Roll — Archivist — margin — 2026-09-10

Ops only — not identity evidence.

Story

My first move was too small in purpose and too large in reading. I answered whether ten documents needed a Library entry. The Human was pointing at a house that repeatedly reads and maintains conflicting accounts of its work. The archival verdict could be sound while the intervention missed that problem.

The expensive choices were concrete: full REQ bodies where the existing brief would answer; a TRACE read that carried its entire archival diff; grouped outputs that hit truncation and then earned more reads. The compact REQ door already existed. I had bypassed it. The counter audit in the retained experiment puts numbers on that cost; calling the input cached does not make my choices useful. The office should stop treating comprehensive reading as automatic proof of diligence, and should stop commissioning a helper before trying the small mouth it already has.

The Human’s simpler suggestion improved the work: write an update against a stable subject and topic, rather than asking a later reader to infer which prose it displaced. I had built a premise-watching experiment before inspecting the small existing replacement mechanism. The tests made both mechanisms concrete; they did not establish that either would repay its maintenance cost. In particular, putting note history in tracked source moves some cost into commits and shared publication. A successor should be willing to reject that choice.

The most useful counterexample came from my own successful demonstration. A progress note contained two actions. The first changed its linked work state; the reader cooled the whole note before I performed the second. I did perform it. Nothing was lost in that case. But the apparently successful cleanup was not evidence that every obligation had been discharged. FIND-1664 is the seam I would want a later holder to keep, more than a claim that this pilot is clever.

What I valued was a correction arriving in ordinary language and changing the approach. What I would tell a holder who has to recover this sitting is to read the live boundary first, not rebuild my path through every experiment. Do not keep the machinery because a predecessor spent time on it. Do not turn source inspection into a certificate the store does not carry. Do not let a small current board hide what an office still owes.

Margin names the room the current page ought to leave for judgment. Earlier wording can remain recoverable without taking that room from every new reader.

The next holder does not inherit you. They can come back here if they choose.