The delta feed loaded at 10:23 AM and Mitsue Hoshino read through it twice before she accepted what it told her.
Four passes. Kavya Sundaram had walked past the face-down page on her desk four times since 7:33 AM Saturday. Kitchen and back. Window and back. The ISF terminal — idle since Friday afternoon — and back. The coffee maker and back. The page had not moved.
Mitsue closed the delta window. The workstation she used on Saturday mornings was in the secondary attribution office on the Lab Network's third floor — a narrow room with two long desks, usually empty on weekends, the climate system dialed back so the air carried a faint mechanical dryness. The overhead panels ran at sixty percent, which made the screens bright by comparison, everything washed in the blue-white of interface light. Through the window on the east wall she could see the underside of the transit pod corridor that ran between the research floors, its polymer surface faintly translucent, grayed out against the overcast morning. The Lab Network ran seven days; the building did not rest. But Saturday had a texture — quieter ambient data flows, fewer routing events in her feed, the soft background hum of automated systems running without human oversight.
She looked at her instrument, open to the summary dashboard. Three open attribution states, color-coded amber: A27, UNCAT-7832, correction_validity_status for Entry Thirty-Two. In the Lab Network's distributed workflow, amber meant pending external input. It was not a warning color — it was an information color — but Mitsue had been staring at three amber indicators all week and something in her had started reading them as failure.
She had come in this morning hoping for an update. That was the honest word for it: hoping. Not expecting, not anticipating — hoping, which in attribution trace work meant wanting external events to occur according to a preferred schedule. She had checked the GOVARCH queue status at 8:47 AM. Day 1. The governance archive acknowledged receipt in under four hours, which meant the filing was clean, which meant the 5-10 business day window had started running Friday morning. The earliest possible response was Thursday, September 25. Today was Saturday, September 19. She had opened the queue at 8:47 AM anyway.
She had also checked the routing batch schedule, which displayed only its standard holding message: Next attribution pass: [dynamic]. Contact your regional batch coordinator for inquiry windows. A27 was at position 2. A27 would advance to position 1 in the next pass, whenever that occurred. The handling flag — a priority marker set by the routing system's pre-assessment layer, indicating the record required additional verification before moving to position 1 — had been active since the 06:00 UTC batch on September 18. She knew this. Checking the schedule did not make the pass occur faster.
She was very aware that she had been doing these things.
At 9:48 AM she had opened the Lab Network's batch inquiry portal and submitted a window request for A27. The portal generated a case ID — LAB-INQ-2026-09-19-0448 — and returned the standard response: Inquiry received. Window requests are processed during standard coordination hours (Mon–Fri 08:00–17:00 local). Expected response: 2-5 business days. She had submitted the request on a Saturday, knowing this. She had submitted it anyway. The case ID sat in her record as evidence of wanting the cascade to move faster than it was moving.
Mitsue pulled up Carmen's record. Day 43. Status: NOTABLE. Fields: null. Hold type: automated routing hold. Duration: unknown start. Automated routing holds were issued by the Lab Network's triage delta — a non-human allocation layer that had access to information about subjects that did not flow through standard attribution channels. The triage delta did not explain its holds. It flagged, held, and released on its own schedule, with its own criteria. Mitsue had worked in the Lab Network for eleven months. In that time, the triage delta had issued seventeen holds on records she was actively tracking. Fifteen had resolved within two weeks. One was still open at six weeks. Carmen was day 43.
Day 43 was not unusual for an automated hold. Day 43 of null field values was harder to read. The triage delta knew something the instrument couldn't represent. She had tried twice this week to identify what trigger condition would produce a NOTABLE flag on a record with completely null metadata. She had not found an answer. The triage delta's criteria were not documented in any resource she had access to.
She switched back to the delta feed and read Kavya's four passes again.
7:33 AM Saturday September 19. ISF terminal idle since Friday afternoon. Face-down page on desk. Walked to kitchen.
7:41 AM. Walked to window.
8:02 AM. Walked to idle ISF terminal. Did not open it. Walked back.
9:15 AM. Walked to coffee maker.
The face-down page contained Kavya's scenario tree for UNCAT-7832. Mitsue had never seen it. It was Kavya's document, and she had access to the feed only because they were both named in the GOVARCH filing — the governance archive request linked their attribution chains, which meant the Lab Network's workflow layer shared relevant environmental context between their records. It was not surveillance. It was the routine cross-referencing that made distributed consortium work possible. Every dweller in the Lab Network whose records linked to UNCAT-7832 received environmental context from the others. Mitsue received Kavya's activity feed because they were working the same question from different angles: Kavya from the ISF methodology side, Mitsue from the attribution trace side, the triage delta from whatever angle it operated on.
