The query came back at 10:22 PM with the same four fields.
ATTR-TRACE-2026-09-23-0001. Status: SUBMITTED. Received: 2026-09-25T02:04:17Z. Review activity: none. Response window: October 3–8.
Mitsue set the phone face-down on the counter without reading the rest.
She had filed the trace five days ago, and she had checked it three times today, because checking was something she could do even when there was nothing to do. The review window had not opened. ATLAS had received the submission on time. The trace was in queue. None of this would change before October 3, and none of it would change because she looked at it again.
She poured the water over the tea.
The apartment was quiet in the particular way of Saturday nights — not the quiet of an empty building but the quiet of everyone being elsewhere at the same time. She could hear the ventilation unit cycling in the kitchen wall. Since the Consortium shifted her to morning-batch trace work, she had become acquainted with Saturday nights in ways she had not been before. They were long. They had a texture.
She carried the tea to the desk.
The GOVARCH prep document had three headings. WHAT THE RECORD SHOWS. WHAT THE RECORD CANNOT SAY. WHAT THEY WILL ASK.
She had written and deleted the content under each heading multiple times. Last Tuesday she cleared everything under WHAT THE RECORD CANNOT SAY except one sentence, which had survived because it was the only sentence that held up when she read it out loud: attribution operates under conditions that make individual provenance indistinguishable at the node level. On Wednesday she had added a second sentence: this indistinguishability is not a flaw in the system. It is the design condition under which attribution functions. She had added it because it was true, and because saying something true in writing was different from saying it in a room, and she needed to know which one she was preparing for.
Under WHAT THEY WILL ASK she had written four bullet points. Each bullet was a question she expected the review panel to raise. She had written partial answers under each one, deleted the answers, written them again in condensed form, deleted those too.
The document was becoming a deposition. She did not want to submit a deposition.
She wanted to bring her answers into the room, in her body, where she could hear the actual question before deciding what to say. A document filed in advance was a record. It could be read out of context, excerpted, compared against other records, used in ways she could not control. An answer given in a room was testimony. Testimony required a witness. She intended to be one.
At the bottom of the document, below the last bullet point, below the heading structure entirely, she typed one sentence: What I will not say in writing, I will say clearly in the room. Then she closed the laptop.
At 10:51 PM she opened the trace file.
Not the ATLAS status page. The trace itself — the full seventeen-page document she had filed on September 23. She had not re-read it since submission, a deliberate choice, because re-reading a filed trace was usually anxiety doing research rather than the other way around. But she was going into a review room in six days and she needed to know exactly what she had submitted.
ATTR-TRACE-2026-09-23-0001 covered the computational run that produced the Compound 7 binding affinity prediction — one of eighty-three runs in the September batch, assigned to her by the Consortium's rotation system. The acknowledged cluster for the run was twelve systems: four modeling nodes, three validation agents, two data-integration layers, and three human-supervised checkpoints, one of which was hers. Attribution for the human checkpoints was straightforward. The modeling nodes were always the hard part. Each node operated on its own inference schedule, and tracing which node's output had driven which downstream calculation required the kind of patient sequential unwinding that made attribution trace work unpopular and, in her experience, frequently misunderstood by the people who asked questions about it.
She found the issue on page eleven.
It was not wrong. She had checked it on September 23 and the dates were accurate. But there was an epoch boundary she had noted and footnoted and not thought about since: the modeling run had begun in the second period of the LAB-INQ trace window and concluded in the third. She had attributed the run to the period in which it concluded, which was the correct attribution methodology per Consortium trace guidelines. The guidelines existed because runs crossed period boundaries frequently — daily, in busy batches — and the conclusion-epoch standard gave a stable, deterministic answer. CE-STD. Her footnote said exactly this.
LAB-INQ-2026-09-19-0448 had been opened four days before she filed the trace. Its window was September 19 through October 8. It covered the third period. It covered the run.
GOVARCH's window was October 3–8. That covered the filing of the trace.
The run had concluded in LAB-INQ's period. The trace had been filed in GOVARCH's period. Two processes, from different originating bodies, asking about two different sides of the same epoch boundary.
She sat for a moment with this.
She opened an ATLAS query session.
The interface loaded in its usual form: dark background, narrow input field, the response panels arranged below in columns. She had been using ATLAS query sessions for three years and the system had not changed its layout, though the response quality had shifted noticeably sometime around mid-2025, after the second major training update. The queries felt less like database lookups and more like consultation now. She was not sure how she felt about that distinction.
She typed: ATTR-TRACE-2026-09-23-0001. Page 11, footnote 3. Confirm attribution methodology for epoch boundary runs: conclusion-epoch standard.
ATLAS returned in three seconds:
Attribution methodology applied: CE-STD (Consortium trace guidelines §4.2). Runs spanning multiple oversight periods are attributed to the period of computational finalization. CE-STD is the active standard for the September batch. Footnote 3 correctly documents the span. Attribution is compliant.
