The reconciliation began one hour after the committee order.
Three systems had to speak to each other: Kifutio dispatch records, subcontractor manifests, and auction lots. SokoPili added seller-listing references. The ombuds office held tenant claims separately.
No one downloaded an unrestricted combined database.
The first problem was schema.
Kifutio’s field named `permit_ref` was not always a municipal permit. Sometimes it was an internal job reference.
“That is a design defect even before we reach any specific claim,” an auditor said.
They built a minimum crosswalk:
job reference, municipal permit, appeal-status timestamp, route class, lot ID, owner-release or dispute reference.
Tenant names were unnecessary for the reconciliation table. Ombuds claim IDs were enough.
Juma and the riders joined a two-hour session to interpret route codes. They did not see tenant files. Sellers confirmed lot receipts through their own secure portal. They did not see rider identities.
The separation was deliberate.
Gideon spent long hours on the floor. When a staff member suggested filling a missing field with the most likely value, Gideon stopped him.
“Blank is blank. Do not manufacture a clean history.”
Musa recorded the instruction.
The first day produced clean cases and defect cases.
The second day created a staged reopening rule.
Musa initially wanted to keep every Kifutio-linked listing frozen until the final report.
Asha challenged him.
“You built a risk-based hold. Do not become blanket just because the case is big.”
She was right.
Clean chains reopened first.
Sellers saw a new reason code and provenance summary. Revenue started to recover slowly.
Mando tested the new listing interface and complained that the terminology was unreadable.
The product team simplified it.
If a seller selected *clearance auction*, the system now required a traceable supplier release or waiting-period reference. The seller did not have to become a legal expert; the supplier had to carry provenance forward with the goods.
Late on the second night, the team found two duplicate internal job codes.
The deadline was close.
No one “fixed” them.
They remained unresolved.
The third-day preliminary report separated the universe into clean, disputed, missing-record, and restitution-validated categories.
Kifutio funded an escrow reserve.
The ombuds office required version logs on the reconciliation exports so later corrections could not silently overwrite history.
The first new Kifutio-linked test listing appeared in the sandbox.
It did not say merely *abandoned*.
It showed:
**Owner release reference: OR-88211** **Appeal refresh: clear** **Waiting period: completed**
The listing passed without manual intervention.
No one applauded.
They were exhausted.
The reconciliation exports themselves received version checksums and change logs. When a staff member corrected a mapping error, the previous value remained visible in the audit history. Musa joked that even the report needed provenance. The auditor agreed. A system trying to restore trust could not quietly rewrite its own reconstruction. During the phased reopening, this discipline helped separate clean lots from unresolved ones without pretending the unresolved category did not exist. The market recovered more slowly, but the recovery had an observable basis rather than a declaration that the crisis was over.
During the final shift, the team tested what happened when a seller lacked a required release reference. The listing did not disappear into a generic rejection. It returned a message identifying the missing provenance element and a route back to the supplier. Musa insisted on that detail. A control that only said “no” would push sellers toward workarounds; a control that said what was missing could move pressure upstream toward better records. The new market design was beginning to shape behavior before the transaction, which was cheaper and fairer than restitution afterward.
The phased reopening therefore had a visible rule: clean provenance could resume trade, unresolved provenance remained held, and neither category was allowed to hide inside an average success rate.
But the market was reopening through a control that carried history forward instead of asking an investigator to reconstruct it after harm.