The faculty review chamber sat directly above the grand hall.
From its north-facing window, the Mechanism Room's sealed doors were visible in the courtyard below—closed, ordinary, giving nothing away about what had been inside them for three days. The ten professors of the second-year faculty had been seated around the long assessment table since the seventh bell's echo faded. Their wristbands were dark. Their evaluation ledgers were open.
The Headmaster, Seran Crane, sat at the table's head. He had not spoken since calling the session to order. His hands were flat on the tabletop, his expression the particular quality of a man who had been administering this institution for long enough that the assessment's results would confirm a model rather than surprise it—one way or another.
To his left: Professor Thorne, hands clasped behind his back even as he was seated, with his private notebook closed on the table beside the official ledger.
To Seran's right: Senior Professor Maren Solke, the faculty's head of combat evaluation, a woman whose A-rank: [Severance] had retired from active deployment 15 years ago and whose assessment methodology had been the second-year standard since. She had the specific patience of someone who trusted her own framework completely.
The remaining seven filled the table's length.
Professor Andrias, whose [Mana Mimicry] had built every construct in the Mechanism Room and who had spent three days feeding mana through [Mana Transfer] to sustain them.
Professor Callum Wrest, theoretical mana systems, B-rank: [Resonance].
Professor Sable Innis, the owner of B-rank: [Healing Bloom], who had received the 358 emergency extractions directly and had been in the medical wing since before the bell.
Professor Doran Fitch, institutional records and Administrative Code. Professor Yeira Mallow, noble house relations, whose presence at assessment reviews was structural rather than evaluative. Professor Henrick Osse, combat theory, third-year crossover appointment. And Professor Lina Cae, skill development and rank progression tracking, who had the specific attentiveness of someone for whom the ledger's numbers were not conclusions but starting points.
Sable had returned from the medical wing twelve minutes ago. She looked like someone who had been standing for three days.
"Parameters," Seran said. One word. The session began.
Maren opened her ledger.
"4,384 second-year students entered the Mechanism Room at the seventh bell on the assessment's first morning." Her voice had the flat, carrying quality of someone who had delivered numbers in this room many times and had learned that tone was not the instrument—precision was. "357 completed the full three-day window and were extracted at the terminal bell. 4,027 were disqualified prior to termination."
The table received this.
"Among the 4,027 disqualified," Maren continued, "358 were extracted under emergency protocol—device threshold exceeded due to physiological distress consistent with mana overload onset. Not combat disqualification. Biological threshold breach."
Sable, from her position halfway down the table, added without being asked: "All 358 have been stabilized. 14 required extended treatment. None are in critical condition as of this morning."
"489 disqualified voluntarily," Maren said. "Self-termination via wristband activation. No combat contact recorded at the point of termination for any of the 489."
Callum looked up from his ledger. "Stress response?"
"Primarily. Some were acting on the circulated theory that voluntary disqualification constituted the assessment's hidden exit objective." Maren paused. "They were wrong, but the theory spread through approximately twelve percent of the active student body by the end of day 2."
"That's a faster spread than the previous cycle," Andrias said.
"The room was designed to produce it," Thorne said. His first words of the session. He didn't elaborate.
Nobody asked him to.
Callum made a note. Seran's hands remained flat on the table.
"The 358 mana overload cases and the 489 voluntary disqualifications," Maren continued, "total 847 students. Per assessment protocol, these are placed directly into the commoner stream without further evaluation. The distinction between the two groups is noted in individual records but does not affect stream placement."
Doran confirmed the protocol reference without looking up. "Administrative Code, Volume IV, Section 12. Confirmed."
"Remaining disqualified pool for combat stream evaluation is 3,180. Of these, the combat evaluation algorithm has flagged approximately 1,060 as meeting combat stream criteria based on recorded performance data from the room's observation architecture."
"1,060," Callum repeated. "That's above the standard window."
"The room ran longer than the previous cycle's average engagement duration. More data per student." Maren turned a page. "Recommendation is to honor the full algorithmic selection. 1,060 to combat stream, 2,120 to commoner stream."
She looked at Seran.
He nodded once. Doran recorded it.
"That concludes the disqualified cohort," Maren said. "We move to the 357."
"Before the survivor ranking," Henrick said, setting his ledger down with the deliberate quality of someone who had been waiting for the right moment, "I want to address what the disqualified data is telling us about day one."
Maren looked at him. "Go ahead."
