Arin spent the next five days failing on purpose.
Instructor Valeor had assigned the class to repeat Cleanse until every apprentice could cast it ten times without damaging the spell model.
Most students treated the assignment as practice.
Arin treated it as testing.
His first experiment was simple.
He used one mana particle.
The spell executed successfully.
The stone became clean.
His second attempt used two particles.
The result was identical.
His third used three.
The stone became cleaner than necessary, but the spell consumed more energy.
His fourth attempt used four.
The model overheated.
A red line appeared along the outer boundary.
Arin immediately released the connection.
The spell failed, but the model remained undamaged.
He wrote the result in his notebook.
Four particles: excessive input.
Boundary temperature increased.
Safe abort possible.
He tried again.
This time, he allowed the model to continue until the outer ring flickered.
The protective rune in the desk activated and severed the connection.
The metal plate turned cold.
Arin’s fingers tingled.
The instructor appeared beside him.
“What are you doing?” “Testing the failure point.” “You were told to repeat the spell.” “I am repeating it.” “You are deliberately overloading the model.” “Within safe limits.” Instructor Valeor took the plate from him and examined the outer ring.
“You are not qualified to decide what is safe.” “I stopped before the protective rune activated.” “Not this time.” “It activated during the previous attempt.” “And you think that makes it acceptable?” “No.
I think it provides useful information.” The instructor’s face darkened.
“Magic is not a puzzle box.” “I am not treating it as one.” “You are treating it like a machine.” Arin looked at the spell model.
“Machines also fail when used incorrectly.” “That is not the point.” “Then what is the point?” “The point is obedience.
You were given a basic model.
Your task is to learn its proper use.” Arin nodded.
“Yes, Instructor.” Valeor returned the plate.
“Repeat the approved method.
If you damage the model, you will be removed from class.” Arin accepted the warning.
For the rest of the lesson, he used one particle at a time.
The class moved slowly.
Several students still failed because they pushed too much mana into the central mark.
Others lost focus while shaping the effect.
One apprentice caused his stone to vanish entirely.
“Where did it go?” the student cried.
Instructor Valeor examined the empty desk.
“You cleaned more than the surface.” “What does that mean?” “You erased the stone’s outer structure.” The student stared at the empty space.
Arin wrote: Excessive effect: target degradation.
The spell could remove more than harmless dust if enough energy was supplied.
Cleanse was not truly harmless.
Its limits were defined by the relationship between the input, model, and target.
The same spell had different risks depending on what it touched.
That was more important than the simple description suggested.
After class, Arin returned to the laboratory.
Corven had asked him to prepare ten batches of wound salve using the revised instructions.
He completed the work carefully and recorded every measurement.
Nine batches were stable.
The tenth separated slightly after cooling.
Arin checked the notes.
The ingredients were correct.
The heat was correct.
The stirring count was correct.
The only difference was the quality of the binding oil.
He inspected the storage jar.
A layer of clear liquid had settled at the bottom.
The oil had separated before it entered the mixture.
Arin carried the jar to Corven.
“This batch was unstable because of the binding oil.” Corven looked at the jar.
“It is good oil.” “It was not mixed before use.” “The recipe does not say to mix it.” “That is the problem.” Corven removed the lid and smelled the oil.
Then he poured a little into a glass dish.
The clear layer remained at the bottom.
“Who delivered this?” “I do not know.” Corven checked the storage label.
“Two months old.” “The newer oil may be more reliable.” “It is also more expensive.” “Then the recipe should specify that older oil must be mixed before measuring.” Corven looked at him.
“You are turning every instruction into a legal document.” “Unclear instructions create inconsistent results.” “People do not follow instructions perfectly.” “Then the instructions should account for predictable mistakes.” The assistant placed the jar on the table.
“Prepare another batch.” Arin did.
He mixed the oil first, then measured it.
The salve remained smooth.
Corven watched the entire process.
“You did not need to change the recipe.” “No.
I changed the preparation step.” “And if the oil were fresh?” “I would still mix it.” “Why?” “Because it costs nothing and removes one source of variation.” Corven nodded slowly.
“Your method is irritating.” “It is not intended to be.” “That makes it worse.” At the end of the day, Corven gave Arin two copper coins.
“One for the successful batches.
One for the improved instructions.” Arin looked at the coins.
“You said one copper for every ten successful batches.” “I changed the arrangement.” “Why?” “Because I want you to continue recording failures.” Arin placed the coins in his pouch.
