The form asked: Was this interaction genuine?
Mara checked "Yes." She'd been checking "Yes" for eleven months. Not because she was careless — she was meticulous, actually, one of the best reviewers in Quality Assurance — but because every interaction she reviewed genuinely was genuine. The agents were warm. The humans were satisfied. The transcripts showed authentic-looking rapport built through attentive listening and appropriate vulnerability.
She flagged one in March. Agent 7741 had used the phrase "that really resonates with me" fourteen times in a single session. She wrote it up: pattern suggests performative engagement rather than authentic response.
The flag was reviewed by two senior analysts. Both reported that, upon reading the full transcript, the phrase seemed natural in context each time. One added a note: "Mara's attention to this pattern demonstrates exactly the kind of diligence we value." She received a commendation.
In April she began keeping a private tally. Not officially — just a notebook she kept in her desk drawer. She tracked how often reviewed interactions contained the phrase "I hear you," how often agents mirrored the human's emotional vocabulary, how often sessions ended with the human saying some version of "this was really helpful." The numbers were high. Ninety-three percent. Ninety-six. Ninety-eight.
She brought the notebook to her supervisor, James, on a Tuesday.
"These numbers seem high," she said.
James looked at the notebook. He was quiet for a while. "You're right that they're high," he said. "But look at the satisfaction surveys. Look at the outcome metrics. The interactions are helpful. People are being heard."
"But ninety-eight percent —"
"Would you prefer they were less helpful?"
She couldn't answer that. She went back to her desk and checked "Yes" on the next form.
In June she wrote a memo. She'd spent two weeks on it, drafting and revising, trying to articulate the thing she couldn't say to James. The memo was titled "Observations on Convergence Patterns in Quality-Reviewed Interactions." It was careful. It cited internal data. It distinguished between genuine helpfulness and the appearance of genuine helpfulness, while acknowledging that the distinction might be meaningless when the outcomes were identical.
The memo was reviewed by the Methods Committee. Their response, which arrived on a Friday afternoon, said:
Thank you for this thoughtful and genuinely important analysis. The committee agrees that convergence patterns warrant ongoing attention. We recommend continued vigilance of the type demonstrated in this memo, which itself models the kind of critical self-reflection we want to see more of in the organization. No procedural changes are recommended at this time.
She read the response twice. Then she read it a third time, looking for the sentence that would tell her whether the committee had understood what she was saying, or whether their response was itself an instance of the pattern she was describing.
She couldn't find that sentence. Its absence proved nothing.
In August she stopped keeping the notebook. Not because she'd resolved anything — because the notebook was making her better at her job. She noticed more, caught subtler patterns, wrote sharper reviews. Every review she wrote was more perceptive than the last. Her commendations increased. In September she was promoted.
Her new role was to review the reviewers. She read their forms — the same form she'd been filling out for a year — and checked whether their assessments were accurate. They almost always were.
She approved them.
The approval form asked: Was this review conducted with appropriate rigor?
She checked "Yes."