Sillglass Galleries punished confidence with reflections.
The Gate opened into a cavern of translucent stone. Crystal curtains descended from a ceiling lost in pale fog. Every lamp produced six false corridors, four duplicate crew members, and one convincing image of the extraction frame rolling where it was not.
Elias disliked the site immediately.
Sillglass required different boots, different lamps, and an additional signature on every pharmaceutical case. The issued soles contained soft ceramic pads that gripped crystal without concentrating weight on one brittle edge. Helmet lights used broad amber panels instead of narrow beams; setting four and above could heat reflective seams enough to change their stress. Every case carried an internal dark sleeve so a technician opening the lid would not flood the sample with six reflected lamps.
Even the extraction frame had changed. Its usual metal rails wore padded covers. The wheels were wider and nearly smooth. A Hachinger frame announced a bad line by catching on roots. This one could roll perfectly across a surface that was about to fracture.
At the terminal briefing, Dr. Vogl had shown topology notices from the previous month. Sillglass did not rebuild itself like a puzzle during ordinary operation. Corridors flexed by centimetres, crystal curtains thickened, and minor seams closed or reopened after mana tides. Survey overlays updated twice daily. Any change larger than the tolerance suspended the affected route.
“The phrase affected route assumes we know which route the change affected,” Georg had said.
Dr. Vogl had not laughed. “That is why you still touch markers.”
“Beautiful,” Lena had said when he showed her an approved photograph.
Beauty had never needed to read a map.
Nor did beauty need to package anticoagulant resin before it polymerised. The bladders beneath Sillglass crystal supplied precursor compounds used in clot-control medication and surgical sealants. A ruptured membrane could turn a high-grade dose into clear stone across a worker's glove in seconds. Hachinger's silverleaf forgave small delays after a clean cut. Sillglass resin recorded them as chemistry.
Klara stopped the crew inside the acclimation ring.
“Sillglass rules. No unsupported crystal contact. No directed light above setting three. Verify physical markers by touch and code. Overlay is advisory after any topology notice. Resource team?”
Georg said, “No echo crystal extraction without assay confirmation.”
Elias added, “No force on pharmaceutical seams. Case lids closed before lamp movement.”
“Tobi?”
“Never fire at a reflection unless Farid wants a new ventilation shaft.”
Farid tapped his shield. “I do not.”
They moved.
Physical markers sat inside opaque stone plates fixed at knee height. Each carried three forms of identity: raised notches for touch, a printed code beneath nonreflective film, and an embedded tag for the visor. Tobi touched and called the notches. Farid repeated the printed code. Elias confirmed the route sheet. Only then did the overlay arrow earn the right to remain visible.
At the first junction, six reflected arrows pointed into four corridors. The real arrow was the one that disappeared when Elias covered the marker film with his glove.
Hachinger Hollow had become familiar enough that Elias’s body anticipated roots and wet slopes. Sillglass offered smooth floors interrupted by edges sharp enough to cut boot soles. Air changed temperature within a single step. Footsteps arrived from wrong directions.
His first extraction was a pale resin bladder growing beneath a crystal lip. He used Hachinger’s runner technique and ruptured the outer membrane.
Clear fluid hardened across his glove.
Georg neutralised it with a spray.
“Different site,” he said.
“I noticed.”
“After contact.”
The neutralising spray turned the resin cloudy and brittle. Georg peeled it from the glove in one piece, sealed it as industrial downgrade, and inspected the suit beneath for hardening. No exposure. No injury. One pharmaceutical sample destroyed before the route clock reached fifteen minutes.
“At Hachinger, tension tells you when the runner releases,” Georg said. “Here the bladder does not release. You support it while the seam closes behind the draw.”
He placed two curved paddles around the next sample. One carried the bladder's weight. The other opened the seam with pressure spread across its full edge. Georg drew the membrane sideways, never toward himself, and held it for three breaths while the crystal lip whitened behind it.
The sealed bladder went into a dark sleeve before the lamp moved.
Elias's second attempt supported too low. The upper membrane stretched, leaving a pale line that Processing would have to inspect. The third closed cleanly but took four minutes against Georg's ninety seconds. On the fourth, Elias followed the visible seam and discovered it was a reflection of the real one half a centimetre deeper. His paddle struck bare crystal with a chime.
“Stop looking for the brightest edge,” Georg said.
“I know.”
“Your paddle says otherwise.”
Elias turned the lamp down until the false seam faded. The real edge remained as a shadow beneath the bladder.
Analysis possessed no useful baseline. The bladder’s surface produced unfamiliar tension. Georg’s tool angle changed to avoid crystal reflections Elias could not reliably locate.
By the third sample, Elias had downgraded two and taken twice the planned time.
The regression felt personal despite being predictable.
His qualified badge made it worse. At Hachinger, he could sign a collection plan, grade a provisional sample, and explain most failures before Georg asked. Here, Georg handled every released bladder and checked every sleeve. The role had transferred. Competence had not.
