would affect."

  "What do you do when they fail?"

  "Repeat the failure pattern until I understand what I'm looking at," he said. "Wait. Approach carefully. Try the simplest thing first."

  "Has that always worked?"

  "No," he said. "The east wing had a section go dark for three weeks about fifteen years ago. I couldn't figure out what had failed. I replaced every accessible node, rechecked every accessible channel. Nothing helped. Then one morning it came back on its own."

  "Something cleared that you hadn't been able to reach."

  "Something cleared. I don't know what. I have notes on what I observed and what I tried, but the resolution section of the notes just says 'resolved, mechanism unknown.'"

  "Resolved, mechanism unknown." I liked that. It was honest in a way most documentation wasn't.

  But the honest part was also the fragile part. Everything Bren had reconstructed over thirty years lived in his head. If he left or died, the next person would start from the same incomplete original notes, rebuilding the same understanding from scratch.

  "Have you written down what you've reconstructed?" I asked.

  "Some of it," he said. "What I'm confident in. The rest..." He paused. "The rest is inference. I'm not comfortable writing down inference as fact when a future maintenance professional might take it as fact and make decisions based on it."

  "You could mark the confidence level," I said. "Document what you're certain of, what you're fairly confident of, what you're guessing. The guess is still more useful than nothing, as long as the reader knows it's a guess."

  He looked at me. "Where did that come from?"

  "From trying to write up my own observations in a form that would be useful to someone else," I said. "The hardest part is always being precise about what you know versus what you're inferring. But the imprecision has to be legible in the documentation or the documentation misleads."

  "Calibrated confidence notation," Bren said, as if filing a term.

  "Yes," I said. "That's a good name for it."

  He was quiet for a moment. Then: "You should write some of my work up. With me. I'll explain what I know and what I'm inferring and you can document the distinction."

  "I'd like that," I said.

  We started that afternoon, working through the east wing's illumination nodes section by section. Bren narrated; I wrote. "Why did you check that node first?" produced an explanation of a diagnostic heuristic he'd been using for decades, never once articulated until now. "What would you do if this approach didn't work?" produced a full decision tree that existed nowhere in the official records. The details had always been there, in his head. Nobody had asked the right questions.

  Thirty years of hands-on knowledge. In the broader mathematics of this kingdom, thirty years was not a long time. The king had been on the throne for four hundred. Bren's entire career was a brief chapter in the ongoing story of this castle. But those thirty years contained something the king's four centuries did not: the specific, ground-level understanding of how this building's infrastructure actually functioned, day to day, node by node. The formal hierarchy said the king's knowledge mattered more. The formation network disagreed.

  The documentation project became part of our regular work. Not every session; sometimes the maintenance itself was pressing. But regularly enough that the castle's formal records, by the end of the year, were significantly better than they'd been when I started.

  By the end of the year, thirty years of reconstructed knowledge had started making its way onto paper, marked clearly: confirmed, inferred, or guessed. Bren still checked every entry I wrote. He'd cross out the word "certain" and write "very likely" in the margin, which told me more about his epistemology than any conversation we'd had.

  The buffer node I had invented went through several refinements before I had a version stable enough for permanent installation.

  The first attempt had a stability problem: the buffer node's own inscription degraded under the sustained backpressure it was absorbing. I'd traded the original problem for a scheduled maintenance requirement, which wasn't a fix. I went back to the geometry.

  The version I eventually brought to Bren used a two-stage inscription approach: a base inscription done outside the confined space of the junction, then a single connector inscription done in place. Stable under load. Practical to install. I'd tested it on three pieces of stone in my room before touching the actual formation.

  Bren watched me install it in the south corridor's problematic junction. The physical work was careful and slow: I held my inscribing stylus, the custom bronze-alloy piece Garret and I had made together, and carved the connector channel into the junction stone with the focused attention that inscription demanded. The stone's mana grain ran roughly north-south through the corridor wall, and I worked with it rather than across it, feeling the subtle give where the grain aligned with my channel and the slight resistance where I had to cross it at the junction point. The buffer node's base inscription, already prepared on a separate stone piece, slotted into the junction and connected to the network through the newly carved channel. I felt the moment the connection took: a slight drawing sensation as the network's mana found the new pathway and began to flow through the buffer's geometry.

  "How long will it last?" he asked.

  "Longer than the inscription material lasts, which is probably fifteen to twenty years in this position," I said. "The geometry is stable under the backpressure loads I've measured. If the backpressure exceeds what I've measured, if the network's load increases significantly, there'd be a failure, but we'd see warning signs in the adjacent nodes first."

  "What warning signs?"

  I described them. He wrote them down.

  "You're thinking about your maintenance successor," he said.

  "Someone's going to be maintaining this in twenty years," I said. "They should know what to look for."

  He made the specific sound he made when something I said intersected with his professional values in a way he hadn't expected from someone my age.

  The buffer node installed cleanly. The south corridor's backpressure problem resolved. Bren filed the installation documentation (mine, which he'd reviewed and annotated with his own observations) in the castle's formal maintenance records.