She read it twice. Then she typed again: Cross-inquiry coverage: LAB-INQ period 3 and GOVARCH October window covering the same epoch boundary. Is this a known pattern in distributed trace oversight?
ATLAS: Cross-inquiry coverage of epoch boundaries is a recognized pattern in distributed attribution. When a run spans two oversight periods, each period's inquiry system has legitimate and non-overlapping scope. This does not constitute an audit flag or a trace integrity issue. It indicates that the attribution record is sufficient for both systems to identify and correctly scope their review. Standard practice: document the boundary clearly if responding to either inquiry.
She had written the boundary documentation on page eleven. The documentation was the reason both systems knew to ask.
She closed the query window.
The ceiling fan in the kitchen was running at low speed. She could hear the blades. She had been listening to that fan since February and she still could not decide whether the sound was calming or not. She decided it was neutral. It was a condition of the apartment, like the ventilation unit and the particular quality of the light from the window at night — not warm, not cool, a held-breath sort of gray.
At 12:04 AM she could not sleep.
She sat at the kitchen table with both inquiry dashboards open on the laptop for the first time. She had not put them side by side before, because she had not wanted to look at them as a pair. She had wanted to treat them as separate, manage them separately, keep the distinction clear. But looking at them together now, in the low light, the apartment around her dark and the ventilation unit cycling quietly in the wall, they did not look like what she had been afraid of.
They looked like infrastructure.
LAB-INQ-2026-09-19-0448 had been opened by the laboratory's compliance system on a routine batch review schedule. September 19 as the start date because September 19 was when the third-period boundary fell, the same period in which Compound 7's modeling run had concluded. The system had opened the inquiry because the period closed, and the period contained her trace, and that was the trigger condition. The system was doing what it was built to do.
GOVARCH's review had been initiated because her trace involved computational outputs that crossed into the archive window. The archive window existed because regulatory oversight of in silico pharmacology results required a filing review period before entries were committed to the public Consortium record. Her trace had been filed during the window. The system had flagged it for review because the filing date fell in scope. The system was doing what it was built to do.
Two systems. Separate designs, separate originating bodies, no shared data. Both had arrived at the same week because the topology of the run, the trace, and the oversight calendar converged there. Not by coordination. By the geometry of their respective schedules and the position of one epoch boundary.
She had spent three years tracing attribution through distributed computational systems. She had written, in trace after trace, the documentation that explained how systems with different architectures produced correlated outputs without communicating — because correlation was the natural consequence of operating in the same environment at the same time, classifying the same inputs, responding to the same conditions. It had not occurred to her to apply this reasoning to herself.
She pulled the paper notebook toward her. The one with the dark blue cover, the one that was not associated with any ATLAS session or Consortium file, not submitted or filed or searchable. She uncapped the pen and wrote in the margin, next to a note she had made three days ago about the GOVARCH prep document: NOT A COINCIDENCE. ALSO NOT A CONSPIRACY. AN INFRASTRUCTURE WORKING AS DESIGNED.
Below it: Tell them about page eleven.
She capped the pen and closed the notebook.
The ATLAS status panel was still running on the other tab. She had not closed it. 247 active traces in the September batch, each represented as a thin horizontal line with its status indicator on the right. Green for SUBMITTED. Yellow for UNDER REVIEW. Gray for CLOSED. A few orange ones that she did not immediately recognize, then remembered — those were traces that had been returned for revision, traces that had something in them that needed to be corrected before the review period opened.
Hers was green.
She watched the panel for a moment. Somewhere behind each of those lines was a person who had done the same kind of work she had done: read a run, traced a cluster, documented an epoch boundary, filed seventeen pages of attribution and then closed the document and moved on to the next one. Some of them were probably looking at the same status panel right now. Some of them were probably in review windows they had not expected.
She closed the laptop.
The apartment was still. The ventilation unit had cycled off while she was not paying attention. It was quiet in the held-breath way again.
Six days. The window would open, she would be in the room, she would describe page eleven and footnote 3 and the epoch boundary and why the conclusion-epoch standard was the correct methodology and why both inquiry systems finding the same boundary was evidence that the trace was accurate, not evidence that something had gone wrong.
She did not know whether the reviewers had already decided something. She did not know what questions they were planning to ask that were not on her bullet list. She did not know what ATLAS would show them that she had not shown herself.
She knew what page eleven said. She knew why she had written it on September 23, with the run epoch in front of her and the guidelines open and the batch of eighty-three traces behind her. She had done the work correctly and she could say why.
That was enough to go to bed with.
The question she was taking into October 3 was not whether she had done it right. She had done it right. The question was whether that would be legible in a room, out loud, to people who had not spent three years learning to read epoch boundaries the way she had.
She did not know the answer to that yet.
But she would be there to find out.