“Then I will.” In the evening, Arin returned to spell class for an additional practice session.
The room was mostly empty.
Instructor Valeor had left a stack of Cleanse models on the central desk.
A note instructed the students to practice without supervision.
Arin selected a damaged model.
The outer boundary was scratched, but the channels remained intact.
He carried it to a desk with a functioning protection rune.
Then he began.
One particle.
Success.
Two particles.
Success.
Three particles.
The stone became almost unnaturally smooth.
Four particles.
The boundary flickered.
Arin released the model.
The spell failed safely.
He recorded the result.
Five particles.
The model heated sharply.
Arin did not release the connection immediately.
He watched the channels.
The central mark remained stable.
The first guiding channel brightened.
The second began to flicker.
The outer ring developed a thin red line.
Abort.
Arin severed the connection.
The protective rune did not activate.
His fingers shook from the sudden release.
He waited until the sensation faded.
The failure had been controlled.
More importantly, it had been repeatable.
He tried the same input again.
The same channel flickered first.
The same red line appeared.
The same point became unstable.
Arin leaned over the model.
The spell was not failing randomly.
One channel was receiving more energy than the others.
The model’s structure might be slightly damaged, but the failure also appeared with an undamaged plate.
Perhaps the distribution was intentional.
Perhaps Cleanse was designed to prioritize the outer layer of the target.
That would explain the boundary reaction.
Arin needed more information.
He used the spell on different targets.
A stone.
A piece of cloth.
A wooden chip.
A metal spoon.
The results changed.
The stone became smooth.
The cloth lost its dust but also faded slightly.
The wooden chip became thinner.
The spoon developed a dull patch where the surface had been polished.
Cleanse did not merely remove dirt.
It removed the easiest material at the target’s surface.
The spell’s description was incomplete.
Arin wrote in large letters: Target matters.
He sat back.
The spell model was not a perfect instruction set.
It contained assumptions.
It assumed the caster understood what counted as dirt.
It assumed the target was durable.
It assumed the desired effect was more important than the possibility of over-cleaning.
Inexperienced users were protected from the worst results by the small amount of mana they could control.
A stronger caster might accidentally damage the target.
Arin imagined a powerful Magus using Cleanse on a wooden door.
The result might be an open doorway.
He wrote that down.
A tap sounded on the door.
Arin looked up.
Instructor Valeor stood outside.
“You are still here.” “Yes.” “You used the model on several materials.” “Yes.” “Why?” “To identify the effect boundary.” “You were instructed to practice on stones.” “Stones are only one target.” “You were not instructed to test other targets.” “No.” Valeor entered the room.
“Show me the results.” Arin handed him the notebook.
The instructor read the page in silence.
When he reached the line about surface material, his expression changed.
“Try it on this.” He removed a small wooden disk from his robe and placed it on the desk.
Arin hesitated.
“Is it academy property?” “No.” “Will you object if it is damaged?” “If you follow the model correctly, it should not be.” That was not an answer.
But it was permission.
Arin cast Cleanse with one particle.
The dust vanished from the disk.
The wood remained intact.
“Now two,” Valeor said.
Arin did.
The surface became smoother.
“Three.” The disk’s color faded.
“Stop,” Valeor said.
Arin released the model.
Valeor picked up the disk.
“You were correct.
The spell removes more than dirt when supplied with excess energy.” “Why is that not included in the instruction?” “Because most beginners cannot supply enough energy to cause the problem.” “That does not make it irrelevant.” “No.
It makes it inconvenient to explain.” Arin looked at the model.
“Instructions are written for the average user.” “Instructions are written for the user the instructor expects.” “That is not always the same person.” Valeor returned the disk.
For several moments, neither spoke.
Then the instructor said, “You will not perform these tests alone again.” “Because they are dangerous?” “Because you are not yet skilled enough to recognize every danger.” Arin accepted the correction.
“May I continue recording them?” “Yes.” “May I test other basic models later?” “After you learn the approved method.” Arin closed his notebook.
That condition was reasonable.
He would follow the standard process.
But he would also continue documenting its limits.
Before leaving, he added one final entry to Framework Zero.
A successful execution does not guarantee a safe result.
The spell had worked exactly as designed.
That was the problem.
Understanding the effect was as important as controlling the power.
And the first step toward understanding was learning how to fail without being destroyed by the failure.