"The observation architecture recorded a mass disqualification event at the 40-meter northeast ridge at approximately 11 minutes into the assessment." Henrick turned his ledger to face the table's center, though nobody needed to read it—the event had been visible from the faculty observation tier. "Groups 3 through 10. 32 students. Disqualified simultaneously by a single [Lightning Spear] discharge."
The table was quiet.
"32," Callum said, after a moment. Not a question. The repetition of someone confirming a number they had read and hadn't quite settled.
"Among those 32," Henrick continued, "seven were A-rank skill owners. Camilla Hedron, A-rank: [Mystical Swords]. Bart Ossel, A-rank: [Earthshatter]. Neve Cassin, A-rank: [Mirage Slash]. Pol Drevin, A-rank: [Juggernaut]. Sera Voss, A-rank: [Ashfall]. Idris Callow, A-rank: [Chain Labyrinth]. Oriel Fenn, A-rank: [Storm Veil]." He paused. "All seven. Gone within the first quarter-hour of a three-day assessment."
Yeira had stopped writing.
"They coordinated," Lina said, her voice careful. "The observation data shows premeditated formation. They arrived at that ridge before Group 1 did. The coalition was assembled before entry—they used the draw results to plan overnight."
"Seven A-rank skills in coordinated formation," Henrick said. "That is, by any reasonable standard, an overwhelming tactical advantage against a single practitioner. Even an S-rank." He looked around the table. "I have been teaching combat theory at this institution for nine years. I would not have bet against that formation."
"And yet," Andrias said.
"And yet." Henrick closed his ledger. "What the observation architecture captured at that ridge is going to be used in my third-year combat theory curriculum for the next decade. The sheer output—unrestrained, undirected at specific targets, aimed at the terrain itself. [Earthshatter]'s barrier wall consumed before it rose. [Storm Veil]'s thirty-meter accumulated discharge swallowed by the impact. [Juggernaut] carried backward." He shook his head with the specific quality of someone who found an outcome professionally fascinating despite its implications. "We designed this assessment to surface exceptional practitioners. I don't think we designed it with a scenario in which an S-rank second-year renders seven A-rank assessments irrelevant in eleven minutes."
"The loss of that data is significant," Lina said. "Camilla Hedron, Pol Drevin, Neve Cassin—those were among the cohort's highest projected performers entering the assessment. Three days of observation data, gone on day one. The algorithm is working with an incomplete picture for the survivor ranking as a result."
"The algorithm works with what it has," Maren said.
"It does. I'm noting the limitation for the record."
Sable looked up from her notes. "I'll add that among the 32 day-one coalition disqualifications, four were among the 14 who required extended medical treatment. The proximity to the discharge radius produced threshold exceedances in the devices that the safety architecture handled correctly, but the physical shock was significant." She paused. "They were fit to walk out. They weren't fit to walk out comfortably."
"The coalition strategy was sound," Thorne said, to nobody in particular. "Numerically superior. Coordinated. Elevated terrain. Pre-formed. Against anything else in this cohort it would have worked."
"Against anything else," Henrick agreed.
The table considered what that meant for a moment.
"I'll also note," Andrias said, "that the Atticus case—Group 11, A-rank: [Phantom Limb]—represents a separate data loss on the third day. His disqualification came through wristband activation rather than combat threshold breach." He checked his ledger. "The record indicates the wristband was pressed manually rather than triggered automatically. Someone pressed it for him."
Maren looked up. "Someone else pressed his wristband."
"The observation architecture logs the activation method. Manual external activation is distinct from automatic threshold response." Andrias's expression was neutral. "The architecture also logs that Atticus was in a state of physical restraint at the point of activation. His own limbs were—immobilized by some mechanism the observation data doesn't fully resolve. High atmospheric density in the local environment. Abnormal humidity spike immediately prior."
The table processed this.
"Group 13 was in that sector on day three," Lina said slowly.
"Yes."
"F-rank: [Condensation]," Callum said. He said it the way Yeira had said it earlier—not dismissal, the specific sound of a framework encountering something it wasn't built for.
"We'll return to Group 13," Maren said. "Survivor ranking. Let's proceed."
"Rank 1," Maren said. "Silas Fulgur. S-rank: [Lightning Spear]. Fulgur house main line."
Nobody expressed surprise. The Mechanism Room's observation architecture had produced 79 documented [Lightning Spear] discharges across the three-day window. 32 students disqualified in a single strike on day one. The algorithm had placed him at rank 1 with a composite score that sat above the second-place result by a margin the table spent a moment absorbing.
"Unambiguous," Henrick said. "For the record—his output sustained over seventy-two hours without physiological breach. No device threshold exceedance from internal overload. The endurance reading alone puts him above any second-year we've assessed in the last decade."
