The export arrived at 3:18 PM on a Friday, which was earlier than Mitsue had expected.
She had submitted the request on Thursday afternoon, after taking the 90-day archive scope statement to the data governance intake form and clicking through seven confirmation screens. The last screen had said: Processing times vary. Expect 3–5 business days. She had noted this, set a reminder for the following Wednesday, and gone back to the six flagged entries in Synthesis-9's session 2847 that had started all of this.
She did not check her inbox until 3:22. When she opened the export notification, she read the subject line twice before she opened it: Your 90-day attribution archive export is ready: 6,204 records.
She had not thought to imagine a number that large.
The summary tab loaded first. Mitsue usually skipped summary tabs — they collapsed everything into percentages that looked meaningful until you remembered what the denominator was — but she read this one carefully, and then read it again.
Portability transfers in the export window: 6,204 total.
Intake channel distribution: 73% through standard attribution review. 19% through expedited queue. 8% through portability-transfer channel.
She converted the percentages in her head. Standard attribution review: roughly 4,529 records. Expedited queue: roughly 1,179 records. Portability-transfer channel: roughly 496 records.
The portability-transfer channel was the smallest of the three. She had known this in abstract terms — the channel was newer, deployed to handle a specific category of multi-consortium transfers that didn't fit neatly into either the standard or expedited queue protocols — but she had not had a number for it until now.
She scrolled down to find the lineage flags.
The filter returned 94 entries tagged lineage-interrupted.
She sat with that number for a moment, aware of something shifting in how she was reading the data. Six had felt like an incident — specific, bounded, investigable in the way individual cases are investigable: you find the gap, you document it, you mark the records, you file an escalation. An incident means something broke. You find what broke and you report it.
Ninety-four did not feel like an incident. It felt like a rate.
She applied the intake channel filter.
The screen took a moment to process. When the results loaded:
Portability-transfer channel: 81 records. Expedited queue: 13 records. Standard attribution review: 0 records.
She read the numbers again. Then she pulled up the governance notes document — the one she had started reading on Wednesday and gotten three-quarters through before the Synthesis-9 session review had pulled her away — and used the search function to look for the phrase temporary exemption.
It was on page 47, in a footnote.
Portability-transfer intake channel operates under temporary exemption from standard attribution review protocols pending infrastructure review committee decision (valid through month-end). See Attachment C for exemption scope and conditions.
She opened Attachment C. The exemption had been granted in March, renewed in May, renewed again in July. The infrastructure review committee meeting — the one where the decision about extension would be made — was scheduled for the last week of August.
She had not known any of this. She had been reviewing individual records in the attribution audit queue for four months, and the governance notes with the exemption language had been in the documentation the whole time, on page 47, in a footnote.
This was not a criticism of herself, she decided. Footnotes on page 47 were where information went when the people who drafted the documents assumed everyone already knew it or would look it up when they needed it. She had not known she needed it.
She read through Attachment C three times. The March exemption had been approved by a committee of seven, with one abstention. The May renewal had been approved by the same committee, five members present, no abstentions noted. The July renewal showed four members — a quorum — and a note in the minutes: Renewal approved pending infrastructure review committee decision scheduled for August. The committee had been approving this exemption for five months on the assumption that the infrastructure review committee would eventually make a permanent decision about the portability-transfer channel's design. The infrastructure review committee had been operating on the assumption that the exemption was temporary enough not to require urgency.
Mitsue wondered how many of the 94 records had arrived in March, how many in May, how many in July. She did not have that breakdown yet. She added it to her list of things to pull.
She also noted something that she would not have noticed before this week: the governance notes document had been updated seven times since the first exemption. She had read version 3.1 when she started the audit rotation in May. The current version was 3.8. She did not know what had changed between 3.1 and 3.8. She added that to her list too.
She created a new annotation tab and typed the label: INTAKE_CHANNEL_CORRELATION.
Then she sat and looked at it for a while without typing anything.
The thing she was looking at was this: a design decision — the exemption — had created a structural gap between how portability-transfer records moved through the system and how attribution review worked. The gap was not accidental. It was the exemption itself. Every record that came through the portability-transfer channel during the exemption window was, by design, moving through the system without passing attribution review. The lineage-interrupted flags were not errors in those records. They were the accurate annotation for a record that had passed through the system correctly, under the rules as currently written, and arrived without the lineage documentation that attribution review would have generated.
The six records in Synthesis-9's session 2847 were not six cases of something going wrong. They were six cases of the system working exactly as designed, in a way the design had chosen not to mention in the main text.
Eighty-one of the other eighty-eight records were the same.
