Chapter 17 - THE SOFTWARE NEVER SAID “DON’T GIVE THE INHALER”
PulseBridge sent a lawyer before they sent a physician.
Usually informative.
Their position:
CALMPath did not diagnose.
Did not deny medication.
Did not override health plans.
Did not instruct teachers to disregard emergency symptoms.
All technically true.
The system generated contextual indicators.
Districts decided implementation.
PulseBridge’s chief medical adviser, Dr. Simon Keller, joined the meeting voluntarily.
Pediatric emergency physician.
Forty-eight.
Calm.
Not evasive.
That made him harder to hate.
I asked:
“Should asthma ever be scored as probable somatic escalation before rescue medication?”
“No.”
Immediate.
“Then why was my daughter?”
His face changed.
“She shouldn’t have been.”
Good.
Then PulseBridge engineers pulled Maya’s record.
The model had not been told:
Maya has asthma, therefore deny inhaler.
Instead it processed historical events.
Her old district had exported anonymized student-support records into a statewide education dataset.
One entry described her collapse.
But not the way it actually happened.
Student reported respiratory distress during academic activity. Staff suspected anxiety. Nurse evaluation initiated. Emergency transport completed. Student discharged same day.
Missing:
teacher confiscated inhaler.
signed asthma plan.
visible respiratory retractions.
prior medication-access warning.
hospital diagnosis.
That was the record used for training.
A sanitized institutional summary.
PulseBridge had learned from it.
Rachel asked:
“What was the training label?”
Engineer hesitated.
Dr. Keller said:
“Show us.”
Outcome class: NON-ACUTE / HIGH ESCALATION
I stared.
“You labeled an ambulance transport non-acute?”
The engineer explained.
The dataset’s outcome definition used:
hospital admission,
ICU transfer,
surgery,
or sustained observation beyond twelve hours.
Maya met none.
Therefore:
non-acute.
I laughed.
“So successful emergency treatment makes the emergency less real?”
Nobody answered.
There was the flaw.
A child uses rescue medicine.
Gets treated promptly.
Improves.
Goes home.
Database sees:
no admission.
Model learns:
probably not severe.
The better the emergency response works—
the more the data can make the response look unnecessary.
Dr. Keller looked genuinely disturbed.
“That is not how the original clinical layer was designed.”
“Original?”
He stopped.
Rachel noticed.
“What changed?”
PulseBridge counsel interrupted.
“We need to review version history.”
That meant something had changed.
Then Naomi sent me screenshots from teacher training.
One slide:
Before escalating, ask: Could this be avoidance, stress, fatigue, attention-seeking, or situational discomfort?
There.
Attention-seeking.
Old poison.
New font.
The slide carried PulseBridge branding.
Dr. Keller claimed he had never approved it.
He sounded truthful.
So who had?
Footer:
Clinical Implementation Lead — Dr. Valerie Sloan
I knew the name.
Everyone in our old district knew it.
Valerie Sloan had been the administrator who stood beside parents after Maya’s case and promised:
“We are going to build a system where children receive emergency care first and adults analyze later.”
She had left the district two years afterward.
May you like
Now she worked for PulseBridge.
And apparently she had taught software the opposite lesson.
Related Stories
