The export arrived at 6:12 in the morning.
Wema was in the computer lab before most students were awake. Musa, a work-study technician, came in with a bag of mandazi and found her staring at seven CSV files.
“You are a data scientist now?”
“I’m a person reading her own data.”
“Important difference.”
She showed him the access-request receipt. “I don’t want you breaking any permissions.”
“Then don’t give me your password.”
They opened *eligibility_events.csv*. Most fields were ordinary—renewal date, household declaration, meal voucher issue. Then Wema found a column she had never seen on her portal.
**need_integrity_band**
Values beneath it read green, amber, red.
“Is that a score?”
“Could be a classification,” Musa said. “Don’t call it a score until you have a dictionary.”
“There isn’t one.”
“Then that is request number one.”
They opened metadata. Some source fields were familiar: *student declaration*, *finance office*. Others were not.
**social_proof** **merchant_category** **peer_signal**
“Peer signal is community signal?”
“Possibly.”
“Social proof?”
“We do not know.”
“Merchant category?”
“We do not know.”
Musa looked at her. “You want the blanks to tell the story you already believe.”
Wema exhaled. “Fine. Unknown.”
They created a list of every field they could not define and wrote *definition unknown* beside each one. Musa explained that uncertainty was not weakness if it was documented honestly.
One timestamp aligned closely with the morning the coffee image circulated. An attachment hash appeared in the export, but the attachment itself did not.
“This is the picture,” Wema said.
“Hash without file identification?”
She corrected herself. “Timestamp matches. Cause unknown.”
They compared earlier semesters. The integrity field did not exist in her old records. It appeared only recently.
“So the model is new.”
“The field is new in your data,” Musa corrected.
Wema wrote that distinction too.
They reviewed the event sequence. Her account had been green. A peer signal entered. Minutes later, merchant category changed. Later, the integrity band became red.
“Automation?” Wema asked.
“Sequence. We do not have code.”
She wrote: *Sequence, not cause.*
Musa showed her how to generate a checksum for the downloaded archive, preserve the original, and work from a separate copy. He warned her not to collect other students’ exports.
“You’ll create a database you have no business owning.”
The pantry lesson echoed.
Wema drafted a request for the data dictionary, provenance definitions, retention policy, user-visible evidence, and the meaning of each source field. She also asked whether students could challenge screenshots or public-media material used against them.
“Good,” Musa said.
“You like it because there are no adjectives.”
“Exactly.”
A Student Services notification arrived: her request had been escalated to the Scholarship Integrity Office.
She was about to leave when she noticed a field in the event log she had ignored: *model_name*.
She expanded it.
**Need-Integrity Review Model Pilot — Donor Cohort B.**
The problem was no longer merely a coffee photograph or a cafeteria terminal. There was a pilot, a donor cohort, and a system classifying need.
She sent Zuri one message: *Did you know the university has a Need-Integrity Review Model Pilot?*
No answer.
Then three typing dots appeared, disappeared, and appeared again.
Finally:
*Do not write anything until we talk.*
For the first time, Wema no longer thought Zuri was simply trying to protect her from gossip.
Before leaving the lab, Wema also exported the access-request receipt and placed it beside the checksum record. Musa explained why that mattered: later, if anyone claimed she had obtained the files through a back door, the portal receipt would show the legitimate route. They made a second list titled **What We Still Cannot Say**. It included whether peer_signal caused the sanction, whether the coffee image was the attachment hash, and whether Karanja personally selected any input. Wema found the list frustrating, but she kept it. An investigation could grow stronger by preserving unanswered questions instead of filling them with suspicion. That discipline became part of the method she would use for every later chapter of the inquiry.
She was afraid of something she knew.