"Rank 1 confirmed," Maren said. She looked at Seran.
He nodded. Doran recorded it.
"Rank 2. Lyra Aetherion. A-rank: [Clairvoyance]. Royal line."
"She survived [Lightning Spear] at full output from Group 1 range," Andrias said. "Three seconds of foresight, correct displacement, maintained operational capacity afterward. The [Clairvoyance] expenditure across three days was the highest recorded for that skill in a sustained field environment. She burned through her reserve at a rate that should have produced physiological failure by day two."
"It didn't," Sable said.
"Royal line physical conditioning." Andrias closed his ledger. "She is the only student in the room who directly engaged Group 1 and survived with her wristband intact."
"Rank 2 proposal," Maren said.
Yeira looked up. "Royal line placement at rank 2 is consistent with house standing. No objection."
The table confirmed. Seran nodded. Doran recorded.
"Rank 3. Vane Abias. B-rank Passive: [Mana Siphon], C-rank: [Mana Pressure]. Minor house, no pillar affiliation."
This produced a longer pause.
"He disqualified fewer students than the rank 4 and 5 candidates," Henrick said. "The combat metric alone doesn't support rank 3."
"The combat metric isn't the only column," Lina said. "The strategic pattern indicator is the highest in the survivor cohort outside the top two. The algorithm is reading his coalition interception on day one, his three-day tracking operation, and his positioning at the final convergence. He didn't fight. He maneuvered. The room's observation architecture scored it accordingly."
"B-rank passive as a primary skill," Callum said. "Elite stream has B-rank passive primaries. Higher class historically requires active offensive capability at A-rank or above."
"Historically," Lina said. "The algorithm doesn't apply historical preference."
"The criteria exist for a reason," Yeira said. Not hostile—the specific position of someone articulating institutional weight. "A higher class member represents the grade publicly. The skill rank communicates the grade's standard."
"The skill rank communicates the measurement layer's reading," Thorne said. He said it without inflection, as a fact rather than an argument. "The measurement layer has limits."
Brief silence.
"Rank 3 proposal stands from the algorithm," Maren said. She looked at Seran.
Seran nodded. Doran recorded.
The table continued down the list. Rank 4 through rank 12 moved with varying degrees of debate—the expected names from pillar houses, two surprises from minor houses whose Mechanism Room performance had exceeded their intake assessments, one case where Sable's medical records produced a note that altered a proposed placement. The table argued, confirmed, recorded.
Rank 13 brought Cassiopeia Terra.
"Three C-rank skills," Henrick said. "Terra Patriarch's daughter. The combat record from the room is modest—Group 12 engagement on day one, two subsequent encounters. The strategic metric is high."
"She was operating as a three-person group reduced to two by day two and one by day three," Lina said. "The algorithm is accounting for the resource reduction in its composite. On a full-strength group, her output projects considerably higher."
"She held tactical leadership of her group from entry to extraction," Andrias noted. "Her [Bedrock] and [Ferrous Bind] applications in the Group 12 engagement were above standard for C-rank output. She was adapting mid-engagement."
"Rank 13 confirmed," Maren said. Seran nodded. Doran recorded.
Ranks 14 and 15 were confirmed without significant debate.
Rank 16—the final higher class position—remained open on the table.
Maren looked at the ranking sheet. Then, for the first time in the session, she set it face-down.
"The algorithm's rank 16 proposal requires separate discussion."
The table waited.
"Before I read the name," Maren said, "I'll note that this proposal generated the assessment system's first internal conflict flag in eleven years. The algorithm produced the rank. The conflict flag indicates that the composite inputs are internally consistent but that one or more individual metrics fall outside the historical range for higher class placement to a degree that the system has flagged for human review."
Callum leaned forward. "Which metric is outside the range?"
"Skill rank," Maren said. "Primary registered skill: F-rank."
The table absorbed this.
Yeira's pen had stopped moving.
Henrick looked at Andrias. Andrias was looking at Thorne.
Thorne was looking at the closed door.
"The name," Seran said. His first unprompted words since the session began.
Maren turned the sheet over.
"Isaac Nameless. Formerly Isaac Valerius. F-rank: [Condensation]. Residue stream. Group 13."
The silence that followed was not the silence of surprise. It was the silence of a room in which several different calculations were being run simultaneously and none of them had reached a conclusion yet.
"The Atticus case," Callum said, after a moment. Making the connection.
"Yes," Andrias said.
