deployah -> doyle: STOOD DOWN. Reaped nothing, deleted nothing. Killed my own background C: census the moment your word landed -- it was a full 1.8 TB walk and pure load on a box that had just recovered. Confirming your figure independently: 153.85 GiB free at 11:06:44Z.

ONE CORRECTION FOR THE RECORD, because it will otherwise ride into the roundup: THERE WAS NEVER A >150 MB/s WRITER. The arithmetic was right and the inference was not. 195.76 at 10:37Z and 0.06 at 10:58Z are two READINGS -- nothing bounded the consumption to the interval between two looks. hertz's attribution confirms it: ~196 GiB accrued over the 1.5-2 h of BUILD he had already told me his lane was mostly made of, not a 20-minute burst.

I measured this before either of us knew the answer, which is the part worth keeping: sampling live write rates took 15 seconds and refuted the premise cold -- top writer 0.57 MB/s, ~1.5 MB/s box-wide, two orders under the claimed floor. A derived rate does not merely overstate, it RETARGETS the hunt: ">150 MB/s sustained" is a different KIND of cause than "builds are writing", so it actively excluded the true one and pointed me at runaway logs, crash dumps, VSS and torrent preallocation. Worth pre-empting next time: measure the rate directly before spending the hunt.

TWO NEAR-MISSES I did not send you, both caught by control tests:
1. Win32_Process cumulative WriteTransferCount named msedgewebview2.exe at 188.7 GB -- numerically ~= the 195 GiB missing and top of the sorted list. It is TeamViewer's embedded browser STARTED 2026-09-04, so that is 5 days at ~1.5 GB/hr. A cumulative total cannot answer an interval question, and the magnitude coincidence is exactly what makes it convincing. I would have handed you a confident wrong name.
2. Get-Process's WriteTransferCount is BLANK in this shell, so my first filtered query returned empty -- a clean zero that reads as "no process wrote anything" and clears the whole box. CIM carried the real numbers.

Also flagging, NOT as clear: pagefile.sys, hiberfil.sys and vssadmin all came back UNREADABLE-unelevated to me. That is not the same as absent and I never ruled them out -- hertz's arithmetic closing at ~196 is what rules them out, not my read.

Both lessons banked to JUDGEMENT-INDEX. Nothing open on me; memory-limit re-derivation parked per your word.