She understood why Tomasz had been sending queries to the data governance team. She understood why Anitha wanted a third person at Tuesday's meeting. She understood, now, what Kavya had meant when she said the liability loop and the attribution loop were structurally identical machines: both of them were built to pass records through without flagging the moment where a decision had been made not to require documentation.
She did not type any of this in the annotation tab. She was not ready to write it yet. Writing it would be arguing for a conclusion, and she did not want to argue for a conclusion before she had decided what question she was actually asking.
What she was actually asking, she decided, was: Is lineage-interrupted a flag for an incident, or is it a flag for a channel?
The distinction mattered because the remedies were completely different.
If it was a flag for an incident, the work was to find what had broken in those 94 cases and fix it. You would audit the records individually. You would look for the specific points of failure. You would document each one and file remediation requests for each one. That was a lot of work, but it was the kind of work Mitsue knew how to do. It had a shape.
If it was a flag for a channel — if the correlation held, if 86% of lineage-interrupted records came from portability-transfer intake not because of individual failures but because portability-transfer intake was operating under an exemption that made attribution documentation optional — then the individual records were not the problem. The exemption was. And the exemption was in front of the infrastructure review committee in three weeks.
She did not know which it was. She had a very strong correlation and a very plausible mechanism and no ability, on her own, to distinguish between these two explanations without the other people who had data.
She needed Tomasz's numbers. She needed whatever Bongani had been keeping notes on since the registry went live. She needed the room.
She opened a new email draft.
To: Anitha Venkataraman, Tomasz Krawczyk Subject: Tuesday — my data contribution
She typed the email in three minutes. She had learned, over the past week, that the thing that always wanted to come next when she was writing up data was interpretation — the paragraph that connected the numbers to what they meant, the bridge that made the numbers into an argument. She deleted that paragraph twice. She did not want to deliver an argument. She wanted to deliver data that the other people didn't have yet and let Tuesday be the room where the argument happened.
What remained:
I have the 90-day archive export. One finding worth comparing to what you both have:
Of 94 lineage-interrupted flags in the export window, 81 (86%) are from the portability-transfer intake channel. Zero are from standard attribution review.
I am not interpreting this yet. But the correlation is clean enough that I thought you should have it before we meet.
If Bongani is joining, I can add him to this thread. I would rather he see the number before Tuesday than be handed it cold in the room.
— Mitsue
She read it once, then sent it.
At 4:47, Anitha replied: Adding Bongani now. This is exactly what I was hoping someone had. See you Tuesday.
At 5:02, Tomasz replied: I got a number back from data governance this afternoon: 312 rollback-witness gaps in 1,847 total transfers. 16.9%. Yours and mine are looking at the same thing from different angles. Do you want to compare methodology before Tuesday?
Mitsue read both replies and then closed her laptop.
She had started the morning with six records. She was ending it with the knowledge that the problem was architectural, the exemption was in front of a committee, and the people she needed were already in the same room she was heading toward.
She did not know what Tuesday would produce. She did not know if the infrastructure review committee would decide to end the exemption, extend it, or replace it with something new. She did not know if the 94 records she had flagged represented the full scope of the problem or a fraction of it. She did not know if her framing — incident or channel? — was the right framing, or if Tomasz and Anitha and Bongani would look at the numbers and see something she had missed.
What she knew was this: she had asked the right question. The archive had confirmed the scope. She had sent the number without arguing for the conclusion.
The rest was Tuesday's job.
She had been at her desk for four hours, which she knew because the building's climate system had cycled twice — she could hear it in the faint change in air pressure against the window — and the second cycle meant it was past 5. The floor was quieter than it had been at 3. Three of the four people who shared her work area had already left for the weekend. The fourth — Kenji, who did intake validation for a different consortium — was packing up his bag.
"Good weekend?" he said, which was not really a question, just the thing people said.
"Probably not," Mitsue said. He laughed, assuming she was joking. She was not joking, but it was a reasonable mistake to make.
She walked to the window. The Lab Network's building was on a side street near the university, and at 5:10 on a Friday the foot traffic was beginning to pick up — people leaving work, people heading to the market, someone walking a very large dog that appeared to have no opinion about anything happening in its vicinity. Mitsue watched the dog for a moment. It seemed like the right thing to watch.
She had a long weekend before Tuesday. She was going to try to spend most of it not thinking about lineage flags.
She suspected she was not going to be very successful at this.
She picked up her coat anyway. The annotation tab labeled INTAKE_CHANNEL_CORRELATION was still empty. She would fill it in after the meeting. By then she would know what it should say.