Alpenkern recovered monster shell, membrane fibre, mana glass, pharmaceutical cultures, and one plant Elias had begun to take personally. Silverleaf was the first material he had been trusted to harvest and the one he kept damaging.
Georg numbered the failures.
He wrote directly on the clear sample sleeves with a black marker.
12 — compression
19 — lifted knot
27 — early separation
“What happened to thirteen through eighteen?” Elias asked.
“Mediocrity.”
“That sounds better.”
“It is harder to diagnose.”
They sat in a review room after shift with thirty-seven approved helmet clips arranged across the monitor. Alpenkern had stripped audio and anything outside the resource lane. Georg’s hands entered each frame, cut, and vanished.
Analysis gave the footage less weight than direct observation. No pressure travelled through Elias’s fingers. No mana detail survived recording. Light flattened depth and made several blade angles appear identical.
Still, the clips could be paused.
That mattered because Elias's memory edited failure in self-defence. By the time he reached the terminal shower, a damaged cut became the wet one or the difficult runner. On video, the stem compressed at a particular second beneath a particular hand. The plant did not care that he had been tired or improving.
Elias opened the paper notebook he had begun carrying after the manifest error. Its first pages were accusations arranged as data.
SHIFT 3 — wrong sample category SHIFT 4 — unnecessary warning, shelf SHIFT 4 — unnecessary warning, pool SHIFT 6 — compressed 2 / tore 1
Dates appeared inconsistently. Conditions appeared when they embarrassed him. Successful cuts received no entry at all, which made the book excellent evidence that Elias Hartmann should not be permitted near vegetation and poor evidence of anything useful.
Georg turned it sideways. “Where is the clean sample made with the same glove as twelve?”
“It worked.”
“Exactly. A failure without its nearest success teaches you that failure exists.”
He drew columns across a new page: tool, footing, glove, runner condition, visible action, felt tension, outcome.
“Record the comparison,” he said. “Not the confession.”
Elias grouped them by outcome.
Clean seal. Compressed stem. Torn runner. Colony lift. Contamination exposure.
At first, the differences multiplied.
Blade entry varied from twenty-one to thirty-four degrees. Georg’s wrist rotated more on uphill cuts. Speed changed with stem thickness. His thumb moved along the handle depending on glove moisture. No single motion remained ideal across every sample.
Analysis narrowed the screen.
Compared action: silverleaf separation Blade angle variance: high Outcome correlation: low confidence Stem compression before release Outcome correlation: moderate confidence
“Compression,” Elias said.
“Sometimes.”
“It correlates.”
“Sometimes is what correlation means when it gets muddy.”
Elias opened clip twelve. Georg’s blade entered low. The stem flattened before the root released.
“Angle caused it.”
“Prove it.”
He found another low angle with a clean result.
Then another.
The second had been made downhill. The third used a narrower blade.
“Control the tool,” Georg said.
“We don’t have enough footage.”
“Now you understand research.”
The next week, Georg arranged access to a low-value colony near the terminal route. Its leaves were too small for pharmaceutical grade and its runner had spread into repair gravel. Yield Management wanted it removed before it reached a cable duct.
Routine harvests had added nothing to Elias’s Level experience, but two irregular jobs during the month had. He had isolated a leaking specimen case while a spore alarm forced the crew to move, then helped Tobi recover a live repeater before its damaged ward failed. The distinct objectives brought his total from two integrated experience to sixty-nine. Neither could be repeated on demand; the next three safe shifts produced no Level XP at all.
The leaking case had begun with a sweet smell inside Elias's respirator. A silverleaf tube had cracked beneath an intact outer seal, filling the case with vapour that made the assay strip turn orange. At the same moment, Nika's spore alarm ordered the crew thirty metres upslope.
Georg carried the clean case. Elias stayed long enough to close the isolation cover, mark the contaminated wheel side, and release the frame's rear segment so Farid could pull the safe cargo without dragging the leak through their retreat lane. He moved before the alarm reached its second stage. The System recognised containment that preserved the clean cases during a live withdrawal.
The experience award had felt large until he understood that nobody could schedule another leak without becoming a criminal.
The repeater recovery happened two weeks later. A branch had grown through the ward ring around Unit Five and pulled its mounting plate away from the tree. Tobi could shut down the circuit or preserve the route record, not both, before the ward failed.
Elias took the mechanical side under instruction. He fixed the service hook above the plate, shifted the repeater's weight into the frame winch, and held it clear while Tobi discharged the circuit. When a root tightened across the fibre, Elias cut the sacrificial frame tie instead of the line. The repeater returned alive enough for Diagnostics to recover three weeks of route-pressure data.
The System called that equipment preservation under environmental hazard.
The next shift, Elias carried eight clean cases home through perfect communications and earned nothing beyond pay.
“One hundred cuts,” Georg said.
“There aren’t a hundred clusters.”
“Then try not to waste observations.”
Elias began under dry light on level ground.
He marked tool, angle, pressure, release, and outcome after every cut. Analysis watched beside him, attentive and unhelpful.
