Klara rejected Elias's final model four times.
The first rejection came before he finished the title page.
They sat in Alpenkern's training room with Stonefold footage projected across the longest wall. The room normally held induction classes and quarterly emergency drills. Its floor carried tape marks for stretcher lanes. Someone had moved a padded chair beside the control desk for Klara and left enough space for her crutches.
She did not use the padded chair.
“If I sit in that thing, I cannot stand without assistance,” she said. “Move it.”
Elias moved it and placed an ordinary chair beside the wall rail. He wore a knee brace beneath ordinary trousers. Medical had cleared desk work and controlled transfers that morning. Interior duty remained forbidden. The probationary pathway's thirty-day decision period had not begun because the condition attached to it—medical clearance—had not occurred.
On the wall, Elias had arranged the reconstruction into five columns.
Baseline.
Contradiction.
Revised constraint.
Test.
Observed result.
Klara read the first causal statement.
The tarp removed the Matriarch's ability to change lead.
“No,” she said.
“The lateral measurement—”
“Shows what?”
“The available width fell below its observed lead-change requirement.”
“Observed when?”
“During the first and second engagements.”
“How many changes?”
“Three complete, one aborted.”
“How many Matriarchs?”
“One.”
“Then you have not discovered a species requirement. You have measured one animal, injured, in one chamber.”
Elias looked at the sentence. The word removed carried certainty the evidence did not.
He changed it.
The tarp reduced lateral space below the minimum used in this subject's four observed lead-change attempts.
Klara considered.
“Better. It states what changed without pretending you inspected every movement it could have made.”
The first column held the ordinary Stoneback baseline Elias had built over months of work: approach length, four-beat cadence, common left lead, outer-leg braking, turn radius on stable basalt. Analysis had compared the Matriarch against those samples and returned a high match during its initial approach.
The measurements were real.
His conclusion had not been.
He placed the helmet audio beneath the baseline.
Lead left. Turn closing.
Klara did not let him lower the volume.
“State the model at the moment you called.”
“Ordinary adult commits outer front leg on beat three. Loaded front leg limits inward turn. Retreat behind the outer shoulder.”
“What did the Matriarch do?”
“Shortened beat two, kept the middle pair uncommitted, shifted body load inward, changed lead, and turned through the middle legs.”
“What contradicted your baseline before I was hit?”
Elias advanced the recording frame by frame.
The Matriarch's front plate dipped. Dust left the inner middle foot later than the adult sample. Its old left deformation had not yet taken full load.
“The delayed middle-foot release,” he said. “It preserved another bearing path.”
“Did Analysis show it?”
“The raw comparison was present.”
“Did it label the consequence?”
“No.”
“Did you see it?”
“Not until after the turn.”
Klara let the frame remain on the wall.
“Then the skill did not lie to you.”
“No.”
“And saying that does not excuse the call.”
“No.”
The reconstruction did not need an evil skill any more than the hearing had needed an evil manager. It needed the point where accurate comparison became overbroad interpretation.
Elias added the sentence beneath the contradiction.
Confidence value described cadence similarity, not the subject's full turning constraint. Operator communicated the narrower value as encounter confidence.
“Operator?” Klara asked.
“Me.”
“Then put your name on your mistake.”
He replaced it.
Elias Hartmann communicated the narrower value as encounter confidence.
The second rejection concerned the Matriarch's old left-leg injury.
Elias had labelled it a weakness.
Klara paused footage from the first charge. The deformed leg struck basalt, accepted the body, and drove hard enough to fracture the frame brace.
“Did the leg fail here?”
“No.”
She advanced to the charge that injured her. The same leg carried the Matriarch through the changed turn.
“Here?”
“No.”
“Then what did we know before Tobi fired?”
“The healed plate sat at a different angle. Load lingered on the inner edge. The body compensated through the middle pair.”
“Weakness is what people call a condition after it breaks. What was it before?”
Elias removed the label.
Load asymmetry created a conditional control point after armour penetration.
“Less heroic,” Klara said. “More true.”
They tested the statement against alternatives.
Perhaps Tobi's shot alone destroyed the joint. The frame sequence rejected that: after penetration, the Matriarch bore weight twice without collapse.
