What He Chose Not To Document
Marcus Veil underlines the sentence at 10:13 on a Friday morning, and the sensation is not one of recognition but of its opposite — the feeling of seeing for the first time something that has been there for forty-three months.
Ambiguity in assessment timing should itself be documented as a methodological parameter, not treated as noise in interpretation.
The paper is a methods review from the Department's research division, published eighteen months ago, sitting in the stack of flagged materials he'd been meaning to read since the spring. The authors are arguing about contextual assessment protocols in long-term environmental monitoring. Branch 12 is not mentioned. He'd had no particular reason to think the paper applied to his situation.
He reads the sentence again.
The window of his office looks east toward the freight corridors, where the automated carriages run on Circuit Mile in commercial allocation time, the ATTESTOR-class beacons catching morning light at intervals. He doesn't look at the window. He is thinking about a decision he made in June of 2025.
Not a single decision. An accumulation of deferrals, each one reasonable, each one undocumented, each one shifting the ground beneath the next.
There had been thresholds in his thinking — implicit ones, ones he'd formed and revised without writing them down. When Branch 12 crossed 80% in late 2025, he'd registered something: a slight tightening of attention, an additional moment each Tuesday on the readings before moving on. He had not pressed the ASSESS button on the Cartograph-7 interface. 80% felt too early. He could not have articulated what 80% was too early for, but the certainty was there, a professional instinct: not yet.
When Branch 12 reached 84% in late 2026, the threshold moved. He'd adjusted without registering the adjustment. 84% felt like the reading was still resolving, still finding its level. The dormant-baseline protocol had been running for over a year by then. He trusted the protocol's patience. He decided — without deciding — that the threshold was higher. 87%, perhaps. Or whatever the system's built-in alert criteria specified. He'd never looked up what those criteria were, exactly. He had reasons for that too. Good reasons, probably. The criteria were designed for a different class of monitoring scenario, and applying them literally to a dormant-baseline case would have generated alerts that were technically accurate and contextually misleading.
None of this was in the record.
The FORMAL notebook is in his desk drawer, the one he'd labeled two days ago after the committee email arrived. Three sections have been forming in his mind since Thursday afternoon: record, methodology, limitation. The first two are straightforward. The third is where he's been sitting.
He reads the sentence a third time.
The authors expand on it in the following paragraph. When the assessor's own uncertainty level is the primary variable driving decision timing, and that uncertainty level is itself undocumented, the assessment methodology cannot be fully reconstructed from the record. The gap between the observed data and the decision not to act is invisible.
He has a specific memory of September 2027, when Branch 12 crossed 85% for the first time. The number had come up on a Tuesday morning in the usual dashboard format: an amber waveform display, the percentage in the upper right corner, the standard compliance notation below. Dormant-baseline readings came in at the level of background, not interruption. He'd processed it the way he processed the transit schedule on the window — visible, registered, not acted upon. Tuesday became Wednesday. Wednesday became the following Tuesday. 85% became 86.2% by October. He noted it and moved on.
He had never written, anywhere, what he thought 85% meant. He had never committed to paper the question he was holding in mind: at what number does this become something I should examine more carefully? The question was there. It was doing real work in his thinking. It was shaping his behavior at every Tuesday reading. It simply wasn't in the record. He was reading the weekly summary on a Tuesday morning, coffee on the windowsill, the transit band below busy with freight at the commercial allocation hour. He had looked at the 85% figure and thought: this is still within the expected range for a dormant-baseline system showing long-term directional drift. He had thought: I should check what the VERIFY-class criteria say about this range. He had not checked. He had closed the dashboard and moved to the next item in the week's queue.
That was not a decision not to act. That was a deferral of a decision, which is different, except that at the end of forty-three months the difference becomes academic.
He picks up his pen.
The limitation section will say this: he had not documented the threshold he was applying to determine when contextual assessment was warranted. He was aware, at various points in the monitoring period, that he held an implicit threshold — a provisional answer to the question of what reading would constitute sufficient reason to run ASSESS. That threshold shifted at least three times. He does not believe the shifts were inappropriate; he believes they were, in fact, reasonable adaptive responses to a system that was changing slowly and consistently in a direction that didn't clearly warrant escalation at any single point. The question of whether 43 months of those responses constitutes a methodology or a pattern of avoidance is exactly the kind of question the limitation section is supposed to flag.
He writes the sentence twice before he writes it once. The first version is too defensive. The second version overcorrects, sounds like an indictment. The third version says what it needs to say:
Limitation: undocumented assessment threshold. The monitoring record captures readings and the decisions made at each interval; it does not capture the evolving internal criteria the Verifier of record applied to determine when contextual assessment was warranted. This threshold shifted during the monitoring period. The shifts were not documented as such, and therefore cannot be evaluated as such.
He pauses. He considers adding: The Verifier believes the shifts were reasonable adaptive responses to the data, but this belief cannot be verified from available records and should not be treated as part of the record. He reads it over. He leaves it out. The record should not contain what he believes about things the record cannot verify. That is exactly the principle at issue.
