The third paragraph had one sentence in it. Marcus had written it at 9:41 AM and it was now 10:13 AM and the sentence was still there.
He had written, deleted, and rewritten the first two paragraphs of Section Two during the previous ninety minutes, and both of them were adequate. The first paragraph described the dormant-baseline monitoring protocol: what it measured, how often, which VERIFY-class instruments handled the primary corridor feeds. The second paragraph described why dormant-baseline had been selected over active contextual assessment in 2032, when Branch 12 was new and the comparison dataset was thin. He had written that paragraph four times because it kept becoming something it was not supposed to be — an argument rather than a record — and finally on the fourth attempt he had simply written what was true: the monitoring protocol was selected to produce a record that could be reviewed without active interpretation artifacts.
That was Section Two, paragraph two. It said what it needed to say.
The third paragraph had begun, as he wrote it at 9:41, with: In the forty-three months of monitoring that followed, active contextual assessment through the Cartograph-7 system was not engaged, and therefore —
That was as far as he had gotten before the problem appeared.
The problem was the word therefore. Whatever followed it would be a claim about what Cartograph-7 would have produced if it had been engaged. He had reached for that sentence because it seemed like natural methodology writing: you describe the tool you chose, and then you note what the tool you did not choose would have told you differently, so the reader understands the scope of the decision. This was how Marcus had been taught to write close-out documents. You show the decision tree. You account for the paths not taken.
But the Cartograph-7 data did not exist. There was no contextual weighting analysis for Branch 12. For forty-three months, the ASSESS button had remained unclicked. The system was there — the dormant interface was there on every VERIFY-class instrument across the Circuit Mile corridor, amber waveform steady, ASSESS gray and untouched — but it had not been activated. Not once. Whatever Cartograph-7 would have said about the rising pattern, about the June inflection point, about the 87.1% at close, was not in any record Marcus had access to. It was not missing from the close-out report. It was missing from existence.
So the word therefore had nowhere to go.
He could write: and therefore no contextual adjustment was applied to the baseline readings. That was technically accurate. But it described only what had not happened, not why the readings themselves were now difficult to interpret without the contextual layer. It made the absence sound like a neutral technical fact rather than a choice with consequences.
He could write: and therefore the relationship between the rising pattern and the surrounding environmental context is not recoverable from the record. That was closer to true. But it described a limitation, not a methodology, and limitations belonged in Section Three. Section Two was about what he had done. Section Three was about what that foreclosed. If he put the limitation in Section Two, Section Three would have nothing to say that had not already been said, and Section Three was the section he had been thinking about since Thursday night.
He tried three more versions. All of them had therefore in them. All of them tried to explain what Cartograph-7 would have done. None of them worked because none of them were true — they were speculative reconstructions of a dataset that had never been created, dressed up in the language of factual methodology.
At 10:08, the ambient monitoring system on the desk terminal cycled its hourly confirmation: Branch 12 — ambient pattern stable — 87.1% — ATTESTOR beacons active: six eastern stations, Circuit Mile north segment. The confirmation text was formatted in the pale-green fixed-width that had been standard since the Branch 12 instruments were installed, unchanged in forty-three months, identical to the confirmation that had appeared every hour of every operational day of the monitoring period. The ASSESS button was gray below it. The button had been gray every time that confirmation appeared. It would continue to be gray now that the committee had voted to close the branch.
The coffee was cold.
Marcus got up to make a second cup. While the kettle heated he stood at the window. The street below was Saturday-quiet — this was a residential block, not the corridor side where the freight-carriage routes ran. No ATTESTOR-class markers here, no embedded count sensors in the pavement, just a dog walker and two people sitting on the steps of the building across the road. He could see a sliver of the Circuit Mile elevated structure four blocks east, the morning light catching its upper edge.
He thought about the third reason he had not written in Section Two, paragraph two.
The first reason for choosing dormant-baseline was the comparison dataset: in 2032, Branch 12 was too new to have the historical baseline that active contextual assessment required. Cartograph-7's contextual weighting methodology was calibrated against long-term Branch averages, and Branch 12 did not have them yet. That was a legitimate technical reason. He had written it in the second paragraph.
The second reason was institutional convention: dormant-baseline was the standard protocol for newly commissioned branches in the period. Active assessment was reserved for branches with documented anomaly flags. Branch 12 had no flags in 2032. That was also in the paragraph.
The third reason was harder.
The third reason was that contextual weighting would have changed the character of what the record could say. Dormant-baseline produced readings that could be audited by anyone with access to the VERIFY-class instrument logs. Active contextual assessment produced readings filtered through Cartograph-7's weighting model — and if the model's parameters were ever questioned, the entire forty-three months of data became contested rather than verifiable. Marcus had wanted to produce something that could be reviewed by the committee without arguments about whether the weighting had been applied correctly. He had wanted the record to be auditable independent of any methodology dispute.
That was reasonable. It was also, he had been thinking since Thursday, why the record did not contain the information that would make it possible to explain what 87.1% actually meant in context. The dormant-baseline had preserved the record's auditability. It had also preserved the uncertainty inside the record. Both things were true at the same time, and that was the kind of thing that belonged in Section Three.
He poured the water over the grounds and waited.
The choice to use dormant-baseline was the methodology. Not what Cartograph-7 would have said. The choice.
He had been trying to explain the absence of data he did not have, rather than documenting the presence of a decision he had made. Section Two was not responsible for counterfactuals. Section Two was responsible for what Marcus Veil, in the autumn of 2032, had decided to do. He had decided to monitor Branch 12 with a dormant-baseline protocol rather than activate contextual assessment. The reasons were in paragraph two. The sentence about therefore was trying to do something Section Three was supposed to do — and Section Three could not do its work properly if Section Two had already colonized its territory with half-formed limitations disguised as methodology.
He took the coffee back to the desk.
He deleted: In the forty-three months of monitoring that followed, active contextual assessment through the Cartograph-7 system was not engaged, and therefore —
He wrote: The monitoring protocol was selected to produce a record that could be reviewed without active interpretation artifacts. Active contextual assessment through the Cartograph-7 system was not engaged during the monitoring period. The decision and its consequences are addressed in Section Three.
He read it twice.
It was twelve words longer than anything else in the section. It was also, he thought, the first honest sentence he had written about the choice since he had made it. The first sentence documented what he had wanted. The second sentence documented what he had done. The third sentence was a forward reference that told the committee something significant was coming — which was true, and which was better than pretending Section Two could contain it.
The ASSESS button was gray on the terminal screen. Branch 12 was at 87.1%, the same number it had been at close, the same number it would remain at in the record until the record was superseded by whatever came after. The ATTESTOR beacons on the six eastern Circuit Mile stations had sent their automated confirmation at 6:15 that morning, as they did every morning, reporting that the sensors were functional and the monitoring was ongoing and nothing required immediate attention. The branch was dormant. The button was gray. The methodology was documented.
Marcus saved the document and opened Section Three.