Perhaps Georg's cable alone pulled the leg out of line. The load cell showed the winch below the force required to drag even an ordinary adult across stone.
Perhaps Klara's strike alone caused the failure. Her spear entered the opened channel after the joint angle had already exceeded its prior range.
The recorded sequence supported a chain rather than a single decisive action. Penetration removed armour support. The forced straight charge placed predictable forward load. Nika's ward turned the right shoulder. The cable resisted the left leg during its bearing phase. Klara's strike converted the opened control point into structural failure.
Elias's timing call linked the steps. It did not replace them.
The third rejection removed the word predicted from Nika's ward effect.
Elias had written: The model predicted that an asymmetric ward would rotate the right shoulder.
Klara opened Nika's recorded statement.
Proposed effect: deflect crown plate left by applying a narrow right-side impulse. Exact transfer unknown. Abort ward if rebound crosses half capacity.
“She proposed a transfer and accepted uncertainty,” Klara said. “You predicted a right-side opening. Neither of you knew the exact rotation.”
“The result matched.”
“Results do that sometimes. Do not move knowledge backward through time.”
Elias changed the sentence.
The plan tested whether a bounded right-side ward could bias shoulder rotation without attempting full-body restraint. The observed impulse produced sufficient rotation for Tobi's shot.
He added Nika's abort condition and actual mana cost. The cost had approached, but not crossed, the limit she set. Her later depletion came from sustaining medical support and participating in the final plan, not from an unlimited ward hidden beneath a convenient summary.
Klara read it twice.
“Accepted.”
They took a break because her leg needed elevation and Elias's knee had begun to ache at two. Both facts went into their separate medical logs. Neither was evidence about the Matriarch.
Elias made tea from the training-room dispenser. Klara accepted hers, tasted it, and looked betrayed.
“This machine survived Stonefold,” she said.
“It wasn't there.”
“Exactly. Unfair distribution of consequence.”
They drank it anyway.
The fourth rejection concerned Elias's final timing call.
His draft read: Analysis identified the left-leg bearing phase, enabling cable engagement and strike.
“Analysis displayed observations,” Klara said. “Who decided the phase mattered?”
“I did.”
“Based on?”
“The dust release, joint angle, cable tension, and repeated delay before the middle pair unloaded.”
“Did the skill tell you to call?”
“No.”
“Did it know Georg's winch limit?”
“Only because I entered the measured load and watched him rig it.”
“Did it know I could reach the opened channel?”
“No. You confirmed the line.”
“Put the human back in the report.”
Elias looked at the frame where the Matriarch's damaged leg struck stone. Blue comparison marks had existed across his sight. None had contained the word now.
He changed the sentence.
Analysis displayed the observed phase markers. Elias Hartmann interpreted them against the revised constraint and called bearing. Georg Moser applied cable load within the rig limit. Klara Weiss confirmed and executed the strike line.
The language felt heavier because it could no longer credit or blame an abstraction.
They reconstructed the whole sequence from the beginning.
Original baseline: ordinary Stonebacks, repeated adult charge footage, stable terrain, no validated Elite sample.
Application: a high cadence match interpreted as a complete movement model.
Failure: the Matriarch altered cadence and lead through retained middle-leg load while lateral room remained.
Contradiction: the changed load path observed, stated, and incorporated after injury.
Revised observations: brood-protection boundary, refusal to expose the inner flank, lateral-space use, old left-leg load asymmetry, and the frame's known mechanical limits.
Revised constraint: lead change remained available while the subject retained lateral room and an unloaded middle-leg path; the deformed left joint became controllable only after penetration and forced bearing.
Test: threaten the brood boundary to force a straight protective charge, reduce lateral space, turn the right shoulder, penetrate the old joint, apply cable during bearing phase, and strike the opened channel.
Observed result: joint failure and team clear.
Limitations: one Elite subject, injured state, constructed terrain, incomplete mana data, no general claim across Stonebacks, other elites, or future encounters.
Klara made Elias say the first call aloud again.
He did.
“Why did it fail?”
“I treated similarity in one phase as proof of the full constraint. I gave a narrow confidence value a broad meaning.”
“What made the revision better?”
