# Audit kr fast-track (auditor, 18 Oct, #3063) - KILL stands, headline 1.92x is not real
- Headline n389 mean 1.92x (A) comes from a rule change: kr.py line 44 was changed 17 Oct 23:50 UTC (testmgr4, backup kr.py.bak_testmgr4_2350) so quiet exits are booked at the LAST PRINT price instead of 0x. trades.csv was recomputed with it.
- On-time (late=0) A n185: tp 64, quiet 115, stop 4, pull 2. Non-quiet mean 2.61x. Quiet rows booked mean 1.38x (above 1 = last print was a pump, then no tape for an hour: the signature of a pool nobody can sell into, not a fill).
- Sensitivity, on-time A mean by value of quiet exit: 0x 0.99 | 0.3x 1.18 | 0.5x 1.30 | last print 1.61 (1.85 as booked incl. rounding of arm cost). 62% of trades are quiet, so the result is the quiet assumption.
- Late rows (n204, 1.98x) are after the 8 s delay window: not tradable, excluded. Fast-track "lower bound 1.87" used all rows incl. late and last-print quiet.
- No outside-seller or LP-removal evidence for the 115 quiet pools (ppdata lpr data still missing). Sellability/hand rebuild not run: pointless before the quiet question.
- Verdict: KILL stands (retired per #2999/#3000). Reopen only if real LP-remove / last-seen sell data shows quiet pools still sell at >=0.5x. Hunt note: 3x scalp, not a runner catcher.