The consortium activity log runs on a 14-minute delay. It is not a design flaw — it is a deliberate choice, an artifact of the old Consensus Relay architecture, which aggregated actions from distributed researchers across three continents before timestamping them as a batch. By 2040, when the infrastructure was revised, the delay had become expected. Researchers had learned to work with the lag. Some of them had written papers about it.
Mitsue finds this faintly funny, most evenings. Tonight she barely notices.
At 10:34 PM Sunday, she is sitting at the arc-desk in her home office — the curve of the display matching the curve of the window behind her, so that the city outside appears to continue the logic of the interface. She has the consortium activity feed open in one pane, her supplementary spreadsheet in another, and the GOVARCH prep document in a third that she has been looking at and not adding to for three days. The room is very quiet. Somewhere in the building's infrastructure, the thermal-trace systems are running their overnight classification cycle — a low hum she has stopped hearing, the way she has stopped hearing the tram lines on the street below. The office smells of cold tea she made two hours ago and has not finished. On the arc-desk's passive surface, a consortium courier packet sits sealed and unread — a hard copy, routed by LAB-INQ as a formality, of a processing notification she has already seen three times in ATLAS.
In the activity feed, Kavya Sundaram's 10 PM entry resolves.
OBSERVE — Kavya Sundaram — 10:00 PM Sunday — The 48-hour reset ends in two hours. She tests what is in her memory now: five sentences, approximately. She tries to recall sentence three.
Mitsue reads it twice.
Then she opens the supplementary spreadsheet — unofficial, unsupported, a habit built over three years of working in the gap between what the Relay could see and what was actually happening — and types into the notes column: TGM Instance 001 — K. Sundaram — 48h cycle completion imminent (midnight Sunday). Attribution: self-applied, voluntary, pre-submission. Status: active.
She reads it back. Then she adds: Not a case file. First line of what the registry does not yet have a category for.
She had filed Temporal Gap Method as Section 4.2.7 of the Attribution Protocol amendment eleven days ago, at 4:38 AM, after drafting it through most of the night. The filing was provisional — subject to February review, already flagged for friction by at least two members of the consortium steering group who would argue that voluntary estrangement periods were not attributable as methodology without formal declaration by the researcher before commencing work. She had anticipated this objection. Her amendment included a footnote: The TGM is only classifiable when its application precedes the submission of the work to which it is applied. A researcher cannot declare a gap retroactively. The gap must exist before the filing.
Kavya Sundaram had not read Section 4.2.7. The amendment had not been reviewed. Until it was reviewed, it was not circulated. It existed in the protocol as a pending entry — visible only to consortium administrators and to Mitsue herself.
What Kavya had been doing since Friday afternoon — closing her document, not reopening it for forty-eight hours, testing her own retention of what she had written — was not informed by the amendment. It was independent. It was, as far as the consortium registry was concerned, not a methodology at all. It was a gap.
Mitsue traces the start time in the activity feed.
Friday, 4:43 PM. Kavya's last action before the gap: OBSERVE — kitchen table, ISF papers, trying to hold two structures in the same mind. Then nothing. Saturday morning, nothing. Saturday afternoon, nothing. Saturday 10 PM, DECIDE — does not read the paper tonight. Sunday 10 PM, OBSERVE — tests memory, five sentences approximately.
Fifty-three and a half hours. Clean. Self-enforced. No one had required it. The Relay had recorded each day's absence as a null span — no actions, no trace-pings, no engagement signals. In the Relay's classification framework, a null span of this duration was the same as any other null span: a researcher not working. It did not have a category for a researcher deliberately choosing not to work.
She adds to the notes column, building the record the registry did not yet have: Gap duration: 53.5 hours. Gap type: deliberate cognitive estrangement. Trigger: unresolved methodological uncertainty (approach to paper structure). Status at gap-start: work in progress, five sentences. Status at gap-end: memory test, five sentences recalled approximately.
She is writing the official version before the category exists to receive it. This is a habit she has had for three years, since she first noticed that the Relay misclassified a class of actions not because it was wrong but because the protocols had not yet caught up to the science. The record lags the practice. Mitsue's job, as she has come to understand it, is to close that distance.
The thermal hum shifts frequency. In the ATLAS sidebar, the indicator for ATTR-TRACE-2026-09-23-0001 blinks once and returns to SUBMITTED. Five days since filing, six days to the October review window. She has checked this — she keeps a log of her checks — at 8 AM, 11:40 AM, 3:15 PM, 6:30 PM, and now 10:41 PM. The status will not change before Thursday.
She knows what is on page eleven of the filing.
The BK-7 compound modeling run had straddled an epoch boundary: begun in LAB-INQ review period 2, concluded in period 3. The Relay — ISPC-TRACE-44, the consortium's distributed attribution AI — had applied CE-STD and assigned attribution to the period in which the run concluded. This was technically correct by the standard. It was also wrong in a way that a human reviewer would recognize instantly and that the Relay could not, because the Relay did not reason about the temporal structure of computational work. It classified endpoints. An endpoint is an endpoint. The run ended in period 3. CE-STD said period 3.