  Someone twenty years from now would open that file and find a buffer node they didn't build, with annotations in two handwritings, one of them a twelve-year-old's.

  The buckle was a smaller project. Kever was one of the guards assigned to the lower wing, the quiet one who'd been walking the corridor outside my rooms since before I started working with Bren. I saw him most mornings, checking latches and door hardware as part of his rounds, and I'd noticed months ago the specific way he worked his belt buckle's latch on damp mornings. A small motion, thumb and forefinger, working the mechanism through the resistance that moisture put into iron hardware. Not a complaint; not even an inconvenience he seemed to register. Just the automatic compensation of a professional managing equipment that failed him in small, predictable ways.

  The corrosion was straightforward. Iron hardware in a stone building; condensation on cold mornings; the latch mechanism developing a friction it wasn't designed for. An anti-corrosion inscription was well within what I could do at this point. Simple channel pattern that shed moisture from the latch mechanism, carved into bronze alloy rather than iron so the inscription base wouldn't corrode around the inscription itself. Functional, not elegant. The kind of work I could finish in an evening with Garret's stylus and a spare buckle from the equipment stores.

  I did it on a Tuesday. Borrowed the spare from the supply shelf (not the one on his belt, which would have required a conversation I didn't want to have), inscribed it in my room, tested the moisture resistance by holding it over a bowl of heated water and watching the condensation bead and slide off the latch rather than settling into the mechanism. It worked. The inscription was crude by my own standards, the channeling slightly uneven where I'd crossed the metal's grain, but it would keep the latch clear for years.

  I held it for a moment afterward. Bronze alloy takes chill faster than iron; the buckle was already warm from the desk lamp, and it sat in my palm heavier than it looked. The anti-corrosion channel ran along the inside face of the latch housing, invisible from the outside, which was correct. If you could see it working, the channel was too shallow to hold.

  Four years, I counted it, standing there with the finished buckle in my hand. Before Bren, before the buffer node, before I had much of anything I could point to as evidence of a direction, Kever had been walking the lower wing corridor. He'd been three steps behind me on the morning I came out of Tova's final assessment unable to form a sentence; he'd walked the full loop at a careful distance, asking nothing, and that kind of calibration took actual judgment to get right. He'd navigated us around a blocked street during last winter's festival crowds without being asked to navigate, recognizing something ahead I hadn't seen, and said nothing about it afterward. Not warmth, exactly. Something closer to professional attentiveness applied with precision.

  He had never once complained about the latch. That kind of uncomplaining management, I'd learned from months with Bren, usually meant the person was quietly handling more than they let on.

  I put it back on the equipment shelf before the morning shift. No note. A note would have required explaining why I'd been paying attention to a guard's belt hardware for months, which was a question I didn't particularly want to answer, even to myself.

  The summer I was twelve, something shifted. Not dramatically; nothing about my path had dramatic shifts. But the warm-wood piece at eleven had been a test. The practice pieces after that, tests. The training spheres, a test of production reliability.

  The buffer nodes weren't a test. Neither was the maintenance documentation, or the gallery repair. Those were answers to problems that existed whether or not I was curious about them.

  Sethara was out there beyond the castle walls, a city of a hundred and twenty million people whose mana presence I could feel on clear days as a distant, complex hum. I hadn't been outside yet, but the city texts in the library described a place that was simultaneously the largest concentration of humanity in the kingdom and a study in contrasts: guild-organised craftsmen's quarters alongside districts where enchanted infrastructure simply didn't exist. A hundred and twenty million people. How many of them had ever had their cultivation assessed? How many carried latent potential that nobody had thought to look for?

  Somewhere in the north wing, in the room that had been cold for months, the air was even. The gallery's illumination held steady, amber and precise along the corridor. Bren's diagnostic heuristics were on paper now, marked for confidence level, filed where his successor could find them.

  Three variant test pieces still sat in my room. Haric's name sat in my cipher notebook. The test pieces confirmed a principle; the name confirmed nothing yet.

  The formal request for city access, which I'd submitted through the household administration office two months earlier, came back that week. Not approved; not denied. Filed. The response, delivered by a page who clearly had no idea what the document contained, informed me that my request had been classified under "low-priority petitions from junior household members" and would be reviewed in the normal administrative cycle, which ran quarterly.

  Three to six months, in other words. I'd expected that. The household administration didn't deny requests from princes, even seventeenth princes; it simply processed them at a speed proportional to how much anyone cared about the outcome. Nobody cared about this one. The correct response to a twelve-year-old asking for supervised city access was not refusal but bureaucratic patience, which accomplished the same thing more politely.

  I noted the timeline in my cipher journal and closed the notebook. Three to six months was fine. I had the buffer node refinements to finish, Bren's documentation project to continue, and the mapping notes for the east wing's deeper formation channels still incomplete. The bench, such as it was (borrowed, undersized, shared), was full.