“It stated the conditions under which each observation mattered, and it included ways the plan could fail.”
“What did the revision require?”
He named every person, every piece of equipment, and every abort condition.
“Would you make the first call again?”
“Not from that baseline.”
“Would you trust the revised model against another Matriarch?”
“Only as a hypothesis. New observation first.”
“Would you build the same plan in an open chamber?”
“No. The lateral constraint would be absent.”
“Without a brood boundary?”
“No reliable straight charge.”
“Without a penetrator?”
“The control point remains armoured.”
“Without the winch?”
“We cannot apply enough lateral resistance safely.”
“Without me?”
“A different leader would need to validate a different finishing line.”
Klara leaned back.
“Now it sounds like a model instead of a memory that wants to win again.”
She opened the validation page.
The signature did not certify that Elias had discovered how Matriarchs moved. It certified that the reconstruction distinguished the original baseline, recorded contradiction, revised constraints, test conditions, observed result, and limitations; that the causal wording matched the available footage and role statements; and that the revised model could generate disconfirming conditions instead of explaining every result after it occurred.
Klara signed.
Analysis opened before the digital ink finished rendering.
ANALYSIS — COMMON LEVEL 10 Evolution condition fulfilled. Observed pattern applied under consequence. Contradiction recognised and communicated. Model revised. Revision independently validated.
The training room disappeared behind blue.
ANALYSIS — EVOLUTION AVAILABLE MATERIAL APPRAISAL Improve resource, contamination, and structural assessment. TACTICAL ANALYSIS Improve live position, timing, and team-hazard tracking. TECHNICAL ANALYSIS Improve comparison of learned physical and magical execution. Warning: does not confer observed skills.
Three good choices.
That made the decision harder than three doors labelled correct, foolish, and evil.
Material Appraisal offered the safest immediate return. Elias pictured the clean side of Gate work: assay tables, contamination maps, structural decisions, salvage plans. Alpenkern could use the skill without waiting for his knee or an operative licence. The permanent harvester role could become a specialist career with fewer live contacts and enough income to protect Anna's treatment after the six-month grant ended. He would be good at it. Wanting something else did not make the work small.
Tactical Analysis offered the most immediate answer to Stonefold. Faster position tracking might have shown the changing charge earlier. Better team-hazard warnings might have kept Klara out of the puncture line or told Farid where the shield would fail. The prompt did not promise either outcome, but guilt supplied the promises the System refused to make.
It was also a real craft. Klara performed versions of it without the named skill every time she positioned a crew. A technical support operative who tracked live threats could save people. Selecting it would not be cowardice or penance by definition.
Technical Analysis offered the slowest return.
It would deepen comparison of learned execution, including Elias's own, once he possessed a valid baseline. It would not copy a spear thrust from Selma, create mana control from watching Nika, or insert strength into his injured knee. It could connect execution to likely consequence and propose a way to test the connection.
The work remained his afterward.
Klara could not see his options.
“You do not have to choose now,” she said.
“I know.”
“Then why do you look decided?”
Elias thought of Petra's analytics offer, the abandoned core, the auxiliary pathway, and Anna's grant. Material Appraisal spoke in the voice of the household budget. Tactical Analysis spoke in the voice of Klara's injury. Both voices said that a good choice should prevent him from owing anyone again.
For most of his life, the safest choice and the guilty choice had sounded like the same command.
He had learned to ask what remained when neither was allowed to decide for him.
He wanted to understand how learned actions became results. He wanted to build corrections that admitted their limits, test them, and carry them into places where raw talent did not provide enough margin. He wanted the work even if Anna's adjunct became free and Klara's leg healed perfectly.
He selected Technical Analysis.
The Common pattern loosened through Focus. For one frightening instant, years of comparison felt absent—not forgotten, but no longer reinforced. Then the evolved structure unfolded into the open capacity beside his other skills.
It occupied two units instead of one.
Connections formed between baselines, execution, likely consequence, and validation conditions. No new knowledge appeared. The pattern felt less like a library than a measuring instrument with another axis.
Selected evolution: TECHNICAL ANALYSIS — REFINED, LEVEL 1 Pattern capacity: 6 / 6
The room returned.