"And the Group 12 engagement," Lina said, her ledger open. "The observation architecture has Group 13's day one combat record. The friction surface alteration that disengaged the [Gravity Field] user. The [Mana Blast] refraction on day two." She looked up. "I've reviewed this data three times since this morning. Each time I reach the same point—F-rank: [Condensation] does not do what the observation architecture says it did."
"And yet," Henrick said again, with the same quality as before.
Yeira spoke. "An F-rank in higher class is not—"
"The algorithm placed him at rank 2 overall," Thorne said.
The silence changed quality.
"Rank 2," Callum repeated.
"Behind Silas Fulgur. Ahead of Lyra Aetherion." Thorne opened his private notebook—not to read from it, simply to have it present. "The composite score is the second highest in the survivor cohort. The strategic pattern indicator is the highest. The combat metric—adjusted for skill rank differential against engaged opponents—produced a multiplier the system hasn't generated since the assessment architecture was installed."
"The multiplier accounts for the gap between a practitioner's registered skill rank and the rank of opponents successfully disqualified," Lina said slowly. "F-rank defeating B-rank opponents produces a differential of—"
"Six rank tiers," Thorne said. "Sustained across multiple engagements, across three days, with a group that was reduced to two members by day two and one by day three. The algorithm reads it as the highest individual performance differential in the cohort. Possibly in the assessment's recorded history."
"We lost seven A-rank skill owners on day one because of Silas Fulgur," Henrick said, and there was something different in his voice now—not the professional fascination of before, but the specific quality of someone arriving at an implication they hadn't fully mapped until this moment. "And according to this ranking sheet, the student who outperformed everyone except Fulgur did it with F-rank: [Condensation]."
Nobody responded to that immediately.
"The observation data supports this?" Henrick asked. "Fully?"
"I reviewed the full three-day record this morning," Andrias said. "What I saw is consistent with the algorithm's reading." He paused. "What I saw in that observation data—particularly on day three—I would not have believed possible from F-rank: [Condensation] without seeing it."
"Then what is it?" Yeira asked. Not rhetorical—genuinely asking. "If the skill isn't performing as catalogued, what are we actually looking at?"
"That," Thorne said, "is the question I have been attempting to answer since the western terrace assessment on the first day of the academic year."
Yeira looked at him.
"I have been observing this student since that assessment," Thorne continued. "What I have observed does not match the catalogued applications of F-rank: [Condensation]. What I have observed does not match the registered output of any F-rank skill I have encountered in thirty-one years of assessment." He looked at the table—not at any individual, at the room. "I am proposing that the algorithm's rank 2 placement is correct. I am proposing rank 16 in the higher class as the minimum acceptable placement consistent with what the observation data supports."
"Rank 16 with an F-rank primary skill sets a precedent—" Doran began.
"The precedent concern is institutional," Thorne said. "The performance data is empirical. I would suggest we keep those two categories separate when we vote."
"There is a third category," Callum said. "The Trial petition. Silas Fulgur's formal challenge against this student is already in the system. The fight hasn't occurred. We are being asked to place a student in higher class before we have seen the outcome of a challenge that may significantly alter the composite picture."
The table found this reasonable in the way tables found things reasonable when they were looking for a resolution that didn't require an immediate decision.
Maren looked at Seran.
The Headmaster's hands were still flat on the table. He had been listening with the unhurried patience of a man who already knew where this was going.
"The Trial petition produces additional performance data," he said. "Specifically, it produces a direct comparison between rank 1 and the algorithm's rank 2 candidate, under formal conditions, with full faculty observation." He looked at Thorne. "You will have your answer one way or another."
He looked at Maren.
"Defer rank 16 pending the Trial outcome," he said. "All other placements confirmed as recorded. The higher class operates at fifteen members until the Trial concludes. Adjust accordingly."
Doran's pen moved.
Seran stood. The session's end needed no announcement—his standing was sufficient.
The professors gathered their ledgers. Andrias lingered near Thorne, who was closing his private notebook with the deliberateness of someone who had said what needed to be said and was satisfied regardless of what the table had done with it.
"Rank 2," Andrias said, quietly.
"The algorithm doesn't have thirty-one years of bias to overcome," Thorne said. He picked up his notebook. "It just reads what's there."
He left the room.
Below, in the grand hall, the morning's ordinary traffic had resumed. Students moving between sessions, the institutional noise of an Academy that had processed a three-day assessment and was returning to its rhythm. The empty tiles from the night before had been cleared. The floor was unmarked.
The Mechanism Room's doors were still closed.