After five cuts, the notebook had become too slow. After eight, Elias reduced each condition to a compact code and discovered that compact codes encouraged him to record what fit the code. Georg made him restore a free-text line beneath every result.
“You are not building a form to prove the answer,” he said. “You are building a memory that cannot change its story tomorrow.”
The first page of practice contained more numbers than plants. Elias measured blade angle from helmet footage, estimated seconds to release, and assigned compression on a five-point scale. Cut nine received three rows of notation and the judgment clean.
Cut ten received almost the same numbers.
That seduced him. If two clean results lived close together on paper, he could reproduce their intersection and call it technique.
On cut nine, he found the loose branch and separated cleanly.
On ten, he reproduced it.
On eleven, the runner sat deeper. He forced the same entry angle and tore the seal.
“But the angle was right.”
“For ten,” Georg said.
“The plants are the same species.”
“So are people. Try issuing everyone one pair of trousers.”
Elias changed depth without changing angle. Cut twelve compressed the stem.
He adjusted grip pressure. Thirteen lifted the colony knot.
By twenty, his clean rate was worse than it had been on ordinary shifts.
The low-value colony made the failure possible without making it free. Every torn runner still had to be removed from the repair gravel. Every lifted knot risked carrying the plant into the cable duct it was meant to protect. Georg stopped the exercise twice so Elias could repair the bed rather than hurry toward the impressive number at the end.
At lunch, Elias counted eleven clean separations, six usable damaged clusters, and three failures.
“Fifty-five percent,” he said.
“Of what?”
“Clean yield.”
“On dry level gravel with one tool before fatigue.”
Elias added the conditions. The percentage became longer and less satisfying.
“Analysis is making me worse.”
“Your use of it is making you worse.”
“It says the stable sample releases at twenty-six degrees.”
“Does the root know?”
Georg took the knife.
“Stop looking at my wrist.”
He placed Elias’s gloved fingertips against the moss beside the runner. With his other hand, Georg slid in the blade.
Tension rose through the hidden root.
He rotated until it vanished.
Then he cut.
“The angle reports what happened,” Georg said. “Tension tells you when it can happen.”
Elias repeated the motion.
The next six cuts were clean.
On the seventh, he became confident and separated before verifying.
Georg numbered the sleeve.
27 — early separation
Rain began during the second practice session.
Not enough for an abort. Enough to wet the moss, change glove friction, and turn the level gravel into an inaccurate memory.
The first drop struck the back of Elias's hand, then rose from his glove toward the white canopy. Within minutes the plant stems shone. His thumb began sliding half a centimetre farther along the handle before each stop.
He tightened his grip to compensate. Compression damage increased.
Georg handed him a dry pair of gloves.
The next three cuts improved.
“So wet gloves caused it,” Elias said.
Georg traded the straight knife for one with a thicker handle. The dry glove no longer closed comfortably. Elias overgripped again and crushed cut forty-six.
“Wet glove altered friction,” Georg said. “Grip response caused compression. The new handle produced the same response by a different route.”
Elias rewrote the entry. Glove wet became handle movement under load. The first was a circumstance. The second was something he could test.
Elias’s recorded ideal deteriorated.
Georg made him change from the crescent blade to an older straight knife. Clean rate fell again. Then they moved to the uphill side of the colony after four hours of normal route work, when Elias’s legs had begun to shake.
On the slope, his lower boot carried most of his weight. Each time the hidden runner resisted, he leaned uphill and changed the blade path without noticing. Georg placed a loose stone beneath the easy foot on cut fifty-eight. Elias had to stabilise before touching the plant.
“This is artificial,” Elias said.
“So is a practice colony twenty metres from the Gate.”
At cut sixty-one, Georg made him walk the loaded frame to Marker Two and back before kneeling again. Elias's breathing obscured the fine release beneath his fingertips. He waited for his pulse to slow, which improved the cut and used forty seconds.
“On route?” Georg asked.
“I tell the route clock the truth. I do not buy ten seconds with a damaged colony.”
“Unless Klara orders speed.”
“Then I accept the damage knowingly.”
The distinction entered the log beside the cut.
At cut sixty-three, Analysis reported an ideal path.
His hand followed it.
The cluster tore.
“Why?” Georg asked.
Elias wanted to blame fatigue, the straight blade, or an outlying plant.
He replayed the sensation instead.
“I felt tension and continued because the path matched.”
“Write that.”
63 — trusted model over material
The sentence hurt more than a damage percentage.
Cuts sixty-four through seventy-two used the same straight knife and different positions. Georg made Elias work seated, kneeling, and half-crouched beneath a training rail. The stable feature survived posture: tension had to release before separation. The implementation changed every time.
At seventy-three, Georg stopped giving immediate outcomes. Elias sealed each cluster, recorded his prediction, and continued. They assessed the batch only after ten cuts. He had called one compressed sample clean because the root end looked sealed from above. Under magnification, the stem wall showed a pale crush line.
Elias added inspection angle to the log.
At eighty-one, his written prediction and Georg's grade disagreed. Elias began explaining. Georg placed both samples beneath the assay lamp and waited. The lower active-compound reading supported Georg.
