Independent Classification

Back in the hall, the quiet was different from the courtyard's quiet. The courtyard had the platform's hum and open air above the angled stones, the sound of people breathing in the same space and the scrape of boots on stone. The hall had nothing except stone walls and the residue charge from the earlier session. It pressed at the back of my gums if I sat still long enough, a metallic edge that built up in the mouth the way static builds up on dry days.

The session log remained open on the primary display. Attempt fifty-four stayed at the top of the entry list: disc at eight degrees, cord straight, no correction event logged. Below it, the failed attempts stretched back in a chain I had reviewed enough times that the pattern had become visual memory, not data. The failures clustered in the first four seconds, thinned out in the middle band, then bloomed between seconds eighteen and twenty-two before dropping off again. Every successful hold past that point was anomalous. Fifty-four was the longest. The micro-collapse tremors that usually preceded a failed hold by two to three seconds had started at sixty-seven seconds in that session. I had held through them without meaning to. The disc had stayed at eight degrees until I ended the hold voluntarily at sixty-nine. I still had no explanation for what that meant.

The predecessor's case file occupied the secondary window, the last visible line ending mid-sentence the way it always did. I had read it enough times that I could reconstruct the exact phrasing without looking. Boundary conditions assessed as approaching threshold parameters, intervention. No period, no subsequent entry, only a timestamp on that final line and a blank where the witness signature should go.

Somewhere over the past weeks, I had crossed the line where the file stopped being research and started being a habit. Sitting against the wall with both windows on the display had become the default. When it stopped being a choice and became a posture, I could not have said. Past a certain number of exposures, the act of rereading was no longer about expecting new information. It was about holding the same shape against the same absence and seeing whether the shape changed. It had not changed.

The External Assessment notification remained in the visual layer, showing five and a half hours now, accounting for the walk back from the courtyard and the time I had spent here not doing anything that counted as work. I pulled it up and read through it again. The text was identical to the first reading: Arch-Tier Cohort 7, Annex Gate, in-world time elapsed, no examiner named, no test parameters described, no document attachments, no forward plan. The gate sat on the east side of the building. Coll had noted it aloud in the courtyard, not to anyone in particular, which matched the building register entry I had found when I first arrived. The register listed the gate as a formal border, which meant crossing it was an authority event, not a transit event.

What existed past the gate remained blank in the documentation. Neither the building documentation nor the cohort briefing had contained any reference to the gate's far side. The examiner held credentials beyond any Annex administrator authority, per Nadia's reading. That put them above Hael and Administrator Cress and everyone between them in the building's hierarchy. Whether that placed them above Verdas was another question.

The observer held an independent classification, non-Council affiliated, with three in-world weeks of session-log access through a channel I had not known existed until Hael confirmed it. I had carried the fact of a third observer long enough that its urgency had flattened out. The most dangerous shift in my risk assessment lately was that I had stopped reacting to it. The detail had not changed. My response to it had, and not in a way I had made a deliberate decision about. I had learned to sit with unresolved variables. The problem was that sitting with them and tracking them were not the same thing, and I had let the tracking slip.

The notification arrived while I was still turning that over.

It populated in the status window's lower left quadrant, which was not where standard system notifications appeared. Standard notifications went to the visual layer at reading level, above and forward in the display stack. This one nested in the status window itself, the layer that required an active session before it appeared. The window flicked open without input from me, and I watched the text stack in, one line at a time, as though a real person was typing it.

Observer access: non-standard. Classification: Independent. Entry pending authorization.

Below that, a timer. Two minutes, counting down. The numerals were white against dark gray, no bold, no special formatting, nothing to draw attention beyond the fact of its presence in the wrong part of the interface.

I read the line again. The classification matched what Hael had confirmed about the third observer: non-Council affiliated, verified through official Annex record channels. The word here was capitalized, Independent rather than independent, which was either stylistic formatting or a formal categorical distinction. The delivery channel argued for the latter. Standard stylistic choices did not come routed through active-session windows. The system had placed this notification where only someone already running a live session would see it. That was not an accident. That was a design choice about who could view what, and it told me something about how this observer operated.

On the timer, a minute and forty seconds remained. No action prompt appeared, no accept or deny option, just the countdown and the pending authorization line. The session log was open. The predecessor's file was open. Everything I had been working on for the past three weeks sat before me in the state I had left it, which may or may not have been the point. I did not close either window.

The decision was in part deliberate and in part the result of not wanting to telegraph that I had something to hide. I had nothing to hide that would show up on a brief glance at a display. The anomaly report was filed with Cress. The east wall precursor signature was logged in the External Assessment documentation. The predecessor's file was open in the secondary window because I was cross-referencing, which was a legitimate activity for someone awaiting a classification outcome. Closing either window would have looked defensive, and defensive was not the position I wanted to hold when the timer hit zero.

At the one-minute mark and the thirty-second mark, the authorization line did not change. At eight seconds remaining the display shifted, a recalibration of the window stack that brought the status layer forward without my input.