The independent review lasted six weeks. Its interim findings came in a thick document filled with diagrams, appendices, and vocabulary that made ordinary mistakes sound scientific. Noor read the summary twice before she understood the central conclusion: the GuestPath pilot had been deployed with inadequate controls over how staff instructions were generated, and some data variables created predictable differences in who was asked to compromise.
The investigators found that the missing user identifier did not indicate a secret intervention by an unknown executive. A software connector had failed to preserve the staff ID when an approved service contact made a request. The missing field mattered because the failure obscured accountability. But the person behind the instruction was traceable through a separate access log: a customer-service contractor who had received Clara Wynn's message and requested an accommodation for Calder.
The contractor had not been promised money and had not received a direct order from the investor. She had believed she was helping a valuable customer. GuestPath supplied a recommendation; the interface represented it as an instruction; she forwarded it without checking whether Mia had agreed. A chain of ordinary decisions produced an extraordinary level of pressure on someone who had never been consulted.
Noor felt both relief and disappointment. She had worried about a conspiracy and discovered a design failure. The distinction mattered, even if the effect on the passenger remained serious. She wrote in her margin: Intent is one question. Consequence is another.
The review also found that several test variables, including approximations derived from spending history, had remained accessible to the scoring service after the project team believed they were disabled. Those variables did not determine Mia's final seat, but they influenced the priority score presented alongside Calder's request. The investigators recommended shutting down the pilot until consent requirements, audit trails, staff training, and data limits were rebuilt.
At a meeting with the airline's leadership, Noor asked whether employees would be able to challenge priority recommendations without being penalized for slow service metrics. The room went quiet. A senior executive asked her to explain. She pointed to a graph showing average handling time and said, "If you measure employees only by how fast they remove a complaint, they'll have an incentive to move the person least likely to object. You can fix the words on the screen and still leave the pressure in place."
Helen Cho nodded. Her team added a recommendation about performance evaluation and staff discretion. Noor felt more hopeful than she had in weeks. A repair that addressed only software would have missed the human system wrapped around it.
Mia received an invitation to review a passenger-facing summary. It disclosed the cause of the attempted seat adjustment in plain language, acknowledged that consent had not been sought, and confirmed that the incident had been recorded in the airline's complaint system. It also stated that the contractor's identity would not be released publicly. Mia appreciated the privacy protection, even as she considered how useful it would have been to know every name.
She called June, who asked whether the findings changed how she thought about Calder. "A little," Mia said. "He was rude. But the system had made his preference seem more important before he said anything."
"Both can be true," June replied. "A person can be responsible for their conduct and still be encouraged by an unfair structure."
Noor read the technical appendix at home rather than in the office. Her kitchen table was covered with a child's colored pencils, a stack of unopened mail, and the half-finished shopping list she shared with her sister. The diagrams showed exactly where the GuestPath team had been careless. She wanted to feel triumphant that the investigators confirmed her concerns. Instead she felt weary, as though she had been carrying a heavy box for so long that setting it down left her arms shaking.
Her sister, Farah, made tea and asked whether the report vindicated her. Noor disliked the word. "It explains what happened," she said. "That's more useful."
"Can you keep your job?"
"I think so. They've asked me to help rebuild the system."
"And do you want to?"
Noor had not considered the question without its practical obligations attached. She liked her colleagues, valued the technical challenge, and believed the airline could improve. Yet the last weeks had shown her how quickly a healthy skepticism could be treated as a threat to momentum. She told Farah she would stay only if the redesign included people who actually faced the consequences of software decisions. Farah said that sounded less like a compromise than a condition worth defending.
The next morning Noor wrote three questions at the top of a design document. Can the passenger understand the recommendation? Can the employee refuse it without penalty? Can an investigator reconstruct what happened afterward? The questions were not mathematically sophisticated. They were the kind anyone affected by the tool might ask. She sent them to Helen Cho as proposed acceptance criteria for a future launch.