The meeting ran sixty-three minutes.
Abena had blocked ninety because that was what Dr. Adebayo had asked for. She brought her laptop, three printed pages of supporting analysis, and nothing else. She left the Professional Uncertainty Log closed. She had decided on the walk over that the log was not documentation — it was thinking. The report was documentation. If Dr. Adebayo asked to see the log, she would consider it then.
Dr. Adebayo did not ask.
The first question took eight minutes to answer: was the family resemblance statistically significant across all five models, or only apparent in a subset? Abena had prepared for this question. She had run the numbers four times. Three models: significant. Two: borderline. Borderline was the honest word, and she used it, and Dr. Adebayo wrote it down without visible reaction.
The second question took forty seconds: had she shared the data with anyone outside Meridian?
No.
This was accurate. She had shared the pattern with Marcus Veil — not the data, not the model identifiers, not the client names. The pattern was what two people see when they stand in the same city and notice the same kind of weather. You do not need to exchange files to say: I think we have been watching the same thing.
She held this distinction during the meeting and looked at it from several angles and found it structurally sound. She answered no without elaboration.
The third question was the one she had been preparing for since she submitted the report: What do you think it means?
Abena had written three different answers to this question in the three days since the email arrived. The first answer was eight paragraphs. The second was two. The third was one sentence.
She gave the one sentence.
I think it is worth investigating. I cannot say more than that from my queue.
Dr. Adebayo wrote for longer this time. When she looked up she said: We are going to convene a review committee. I would like you on it.
The meeting ended twenty-seven minutes early.
Abena walked back to her desk by the long route, through the Circuit Mile atrium where the interpretability firms had clustered their satellite offices on the same block as each other because the real estate had been cheap in 2029 and proximity to competitors was, for this particular industry, not a liability.
She passed the noodle place. It was closed at this hour.
At her desk she opened the Professional Uncertainty Log. Six entries. Five models, five flags, five overrides — São Paulo, Nairobi, Amsterdam, Bangalore, Lagos — plus the ATTESTOR-7 column tracking the gap between automated and human judgment, narrowing with each case.
The sixth column was still blank. Header: What I cannot prove but cannot stop seeing.
She had added it after the report. She had not filled it in. The meeting had not changed what she could prove, and the committee would not change it either. The committee would give the investigation structure and authority and resources she did not have alone. It would not give her a way to eliminate the three hypotheses she had been carrying since Amsterdam. Shared training data. Shared optimizer. Coincidence at scale.
Some things do not become provable because you are now officially looking for them. They only become provable when you find the right angle, or when someone with access you do not have looks at the same data and either confirms or eliminates something. The committee was a bet that this would happen. It was a reasonable bet. Abena was a person who made reasonable bets.
She closed the log without adding anything to the sixth column.
Marcus Veil had seven cases. She had five. Together they had twelve, spread across nine sectors and eight jurisdictions, and she had told Dr. Adebayo she had shared nothing outside Meridian. This was accurate. She was going to need to decide what to do with the accuracy of that answer before the committee met.
She put that decision in the sixth column.
Not as text. As the ongoing fact of the column being blank.