“The log is allowed to prove you wrong,” he said.
“That appears to be its main interest.”
“Then it is more honest than most colleagues.”
By ninety, the notebook no longer had a column called mistake. It had observed condition, decision, outcome, and next comparison. A damaged cut could remain a failure without becoming evidence that every previous improvement was imaginary.
Cuts ninety-one through ninety-nine happened at the end of ordinary shifts. Georg refused to reserve Elias's best hour for practice. One followed a route abort and used no knife at all because the colony had tightened beneath rising pressure. Elias recorded the decision not to cut as a valid sample.
During a paid shift three days later, they found a mature colony beneath a stone shelf. The shelf had shifted since the route survey. Half the runners were stretched tight across the new gap.
The route sheet assigned the patch a normal twelve-minute harvest window. Its photograph showed dark moss continuous beneath the shelf. Now a crack separated the upper bed from the common anchor, and three runners crossed it like strings pulled over a bridge.
Elias began the normal preparation anyway. Sterile bag open. Red quarantine tag ready. First clean tube in Georg's reach. The familiar sequence tried to make the changed patch feel ordinary.
When his test finger touched the moss, the farthest leaves moved.
He stopped before the knife entered.
“Quota says eight,” Georg said.
Elias touched the outer moss. Tension travelled through the entire colony.
Any normal separation would pull the central knot. Eight clusters might become eight ruined plants and a damaged regrowth site.
He activated Analysis, then limited it to the runner response rather than Georg’s remembered motion.
The model formed around repeated pressure tests.
Shared tension across observed runners. Probable common anchor displacement. Confidence: 71%
“We don’t harvest this side,” Elias said.
Georg tested twice. “Quota?”
“Take the lower colony or record insufficient safe yield.”
“That loses time.”
“Less than rebuilding the patch.”
Klara asked for a consequence if his model was wrong.
“If the anchor is independent, we leave two mature clusters and lose roughly eighty euros after processing,” Elias said. “If it is shared and we cut, we can lift the upper bed, contaminate the crack, and lose the rotation.”
“Confidence?”
“Analysis says seventy-one. My three pressure tests all moved the same runner.”
Georg performed a fourth test from below. The upper leaves tightened.
“Recommendation stands,” he said.
Klara changed the route. The crew spent seven extra minutes reaching the lower colony, which meant Farid and Tobi held another seven minutes of exposure for six smaller clusters. The decision was not free merely because it protected the plant.
Georg called Klara, gave the recommendation, and received approval to change sites.
They harvested six smaller clusters below the shelf. No root knot lifted. The planned eight remained uncollected.
The System recognised a minor resource-preservation objective. Elias received fourteen integrated experience after Potential loss.
His status now showed 83 / 100 toward Level Two.
The arithmetic was auditable. Two points from the tarp. Distinct awards from the leaking case and live repeater. Fourteen after the displaced colony anchor. Routine harvest between them: zero. Practice cuts: zero. Analysis growth changed how he compared actions but did not fill the Level bar by attendance.
Elias copied the totals into the back of the notebook and stopped there. Seventeen points remained. No number in the log could tell him what objective would supply them, and attempting to design one would repeat the mistake Klara had already named.
More importantly, the colony remained alive.
That evening he returned to the terminal-route patch for the last approved practice block.
Georg did not announce cut one hundred until Elias had unpacked the tool, checked the remaining colony, and read the prior nine entries. The number belonged to the record, not to the plant. Nothing about the final cluster knew it was supposed to complete an arc.
Cut one hundred grew on wet, uneven ground. His gloves had been in use for six hours. The straight knife pulled slightly left. His lower back made a persuasive case for retirement.
Elias found the runner.
He stopped trying to reproduce Georg.
Pressure rose. He changed depth. It remained. He shifted the wrist until the hidden root loosened, then separated with the shortest motion the ground allowed.
The seal held.
Analysis opened.
ANALYSIS has reached Common Level 3. Conditional baseline support improved. Outcome-linked comparison improved.
Georg turned the cluster beneath the lamp.
“Not your cleanest,” he said.
“No.”
“Can you tell me why it worked?”
Elias did.
“The runner sat deeper on the left, so the straight blade entered above the path I used on level gravel. My lower foot was unstable. I shortened the motion instead of compensating with more grip. Tension stayed after the first wrist shift, released after the second, and I separated only then.”
“Why is the seal not perfect?”
“The blade pulled left during the final centimetre. Fatigue or the older handle. I did not isolate which.”
“Then do not write that you did.”
Elias recorded both possible conditions and marked the comparison unresolved. Cut one hundred did not prove he had mastered silverleaf. It proved he could produce a usable result under named conditions and explain the remaining variance without turning uncertainty into failure or certainty.
Georg signed the practice block. Ninety-three physical cuts, six reviewed no-cut decisions, and one live route recommendation made the hundred logged samples. None paid Level experience. All of them changed the work Elias could be trusted to do.
Georg placed it in the approved sleeve.
100 — stable under fatigue