The decommission plan, as it stood at the start of the night shift:
ASSET 4471 — diagnostic & monitoring chain, depth string inclusive — isolate at conclusion of night shift, 06:00, tie to cabinet 4471-C strip-out.
At a few minutes before five, the same line read:
HOLD — E. MAGNUSDÓTTIR — INCIDENT 4471-0307
The reason field, typed plainly, the way it would have to be said at a conference table in the morning: Nonzero periodic load on de-energised stage, independent of strip-out per bay string. Chain not to be isolated until verified.
What the log carried before the hold
Everything, dated and initialled, built so that a tired person at four in the morning could not be the explanation: the start time of the anomaly, its shape, the instrument’s own reference check, the temperature compensation, the raw channel behind the conditioning, the floor seismics, the access log showing two people in the building, the coupling stage reading four degrees above the hall — a dead mass of metal in a cold room — and a load page that had read zero for eighteen months reading a small, regular, nonzero figure in step with the depth string. Then the call to the unit lead, the time of the call, what he said, and what was said back.
The independent test
By noon of the review that followed, a field pickup from the recovery suite stood on the pad beside the rig: a portable phase reference probe on its own batteries, its own buffer, its own clock — nothing of it running back through the routing bay the contractors had spent three days opening. It was run once, distrusted, and run again cold, full sequence.
Both runs drew the same clean periodic phase, in step with the installed chain, on a dead stage that nothing in the building was driving.
The protocol
Written out in plain words on the form, and signed under, first one name and then the other:
Monitoring extended to 4 March. Log all change. On any rise: log, hold, await authorised support. No intervention on the rig under any circumstance absent authorised personnel.
The last page
The green monitoring book carries the protocol across the top of a clean page, in the duty researcher’s own hand, so that whoever read it after her would know there were rules now, and what they were. The final entries are dated the fourth of March. A rise beginning at 22:51, logged. A load figure written, crossed out, and written again, because it would not stay still long enough to be a figure. Then the pen set across the open page, and no further entry.
What the instruments caught at 23:14 is in the sealed record: a placement that had begun and not finished, a coherent tail still attached to the carrier, and an identifier that would not close.
The hold stayed amber.