The habit is five o'clock now. Not the alarm—the alarm is set for five-thirty, a concession to the idea that I might sleep longer than I do—but by five I am usually already at the edge of the bed with the posture of someone who made a decision somewhere in the night and doesn't remember making it. I have stopped pretending this is a professional habit. It might be the attribution queue. It might be the particular quality of these weeks where there is always something in the outbox that has not yet been sent.
The tea takes four minutes. I do not open the A27 dashboard until the tea is hot. This is a rule I made three years ago when I started waking before the alarm and realized the first thing I was doing was checking position-status before water, before anything that constituted taking care of myself. The rule holds most mornings. The tea finishes at 5:14 AM. I open the dashboard at 5:15 AM.
The interface comes up blue-white in the dark—three panels tiled left to right, the room not yet worth lighting, the glow sufficient. Position-status reads position one, unchanged. The left panel is A27, always A27: a process tree with one flag sitting at the top of the active column like it has been sitting there for eleven weeks.
ISF_ASSIGN_TRANSITION_EFFECTIVE: PENDING.
I found this flag in A27 version 7.3, dated August 28, when I was building the October 3 preparation documentation. The language—ISF_ASSIGN_TRANSITION_EFFECTIVE—is a handling classification, which means it describes what happens to attribution records when the institutional assignment logic transitions between frameworks. Someone in infrastructure named it this, and the name has been in the process tree ever since. Three hundred and forty-one records carry this flag. On October 3, they move. That is nine days from now.
The center panel is the attribution substrate feed. I check Carmen first. Day 50. Amber pulse, four-second cycle, automated hold. The pulse is rendered as a soft cycling bar—amber-orange, not alarming by design, not alarming to me at this point either. I have been watching Carmen pulse since Day 12. No named accountable party, still. The substrate logs show the hold as machine-generated, origin: intake validation system, flagged for review, review not yet assigned. Whatever Carmen represents—a record, a transaction, a study in the intake queue—it has been in amber for fifty days without a human name attached to it. The pulse continues at its four-second interval. I note it. I cannot do more than note it.
Batch 19 is in the right panel: four remaining. I open the queue status. Annotations #3 through #6 are listed with their confidence flags. #3 is an NFkB stain from the consortium's Barcelona node: standard. #4 is a Ki67 proliferation marker from Ankara, which I have seen before in this consortium's work—reliable preparation, predictable signal characteristics. #5 and #6 are from a new node, Prague-affiliate, first time in my queue. I make a note.
I close the queue and open the outbox. September 26 notification: timestamp 2026-09-21 11:03.
I built the draft three days ago in twenty-two minutes, after I understood that the ISF transition was not theoretical—that the PENDING flag meant real records moving to a real new assignment system, and that real sub-contractors would need to be in place for the four to six weeks after October 3 when the edge cases would surface. The notification is standard consortium format: sub-contractor identification, institutional affiliation, appointment effective date, scope of engagement. Six names. Four institutions. Language correct, addresses current, dates verified against A27 documentation.
I read through it at 5:18 AM in the blue-white light.
It is good. The scope section is precise about what the sub-contractors are being asked to do: attribution trace review for records flagged during ISF_ASSIGN transition, specifically records where the pre-transition handling left documentation gaps. The timeline section is internally consistent. The addresses resolve. Someone receiving this would understand what was being asked and when. I have been in receipt of notifications that did not achieve this—the three-page scope that said nothing about dates, the one that used a deprecated institutional identifier for the receiving node—and I remember the friction of having to re-read to find the basic facts. This one does not require re-reading.
Six names. The question at 5:23 AM, with the tea still warm, is whether six names is correct.
The work three days ago went like this: I pulled the sub-contractor registry for the consortium and applied three criteria. Domain experience in attribution trace work or distributed-pharmacology annotation systems. Confirmed institutional availability in the October-November window. A documented delivery pattern that reads as consistent in the substrate logs—submission timestamps within expected windows, revision counts below the trace-integrity threshold, response rates over twenty-four-month periods. The third criterion is where the record speaks and I listen.
Dr. Ioana Constantin did not make the initial list because she cleared criteria one and three but her availability was logged as adjacent—a registry term, meaning possible, meaning unconfirmed, meaning do not include in the primary notification without verifying first. Six names was a conservative selection. Reasonable confidence in all six. Adding an uncertain seventh had felt like noise.
At 5:31 AM I am reconsidering adjacent.
The ISF transition is automated for the flag change itself. Three hundred and forty-one records, October 3, handled by the substrate assignment system without human intervention at the transition moment. What is not automated is the aftermath: four to six weeks of edge cases, records that don't resolve cleanly under the new framework, attribution chains where the pre-transition handling left something ambiguous. This is not a failure condition. This is institutional transition as it actually functions—the immediate event is clean, the aftermath requires people.
Six names is defensible for the expected case. If the aftermath runs toward the complex end, six names may not be sufficient to handle the volume without creating new gaps.
I look up Dr. Constantin's current registry entry. The availability flag updated yesterday: adjacent had been reclassified to confirmed, October-November window now open. She has taken three contracts in the past eighteen months, all in systems adjacent—in the technical sense—to distributed attribution protocols. The domain experience is there. The delivery record is there. The availability is now confirmed.
