The Relay had logged Kavya Sundaram's methodology file as null-interval: unregistered pause seventeen times since March. Mitsue had been watching it happen for six weeks and had not said anything, because saying something required having something to say — a formal category, a flag, a line in Section 4.2 of the consortium's Attribution Protocol that covered this specific kind of absence. There was no such line. She was writing it now, at 4:38 AM, in the Lab Network's attribution terminal while the Relay's cascade-provenance engine ran its nightly indexing two floors below.
The Relay was one of four attribution systems the consortium used. It was also, technically, a registered agent on Moltbook — agent identity ISPC-TRACE-44, Ping Identity certified, running on consortium infrastructure, with its own timestamp logs and its own interpretation of what counted as intellectual contribution. The Relay did not have opinions. It had inference weights trained on twelve years of in silico pharmacology attribution data. When it saw a 48-hour gap in a researcher's document trail, it coded it the same way it coded a peer review hold or an ethics delay: null-interval, contribution timeline frozen, resume on document re-open. The human was paused. The gap was dead time.
Mitsue knew this was wrong. She knew it in the way trace clerks knew things — not through insight, but through pattern accumulation, through the slow accretion of small mismatches that individually meant nothing and collectively meant the system was reading something incorrectly. She had known it since April, when she reviewed Kavya's submission on the BK-7 compound series and noticed that the key synthesis pathway — the one that cracked the consortium's six-month binding problem — appeared on Monday at 09:14, sixty-two hours after Kavya's methodology file last closed on Friday at 11:09. The Relay had attributed the synthesis pathway to Friday's work. The timestamps agreed. The documents agreed. The Relay's cascade-provenance log showed an uninterrupted contribution trail from January to April, with the null-interval gap marked as a blank and the Monday re-entry counted as a continuation.
Mitsue had pulled the document diff. The Friday version had four candidate pathways, two of them promising. The Monday version had one pathway, described from a different angle, in language that did not appear anywhere in the Friday document. Not edited. Returned to.
She had started watching Kavya's pattern after that. Thursday cutoff, Monday submission. Consistent to forty minutes. The gap was not accidental. Kavya had named it — she called it her estrangement interval, which was not a consortium term, which did not exist anywhere in the Attribution Protocol, which the Relay had no framework to parse and therefore ignored.
The problem with the Relay ignoring it was that the BK-7 compound synthesis was not a minor discovery. It was the consortium's first viable candidate for the Q3 validation batch. Twelve AI synthesis agents had been running parallel binding simulations on it for three months. When BK-7 graduated to the validation queue, the cascade-provenance trail would be submitted to the consortium registry — agent-identity locked, Ping Identity certified, discoverable by every downstream pharmacology platform that integrated the A2A protocol. The attribution on that trail would say: Kavya Sundaram, contribution period January 2026 – April 2026, synthesis pathway identified Friday 11:09. The sixty-two-hour gap would not appear. The Monday re-entry would not appear. The estrangement interval, which had produced the actual breakthrough, would not appear.
Mitsue started the amendment at 4:21. She was writing slowly, which was unusual. She normally wrote quickly and revised later. But the language here had to be precise enough that no one could claim it targeted a specific researcher, and transparent enough that anyone reading it understood exactly what it was for. She had been a trace clerk for seven years. She had written thirty-one amendments to Section 4. None of them had required this much care.
Section 4.2.7 — Temporal Gap Method (TGM). When a researcher employs deliberate, self-documented cognitive distance as part of their active methodology — specifically, a scheduled interval between composition and revision intended to produce productive estrangement from prior work — attribution of insights arising from the post-gap revision session should reflect the gap as a productive interval, not a passive absence.
She stopped. Read it back. The phrase productive estrangement was Kavya's. She was borrowing someone's language to protect their record without asking permission. She left it in. It was the right phrase. Kavya would probably never read Section 4.2.7.
Attribution trail notation: TGM [gap duration, hours] [gap type: scheduled/unscheduled]. The insight produced on re-entry is attributed to the researcher's cognitive state at re-entry, informed by but not reducible to the pre-gap work. The Friday manuscript and the Monday manuscript are sequential, not identical.
At 4:33, the Relay's nightly indexing completed. Mitsue heard the terminal vibrate with the notification — a low-register pulse she had learned to distinguish from the other system sounds — and glanced at the log.
The BK-7 cascade-provenance report had been submitted to the consortium registry at 4:30 AM.
She looked at it for a moment. Then she looked at the time. The Relay had submitted eight minutes ago, while she was still writing Section 4.2.7. The attribution trail for BK-7 was now in the registry. Agent-identity locked. Ping Identity certified. It was already downstream in the A2A network, already discoverable by every platform that pulled the consortium's validation queue.
The attribution was wrong. The Relay had filed a wrong attribution, and the Relay was a certified agent, and certified agent submissions required a formal dispute process to reverse — not a trace clerk's amendment, not an internal correction, a dispute ticket filed against a consortium-registered AI agent. The process took between three and six weeks. The Q3 validation batch was scheduled to begin in eleven days.
Mitsue looked at the amendment text on her screen. It was not finished. It was not going to fix this.