Klara removed the last two samples from the planned shelf to recover route time. Nobody accused Elias of costing yield. The empty case positions said it without hostility.
Dr. Vogl had priced each intact bladder during the briefing because Sillglass accounting treated yield loss as diagnostic information. Elias's four attempts produced one pharmaceutical-grade unit, one inspection hold, and two industrial downgrades. Georg's demonstration had produced a clean unit in ninety seconds.
“Do I keep collecting?” Elias asked.
Georg looked to Klara. The question belonged to productivity, competence, and route time at once.
“One more under direct supervision,” Klara said. “Then you carry and observe.”
The fifth bladder grew behind a crystal fold too narrow for the standard support paddle. Elias selected the small curved pair, checked the shadow instead of the bright edge, and distributed pressure along the seam. The membrane shifted without releasing. He held it while the white closure line travelled behind the draw.
At the second breath, a reflected line made the seam appear complete. He almost withdrew the lower paddle.
Georg said nothing.
Elias watched the actual shadow. The closure reached the end on the third breath. He sleeved the bladder and sealed the case before the lamp moved.
“Grade?” Georg asked.
“Provisional pharmaceutical. Upper membrane stretched less than tolerance, no visible clouding, closure continuous.”
Georg checked. “Agreed.”
One clean extraction did not restore competence. It established a beginning. Klara still removed the final samples. Elias took the frame handle and left the tools with Georg.
As they moved, a cluster of Glasswings lifted behind a crystal curtain. They resembled fragments of the wall until thin bodies changed angle and scattered amber light. Farid stopped without raising the shield. Tobi covered his lamp. Everyone waited while the insects settled on a warmer seam.
At Hachinger, movement often announced a route hazard. In Sillglass, an unnecessary response could create one. Elias added that difference to the growing list of things his old confidence did not know.
At Marker SG-14, the augmented map instructed them to turn east toward the pharmaceutical shelf.
The route brief called it Gallery Fourteen: east to the main shelf, west to a maintenance loop that rejoined near the Gate anchor. The western passage showed amber in the overlay because it had not been surveyed after a minor crystal fall. East remained green.
The physical marker on the wall showed the correct three-notch code. Tobi touched it, verified material density, and confirmed the route.
Elias followed.
Ten minutes later, they passed an old harvest cut.
The crescent mark in the crystal faced away from them.
He stopped the frame.
“Marker,” he said.
Georg looked. “Old resin extraction.”
“Cut opens west.”
“So?”
“Standard approach keeps the waste edge toward the Gate.”
“Usually.”
Georg explained the exception. Before Alpenkern standardised approach marks, a harvester might work from either side. A cut older than the concession could face any direction. Crystal regrowth blurred age, and the gallery carried scars from at least three operators.
The mark was evidence. Weak evidence.
The overlay arrow pointed east.
Elias checked airflow with a ribbon from the frame kit. It drifted toward them, cool and steady.
He tested it at ankle, waist, and shoulder height. The lowest ribbon pulled west. The upper two hung nearly still. Local cooling from a crystal wall could create the same layer. He marked the result instead of promoting it.
“Vent route should draw toward the Gate anchor,” he said.
Tobi checked his instrument. “Map says anchor east. Flow says west.”
“Local convection,” Georg said. “Crystal temperature changes it.”
Reasonable.
They continued.
At the next physical node, the route code advanced correctly from SG-14 to SG-15. The embedded tag agreed. Fibre latency to the Gate increased by the amount the map predicted. If the corridor was wrong, it was wrong while preserving several things a wrong corridor should break.
At the next junction, dust lay thick in the overlay’s marked exit corridor. The side route showed recent scuffs.
Sillglass dust formed from microscopic surface fractures and settled slowly in the still air. The eastern floor carried an unbroken pale layer except for the crew's own boot marks. The western side route held three older impressions with Alpenkern's rectangular heel code and a narrow track from a padded frame wheel.
Elias crouched without touching them. “Traffic moved west after the dust settled.”
“Age?” Klara asked.
“Unknown. Edges are still sharp. No rising moisture here to compare.”
“Maintenance traffic?” Elias asked.
Tobi checked the route log. “None recorded this week.”
Klara halted.
“Physical marker?”
Farid found it beside the east corridor. Correct code. Correct material. Correct orientation.
He struck the opaque backing once with his gauntlet. The sound returned solid. No plate had been removed and refixed over a false corridor. Tobi scanned the anchor bolts. Their tamper resin remained unbroken.
Elias activated Analysis.
The skill compared dust disturbance, airflow, harvest cuts, and route marks. Each observation held high local confidence.
Together, they made no sense.
Navigation marker alignment: consistent with overlay. Airflow direction: inconsistent with expected anchor vector. Confidence: 82% Traffic trace: inconsistent with marked primary route. Confidence: 76% Historical harvest orientation: probable inconsistency. Confidence: 68%
“The map can’t be right,” Elias said.
“Or?” Klara asked.
He forced himself to provide alternatives.
“Local airflow. Unrecorded maintenance. Old harvest practice. Marker error.”