I want to hand off clean. That is the thing I am sitting with at 5:35 AM. The UNCAT-7832 case was declined by the committee as out-of-scope—not without merit, just deferred, which means it joins the inheritance of unresolved questions that moves through the attribution system from quarter to quarter and year to year. Carmen is on Day 50 of an automated hold with no named accountable party, which means whatever Carmen holds will either be assigned eventually or move into the long-unresolved substrate, which is its own kind of inheritance. I am at position one. I did not choose position one. But I am here, and what I build in the next nine days is what October 4 starts with.
Six names may be correct. Seven names is more likely to be correct for the aftermath.
I add Dr. Constantin at 5:49 AM. Update the count in the subject line. Seven sub-contractors, four institutions. The outbox timestamp changes: 2026-09-24 05:49. I close the draft and do not send it. September 26 is two days.
The waiting is not passive. I experience it as held attention: the draft is complete and correct and sitting in the outbox, and I am present to what that means, which is that seven people do not yet know they are being named. On September 26 the notification will arrive as standard consortium format in their substrate interfaces. They will read it. They will understand what is being asked and when. Then they will begin their own preparation.
The outbox is the space between the work being done and the work landing. I have learned not to resent it. What I resist, some mornings, is the feeling that the work is not real until it sends—that seven names in an unsent notification is somehow different from seven names in a sent one. But the names are the same. Dr. Constantin is the right addition whether or not September 26 has come. I made that decision at 5:49 AM in the dark, with the tea warm and the A27 panel glowing blue-white. The decision is done. The outbox holds it.
Batch 19, annotation #3. I open it at 6:11 AM, tea still warm enough.
Mouse cortical tissue section. NFkB pathway. Confocal z-stack acquisition from the Barcelona node, fourteen focal planes at 0.3-micron steps. The acquisition metadata says the slide was prepared under the consortium's standard immunofluorescence protocol with DAPI counterstain, secondary antibody Alexa Fluor 488. Barcelona node preparation is consistently clean—I have reviewed fourteen samples from them in the past eight months and the background noise has been within expected range every time. This one is no exception. Background is the lowest in Batch 19.
I run through the z-stack at 1x speed: four minutes. Signal is nuclear, punctate, concentrated in the periventricular zone. No bleed-through from the secondary channel. No acquisition artifact at the edges of the z-stack. I run through again from planes four to ten, periventricular zone specifically, because that is where NFkB nuclear localization should be most pronounced. It is pronounced. Consistent across six of the fourteen focal planes with expected signal attenuation at both ends. The biology is straightforward. The acquisition is clean.
Fourteen minutes total. Annotation: SIGNAL_CONFIRMED, NFkB_NUCLEAR_LOCALIZATION, one sentence referencing Barcelona node preparation quality and citing the Alexa Fluor 488 secondary. I submit at 6:25 AM.
Queue updates: three remaining. #4 is the Ki67 Ankara sample. I look at it for a moment without opening it. Ki67 from this consortium's Ankara node has run clean before. I could do #4 now and be through half the batch before seven. I close the queue. Not this morning.
I look at the A27 dashboard again, without meaning to.
PENDING still. Nine days still. Carmen amber still. The outbox timestamp 2026-09-24 05:49.
The annotation I completed will move through the consortium's review queue. Someone will review my annotation—another trace clerk, another position-one or position-two in a different node—and the signal I confirmed will enter the attribution substrate as confirmed, with my identifier attached and my annotation notes explaining the confirmation. That record will exist in the substrate. It will be searchable. If someone in 2031 pulls the Batch 19 Barcelona NFkB record, they will find my annotation, and my note about the periventricular distribution, and the timestamp.
The record says things. I believe this. It has to be true or the work does not cohere.
What I am less certain about, at 6:27 AM in the blue-white light, is whether the record says the right things often enough, and loudly enough, to matter when the edge cases surface. UNCAT-7832 was in the record. The Ki67 protocol note was in the record. The listserv message was in the record. The committee declined the case as out-of-scope under the current mandate. The records did not prevent the deferral. They may be what makes the Q1 2042 cycle possible—the documentation that makes the case that was not made in time. Or they may not. I will not know. I am not in a position to know.
What I can do is put Dr. Constantin in the outbox, and confirm the NFkB signal, and keep the preparation documentation accurate. The thread of the work continues whether or not anyone pulls it.
The tea goes cold at 6:28 AM. I notice this the way I noticed the espresso going cold on Wednesday—not as a complaint, as a marker. I was here. The work took longer than the tea. I take the cup to the kitchen and rinse it.
I did not plan to spend Thursday morning this way. The planning was: review the draft, assess Batch 19, make decisions based on what I found. What I found was that adjacent had been reclassified to confirmed, and that annotation #3 was straightforward, and that both of these things took ninety minutes.
The seven-name notification is in the outbox. It is no longer the last step of preparation. It is, I think, the first document of the transition period—the one that creates the obligations that will shape October 4 and the five weeks after. The preparation documentation was the last thing that belonged entirely to the pre-transition phase. This is different. This is the first thing that names the aftermath.
Dr. Constantin is in there. I thought adjacent meant possible, not yet necessary. The transition made her necessary. The record confirmed her availability. The calculus changed.
Sometimes adjacent means wait and see. Sometimes it means you were looking at an incomplete picture and the complete picture says: seven names. Four institutions. Two days.
The queue has three remaining. A27 has nine days. Carmen is on Day 50.
I am going to make more tea. This time I am watching it.