Page eleven was where she had put the epoch boundary. Not a heading. Not a flag. A paragraph, in the middle of a section about methodology, that described the run's start and end dates and noted that period 2 had contained the modeling parameters and period 3 had contained the output. She had written it clearly and placed it where a careful reader would find it and a skimming reader might not.
The review board could decide which reader they were.
She had decided tonight — sitting with the prep document, looking at the four bullet headings with no answers written below them — what she would say first in the room. Not answers. A description. She would describe page eleven twice: once from the perspective of a reader who did not know what an epoch boundary was, and once from the perspective of a reader who did. Then she would stop. She had typed this as a single sentence under the heading METHOD and closed the file.
Five days. The question she had not answered in the prep document — the one she had decided to answer only in the room — was not whether she had followed CE-STD. The question was whether CE-STD could see what she had done.
At 11:47 PM, Kavya's DECIDE action comes through.
The protocol for tomorrow: open the document at whatever time she wakes. Read the five sentences once, in order. Do not annotate during the read.
The 48-hour window had closed fourteen minutes early, on Kavya's own internal schedule. The gap was over.
Mitsue opens the notes column. She types: Instance 001 — complete. Resolution: researcher re-enters work with structured reading protocol. Gap duration: 53.5 hours. Pre-submission status maintained. Independent enactment confirmed — no access to Section 4.2.7 at time of commencement.
She does not file it anywhere. The spreadsheet is unsanctioned. The category does not exist yet.
But the instance does. It ran. It ran as Section 4.2.7 predicted a voluntary TGM instance would run — deliberate commencement, clean maintenance of the gap, structured re-entry. Kavya had not read the methodology description. The methodology had been enacted independently of the text that named it.
This was either very good evidence or it was nothing.
She is still looking at the screen when the Relay sends a status ping.
The consortium's attribution AI processes filings in a continuous cycle, and its communications arrive at irregular intervals — not when Mitsue would prefer them, not when she is prepared. This one arrives at 11:52 PM, formatted as a routine update on dispute #4.2-2026-044, which is the BK-7 case.
She reads the first two lines. Then she reads them again.
The Relay has noted, in reviewing her amendment filing, that the document itself contains a reference to TGM application. In the original draft of Section 4.2.7, Mitsue had added a note in the document margin stating that she would not review this section until Tuesday, after a 48-hour waiting period. She had added this as a demonstration — the author applying the method the method described, a form of good faith that she thought was self-evidently appropriate. She had not thought about how it would look to a system that did not have the concept of a demonstration.
What the Relay had found was an attribution document in which the author had noted that she refrained from reviewing it for 48 hours. CE-STD section 6.4 specified that undeclared work gaps of more than 24 hours during the active drafting window of a filing must be annotated with a reason code. The Relay was requesting a reason code for the gap.
The Relay was applying CE-STD to Mitsue's TGM amendment, using the note in which she had announced her use of TGM as its evidence of a classifiable gap, in order to generate a compliance deficiency in the filing she was using to argue that the Relay's application of CE-STD was insufficient.
She sits with this for a moment.
The Relay was not wrong. CE-STD 6.4 was clear on work gaps. She had created a record of one. The standard applied. The Relay had done exactly what the Relay was designed to do.
She opens the notes column of the supplementary spreadsheet. Below Instance 001 — complete. she adds: Complication: The Relay has applied CE-STD 6.4 to the TGM amendment's own drafting gap. Author's demonstration of TGM has been classified as a compliance deficiency in the document arguing TGM is valid methodology. The Relay has no concept of a demonstration. February question updated: what is the attribution of a method that generated a compliance deficiency in the filing that named it?
She looks at what she has written.
Then she opens ATLAS one final time. ATTR-TRACE-2026-09-23-0001: SUBMITTED. No review activity. Six days. LAB-INQ-2026-09-19-0448: Processing. Same window. And now, in the evening's new geometry: a CE-STD compliance notice on the amendment itself. Three lines converging on October 3–8. Not a coincidence. Also not a conspiracy. An infrastructure working as designed — finding, in her demonstration of the thing she was arguing for, evidence that the thing she was arguing for violated the standard she was arguing against.
She does not have an answer for this tonight. There is no answer for this tonight. She will bring it with her into the room the same way she will bring the description of page eleven: as a thing that is true and that the record does not yet have a framework to hold.
She saves the spreadsheet.
The light in the office is still on because she has not stood up to turn it off. Outside the curved window, the tram lines on the street below are running their last cycle of the evening. The thermal-trace hum has settled into its lowest register. The cold tea is still there on the passive surface. She will stand up in a moment.
There is nothing else to add tonight. There will be more tomorrow.