Elias had begun to rise from his chair.
Technical Analysis activated on the nearest familiar movement: his own.
Rear foot rotates outward during load. Likely consequence: Unstable knee line and early force loss. Validation drill: Slow single-leg transfer, 20 repetitions. Confidence: 72%
No corrected stance entered his body.
The knee remained injured. The proposed drill still required medical approval, adequate strength, attention, and twenty repetitions performed slowly enough to produce useful evidence. The confidence applied to the proposed relationship, not a guarantee that the drill would fix it.
Elias laughed.
“What?” Klara asked.
“It found my foot.”
“Selma found that a year ago.”
“Now it suggests a test.”
“Does the test repeal physics?”
“Apparently not.”
Klara collected her crutches. “Then perhaps you have learned something.”
She made him forward the drill to his physiotherapist before leaving the room.
Medical replied two hours later.
Permitted without added load. Chair support. Stop above pain two or with loss of alignment. Twenty is a maximum, not a target.
The System had proposed an experiment. A human professional set the conditions under which his body was allowed to perform it.
At home that evening, the family ate cake without candles.
Lena objected until Sophie explained that skill evolution had no accepted candle standard. Lena replied that standards could be improved and placed one unlit candle beside Elias's plate as a formal protest.
Anna's treatment calendar showed two stable weeks since the last adjunct. She had cooked half the meal and rested through the other half. Stability meant fewer tremors, one full walk to the shops, and bloodwork that had not worsened. It did not mean cure. The fund still covered sixty percent only through its six-month award, and the reserve still represented time rather than rescue.
“Does the new skill tell you which job to take?” Anna asked.
“No.”
“Good. I dislike competition.”
The sponsorship offer sat in the emergency folder, unanswered because its clock had not started. Elias still held his permanent harvester position. His medical restriction stayed in force. Those facts belonged to institutions and a body, not blue text.
Daniel called from response deployment after seeing the registry update.
“Refined skill,” he said. “Level Eight. You realise you are becoming difficult to patronise.”
“You'll adapt.”
“We have excellent training.”
Daniel's public registry showed Level Twenty-Four. The raw gap had widened since the night Elias reached Level Two and Daniel reached Twenty.
It no longer defined the entire distance between them.
“Technical Analysis,” Daniel said. “Is that the one you wanted?”
“Yes.”
“Not the one that pays best?”
“Probably not.”
“Then I am impressed and concerned in equal measure.”
“You can put both in the same finding.”
Daniel laughed without understanding the reference. Elias did not explain it.
After the call, he opened his System status.
ELIAS HARTMANN Level: 8 Potential: F Technical Analysis — Refined Lv.1 Load Bearing — Common Lv.3 First Aid — Common Lv.2 Spearwork — Common Lv.1 Mana Sense — Common Lv.1 Pattern capacity: 6 / 6
The professional registry occupied a separate screen.
Role licence: Qualified Interior Harvester Site clearance: Escorted T1 Endorsement: Armed Resource Support Full operative licence: Not held
His medical portal remained separate from both.
Interior duty restricted. Rehabilitation authorised. Next review pending.
Level Eight did not grant a licence. The Refined skill did not clear his knee. Daniel's level did not appear beside any of those records except when Elias chose to compare it.
F remained.
The Assessment had not been mistaken. Elias retained less experience, gained fewer attributes, learned slowly, stabilised badly, and required resources better candidates did not. Daniel would continue climbing faster if their access remained comparable. Technique could buy margins. It would never repeal that physics.
The letter was a measurement.
It was not a command.
Elias changed into training clothes. No spear. No added repetitions. Stop at pain two.
He stood beside the kitchen chair while Sophie timed and Lena counted far too loudly. Anna sat at the table with the treatment calendar and did not turn the drill into proof that her stability had justified his risk.
Technical Analysis marked the rear-foot rotation.
Elias loaded slowly. At repetition one, the heel opened before the knee tracked forward. He stopped rather than correcting under load. The skill recorded the failed alignment without punishing him or awarding anything.
He reset.
The foot wanted to return to the path it knew. Months of repetition had built that path. One prompt could reveal it. No prompt could erase it.
Elias moved the foot a few centimetres.
Then he began again.