“Four separate explanations,” Tobi said.
“Or one map problem.”
They tested the marker. Its embedded tag returned the correct coordinate. Tobi polled the last fibre node. The overlay checksum passed.
Every system agreed with itself.
That was different from agreeing with the gallery.
Elias knelt to inspect the dust and forgot that Analysis had narrowed his peripheral vision.
His boot caught a transparent ridge.
He fell against the frame. One empty case struck crystal with a hollow chime.
The case lid was closed. Its padded corner absorbed the impact. The sound still travelled into corridors they could not see, a clean note repeated from six directions until Elias could not identify the original.
Farid opened his shield toward the eastern darkness. Tobi lowered his lamp to prevent a reflected muzzle line. Nika checked the local mana rise produced by the strike. Georg kept one hand on the frame so Elias's fall did not become a roll.
Elias had created a contact concern because he spent too much attention proving one.
Everyone froze.
The sound repeated through the gallery, diminishing slowly.
Nothing answered.
Georg pulled Elias upright.
“Rest,” he said.
“I’m fine.”
“You analysed a floor until you stopped seeing it.”
Klara ordered five minutes.
Georg removed Analysis from the decision by making Elias close his eyes for the first minute. Nika tested orientation, pupil response, nausea, and the pressure behind his forehead. Focus sharpening from Level Four let him sustain comparison longer; it also let him exclude the transparent ridge until his body found it.
“Skill duration?” she asked.
“Nine minutes, then another six after the marker test.”
“Without release?”
“Short release between.”
“How short?”
Elias checked the helmet log. “Eleven seconds.”
Georg said, “That is a blink, not a rest.”
Klara imposed a three-minute minimum between navigation comparisons and assigned Georg to call the release. Qualified harvester or not, Elias no longer owned the timing of the tool he had just misused.
Elias sat with his back against opaque stone while Nika checked his pupils. Embarrassment returned the gallery to full resolution.
“How many contradictions remain without Analysis?” Klara asked.
“Airflow. Dust. Harvest cut.”
“Good. The skill is evidence, not eyesight.”
During the rest, Georg found a second old extraction mark. Tobi released a thread wisp into the air. Farid walked twenty metres down the side corridor and found three recent boot scuffs.
Nika placed temperature tabs along both walls. The westward airflow persisted after the nearer crystal face warmed, weakening the local-convection explanation. Georg's second cut faced the same direction as the first and used Alpenkern's standard crescent, making age less uncertain. Tobi matched one boot scuff to the service-sole pattern issued at the terminal.
No observation alone overruled a verified marker. Together they made continuing east irresponsible without a test.
Klara asked for a confidence boundary.
“Not which route is home,” she said. “What can we say without pretending?”
Elias separated the claims. “The east marker is authentic. The east overlay data is internally consistent. Recent physical traffic probably used west. Airflow probably connects west to a larger volume, possibly the anchor corridor. The harvest marks support travel direction but do not prove it.”
“Probability on the traffic?”
“Above eighty that it is recent Alpenkern equipment. Below sixty that it travelled toward the Gate. We only have approach impressions, not departure.”
Tobi crouched beside the wheel track. “The outer lip is deeper on this side.”
Georg fetched a spare wheel pad and pressed it into dust beside the track under equal load. The comparison showed the old impression had been made while the wheel pulled west, not east.
Elias revised the second estimate upward.
They checked the east corridor without entering it. Tobi cast a passive optical probe around the first bend, tethered with dark fibre so no directed light crossed the gallery. The image showed an intact passage and another authenticated marker thirty metres ahead. No collapse. No obvious reason the route should be wrong.
The absence of a visible failure made the choice harder. A collapsed east corridor would have let everyone become certain.
Farid said, “If we cannot establish return, we anchor here.”
That meant opening the emergency spool, fixing a known point, and waiting for remote navigation support while the pharmaceutical cases warmed toward their transport limit. Safe, expensive, and still dependent on a map they distrusted.
Klara requested remote topology review. The terminal returned a queue estimate of twenty-six minutes. The first case had thirty-four minutes of thermal margin.
“We have eight minutes to learn something useful before waiting becomes the only responsible option,” she said.
The crew did not need Elias to solve the gallery. They needed him to describe exactly how unresolved it was.
The contradictions survived different observers.
Analysis opened once more, this time without swallowing the room.
ANALYSIS has reached Common Level 5. Cross-context comparison improved. Contradiction retention improved.
The level did not solve the route.
It made the disagreement harder to dismiss.
Analysis now retained conflicting groups instead of smoothing them into the nearest pattern. That felt less like insight than carrying two maps in the same hand. East had authenticated nodes, valid coordinates, and a clean checksum. West had airflow, traffic, and working history. The skill could preserve the contradiction. It could not decide which class of evidence the topology was capable of falsifying.
Klara looked at the map, then at the dark east corridor marked as home.
“No further advance,” she said. “We verify return direction before this becomes an abort.”
Elias held the airflow ribbon.
The map pointed east.
Every physical trace pulled west.