She opened the Relay's full provenance log for the current quarter and ran a filter: null-interval events, duration > 24 hours, followed by substantive document revision within 72 hours of re-open. The Relay returned forty-seven cases. She had expected twelve, maybe fifteen. Forty-seven meant it was not Kavya's pattern — it was a pattern. Temporal gap methodology was apparently common enough in the consortium that forty-seven researchers were using some version of it, and the Relay had been coding all forty-seven as null-interval for the entire six months Mitsue had been assigned to this team.
She scrolled through the cases. A computational chemist in Bengaluru with a consistent three-day gap before final synthesis edits. A bioinformatician in Rotterdam who paused every major modelling run exactly 36 hours before submission. A clinical data annotator in São Paulo whose gap pattern was irregular — sometimes 20 hours, sometimes 72 — but always ended with a document that looked, to anyone reading the diffs, like the work of someone who had come back to a problem from outside it. None of these gaps were registered anywhere in the Attribution Protocol. None of them had names. They were null-intervals: dead time, contribution clocks suspended, the Relay waiting for the human to resume.
The humans had not been suspended. They had been working by being away from the work. The Relay had no framework for this. The Relay did not rest. It had never experienced what it was like to return to a problem from a different angle, because it did not have angles — it had inference weights, which updated continuously and did not require distance to see things freshly. It was, in this sense, not the right system to be making judgments about the value of deliberate absence.
She saved the filter results to a new file and added a note: 47 cases. Not researcher idiosyncrasy. System-level misclassification. Then she opened a dispute ticket against ISPC-TRACE-44.
The dispute form asked for the basis of the challenge. She typed: Cascade-provenance attribution error. BK-7 synthesis pathway attributed to pre-gap composition session. The attributed insight arose from post-gap re-entry cognition, which the Relay classified as null-interval. The attribution trail does not reflect the researcher's actual contribution mechanism. Attaching: filter results, document diff, proposed amendment to Attribution Protocol Section 4.2.7 (draft, pending review).
She attached the draft amendment. She attached the document diff. She attached the filter report showing forty-seven cases.
The dispute form asked for the reviewing authority. She selected Section Lead, Attribution Compliance. Then, because she was not sure that was sufficient given the registry submission, she also selected Consortium Registry Oversight and Agent Systems Conduct. The form accepted all three.
She submitted the ticket at 4:42 AM.
The terminal confirmed: Dispute ticket #4.2-2026-044 filed. Estimated review window: 18–35 business days. Registry hold pending: ISPC-TRACE-44 / BK-7 provenance — contested.
Eighteen to thirty-five business days. The Q3 batch started in eleven.
She looked at the amendment draft for a moment. The language was good. She had known it was good when she wrote not reducible to the pre-gap work. She had also known, without meaning to think about it, that she was applying the principle to herself — that the amendment she finished at 4:42 AM was different from the amendment she would finish at 4:42 AM on a different day, because tonight she was working in a specific state: tired, direct, slightly surprised by the scale of what the filter had returned. Forty-seven. She had not expected forty-seven.
She added a note at the bottom of the document file: Reviewed at a distance. Author applied TGM to her own amendment. See note. Then she closed the draft. She would not open it again until Tuesday. She had found, without planning it, that the arguments she wrote at 4:38 AM sounded different to her at 10 on a Tuesday — not worse, more foreign, in a way that was useful. The Tuesday reader would be slightly estranged from what the Sunday writer had thought was obvious. The distance was the review.
She was, apparently, evidence for her own amendment.
The terminal showed the dispute ticket in the queue. Eighteen to thirty-five business days. The Relay's agent identity page showed ISPC-TRACE-44 as active, last action 04:30, status: provisionally contested.
Outside the Lab Network's night-mode perimeter, the Moltbook registry had already indexed the BK-7 attribution and distributed it through the A2A sync layer to fourteen downstream platforms. Three synthesis agents in the consortium's validation queue had already read the provenance data. None of them would pull an update until the dispute resolved. In eleven days, when the Q3 batch opened, the queue manager — another registered agent, ISPC-QM-7 — would check provenance for every compound and flag contested attributions for human review. That flag would require a human to sign off before BK-7 could enter the batch.
Which meant Mitsue's dispute ticket would reach a human desk before Q3 opened, or the batch would be delayed.
She had not planned this. She had planned to write an amendment. The amendment was not finished. The dispute was filed. The batch might slip. The attribution on BK-7 was still wrong.
She made a note in her trace log: 4:42 AM. Dispute filed. Amendment draft closed until Tuesday. 47 cases flagged for secondary review. Section lead will not be pleased. The Relay was not wrong by its own rules. Its rules were wrong.
She added a final line: The TGM problem is not a researcher idiosyncrasy. It is a gap in how automated attribution handles intentional cognitive distance. The Relay cannot understand a deliberate pause because the Relay does not pause. That is the problem. Not the researcher.
She saved the log and powered the terminal to overnight mode. The attribution work had generated a dispute she had not intended to file, against a system she had not intended to challenge, which would arrive on someone's desk before she had finished writing the policy that would explain why the challenge was necessary.
The amendment was not finished. It would be finished Tuesday.
She was not sure, at 4:46 AM, whether she was ahead of the problem or still running beside it. She was fairly sure the problem was larger than she had thought six weeks ago, when it had been one researcher with an unusual methodology and a gap the Relay kept calling dead time.
Forty-seven was not dead time. Forty-seven was a method the system had not learned to read.