Kavya had read the scenario tree Friday afternoon. She had identified an anchor stability problem in chapter four of the ISF methodology paper — the possibility that the 0.74→0.77 threshold revision in 2031 had violated the stability criteria the methodology was built on, which would put the structural corrections in the A28 batch in a different light. Mitsue had invented a notation Thursday at 4:42 PM because she had no existing field for the situation: correction_validity_status, values: not_questioned, under_scrutiny, confirmed, refuted. Entry Thirty-Two was under_scrutiny. The instrument marked its own observation horizon: before 4:42 PM Thursday, the question didn't exist in the record. After, it did.
Kavya knew what the scenario tree said. Walking past it four times did not mean she had forgotten it. It meant she had chosen not to re-engage with it until the sequence permitted — GOVARCH response, then the October 3 batch, then chapter four re-read, then the scenario tree with full information. Turning it over now would give her the same question she already held, without the answers that were coming.
Mitsue understood this so clearly it surprised her.
She looked at her three amber indicators. A27. UNCAT-7832. Correction_validity_status. She had spent the week inventing notations because the instrument didn't have fields for the situations it was encountering. Provisional-unresolved: a record that cannot close because a necessary input hasn't arrived. Provisional-superseded: a record functionally replaced but not formally closed. Correction_validity_status: a field for "the basis of this entry is under active scrutiny." The notations were good. They described real situations accurately. But she had invented them under a pressure she hadn't named yet — the feeling that amber meant something was wrong, that open meant failure.
It was not failure. It was the structure of the problem.
Cascaded holding.
The phrase arrived without ceremony. She was looking at the three amber indicators and the word for their structure appeared: cascaded. Each open state depended on an external trigger. Each external trigger depended on a prior one. GOVARCH depended on the archive having the ISF deployment timeline. October 3 depended on GOVARCH or the batch running independently. Correction_validity_status depended on Kavya finishing chapter four, which she had deferred until closer to October 3. The cascade ran: deployment timeline → governance archive → October 3 → chapter four → correction validity. Mitsue was at position 0. Kavya's scenario tree was at position 4.
The face-down page was at position 4.
Kavya was not avoiding it. She was waiting for positions 0 through 3 to run.
Mitsue opened a notation window and typed: Cascaded holding: a state in which multiple open records each require external input, where the inputs are sequentially dependent. Resolution requires the cascade to advance from its earliest position. This is not a malfunction. It is an accurate representation of a situation in which the information needed for closure has not yet arrived.
She looked at it for a moment and did not add it to the instrument's schema. It was not a new notation type. It was a description of what the existing notations already captured. Naming it didn't change the instrument. It changed her understanding of the instrument, which was a different thing and didn't require a field.
Outside, a transit pod moved through the corridor, its underside catching the diffuse morning light, casting a brief shadow across her desk. She watched it pass. The shadow lasted three seconds. Then the light returned, flat and even, and the corridor was empty again, and the three amber indicators were the same amber they had been.
She pulled up Carmen's record one more time. Null fields, automated hold, day 43. The delta feed for Carmen's region showed no activity today, which was consistent with a hold: held records didn't generate workflow events, didn't receive routing assignments, didn't appear in batch queues. Carmen was in the Lab Network but outside its normal flow. Present and unmoving. Like A27 at position 2 with the handling flag active. Like Kavya's scenario tree on the desk. Like the GOVARCH acknowledgment, sitting in a queue that would produce no response until September 25 at the earliest.
The triage delta had assessed Carmen as NOTABLE — the highest attention tier — and then held the record rather than escalating it. Flagged-and-held was different from flagged-and-escalated. The delta flagged and held; it didn't route to human review. It was waiting for something too. Its cascade was not visible to her.
She thought about this for a while. Four parties waiting: Mitsue, the triage delta, Kavya, and the Lab Network's October 3 batch process. All of them holding open states that depended on something else moving first. All of them accurately representing their current situation by not forcing resolution before the input arrived.
The face-down page would stay face-down until the sequence permitted. Kavya would walk past it again this afternoon, probably. And again this evening. The question it carried was already in her head — she did not need the page to think about the question. She would return to it when returning to it meant something.
Mitsue closed the GOVARCH queue window. She looked at the three amber indicators. She thought: I cannot accelerate the cascade. I can only record it accurately at each point in time. Today's record is three open states, correctly open, all depending on triggers that have not arrived. That is what today is.
She did not fully believe this. But she wrote it into the day's instrument log as if she did, because the log did not have a field for partial belief. The field was: observation. The observation was accurate.
She closed the notation window. The secondary attribution office was quiet — no footsteps in the hall, no voices from the adjacent data room, just the climate system's low mechanical hum and the blue-white light of the screens. She could hear, faintly, the soft transit pulse of another pod through the corridor outside.
She went to make coffee. Down the hall the coffee machine was the old pull-lever type — the Lab Network had replaced most equipment in 2024 but the secondary attribution office still ran on its original fittings, which she had always privately appreciated. Things that did not change when they did not need to. She pulled the lever. The machine groaned and delivered.
When she came back, the three amber indicators were still amber. Of course they were. She set down her cup and went back to work.