He writes instead: This limitation does not alter the factual content of the record. Branch 12 readings increased consistently over the monitoring period. The absence of documented assessment criteria does not establish that the criteria were arbitrary — only that they are unavailable for review.
That is as far as honesty takes him and no further.
Outside, the Circuit Mile freight corridor is running on the mid-morning schedule now, the commercial allocation window past its peak. Two ATTESTOR-class beacons register each other's proximity and log the interaction. The carriages continue routing without deviation. He is aware of none of this; it is the city's ordinary background texture, present as birdsong is present, unlistened-to.
He closes the FORMAL notebook and picks the contextual weighting paper back up. He is on page forty-two now. The paper is better than he expected. The authors are careful, their argument moves without shortcuts, they are genuinely interested in the problem of when to formalize what is already implicitly known. He finds himself wanting to argue with a section on institutional incentives — not because they're wrong, but because they've taken the charitable interpretation without flagging it as such. He makes a note in the margin.
This is how the close-out report weekend will go. He will read the rest of the paper today. Tomorrow morning he will open the FORMAL notebook and write Section One: the record. The facts: forty-three months, seventy-one percent to eighty-seven, consistent directional upward since June 2025, no threshold events triggering automatic escalation, no contextual assessments run via Cartograph-7, manual annotation throughout. He will write it in the past tense, which is how all monitoring records end up — accurate, precise, and as flat as stone.
Section Two: the methodology. Manual annotation, not VERIFY-class. Daily readings from dashboard review, which is the standard approach for dormant-baseline monitoring in Traced. The decision to rely on visual inspection of the dashboard's amber waveform rather than querying the ASSESS function — he will document this as a choice, not a default, because it was a choice, even if it was a choice made so consistently that it stopped looking like one. He will note the absence of independent confirmation infrastructure. He will note that the monitoring was conducted by a single Verifier without secondary review. These are not admissions; they are facts about the methodology that any reader of the record should know.
Section Three he has already written in draft.
The two paragraphs in the FORMAL notebook are the most precisely true things he has written in the entire forty-three months. Precision of the backwards-looking kind is easier than precision in the moment; he knows this. In the moment, the uncertainty is real and the threshold is moving and the pressure is to keep watching, keep waiting, keep trusting the protocol. The record looks like a series of correct decisions made with adequate information. It is. The limitation section is what the record cannot show about itself.
He reads. The paper's argument develops: the problem of undocumented thresholds is not a problem of bad practice, but of the way monitoring systems are designed to record data without recording the interpretive frameworks applied to that data. The record shows what was measured. It does not show how the measurements were interpreted, or what the Verifier thought the measurements meant, or where the implicit line between normal-variation and threshold-event was being drawn at any given moment.
He has read enough methods papers to know that arguments like this are common in research contexts and rare in operational ones. In research, the methodology section is understood to be a constraint on claims; in operational monitoring, the record is understood to speak for itself. The gap between these conventions is where thirty years of institutional practice lives, and he is not going to fix it in the limitation section of a single close-out report.
What he can do is name it clearly. That threshold was not documented. That is the limitation. The reader can decide what to do with it.
He thinks about a conversation with a colleague, three years ago now, in the cafeteria on the lower level during the early-shift lunch hour. The colleague had been overseeing Branch 7 at the time — an active-monitoring case, the kind that came with VERIFY-class audit cycles and a secondary review team. She'd asked him how the dormant-baseline work was different from what she did. He'd said: quieter. She'd asked if quieter meant less rigorous. He'd said: no. She'd accepted that. He'd believed it at the time.
He still believes it, mostly. The rigor of long-term observational monitoring is not the same as the rigor of VERIFY-class active assessment, but it is a real kind of rigor — sustained attention, resistance to false positives, patience as a professional discipline. The Empiricists in the Department's theoretical division have argued for years that dormant-baseline monitoring is undervalued precisely because its rigor is invisible in the record; all the record shows is the absence of escalation, and absence is easy to mistake for inattention. He had always found that argument persuasive.
He finds it less purely persuasive today. The record also fails to show where the line between disciplined patience and undocumented avoidance was drawn, and that is a problem the Empiricists don't quite solve.
Branch 12, in the system two floors down that he cannot see from this office, continues running at 87% on its ambient sensor array. In thirteen days the dormant-baseline protocol will officially close. The active-branch protocol will take over, quarterly reviews, an assessment team, the full VERIFY-class apparatus finally applied to the thing he has been watching by himself since 2032. He does not know whether the new team will find the 87% figure significant or ordinary. He does not know whether what he has been watching is a system in slow deterioration or a system finding a new equilibrium that simply registers higher. He never ran the contextual assessment that would have told him.
He has underlined the sentence about assessment timing and moved on. The paper still has seventy pages. The freight carriages run on the corridor. The city does what it does on ordinary Friday mornings.
He reads.
Editor's note: This document is a monitoring record. It does not cover everything that was thought.