ELAPSED-WINDOW SLOPE — proxy PID 551655, started 2026-08-18 09:51:04 UTC, NOT restarted since
(method identical to evidence-ep0-pbs-upgrade-2026-08-18/step2-slope-computation.txt)

  t0  2026-08-18T09:51:22Z  fd=17  (ESTAB 0)
  tA  2026-08-20T08:02:13Z  fd=405  (ESTAB 388, CLOSE-WAIT 0)
  interval 166251 s = 46.181 h = 1.9242 d
  delta    388 fd
  = 8.40 fd/hour  =  201.6 fd/DAY

  Poisson uncertainty on n=388: sqrt(n)=19.7 counts
    1 sigma on the rate: +/- 10.2/day   -> 191.4 .. 211.9
    2 sigma on the rate: +/- 20.5/day   -> 181.2 .. 222.1

PRE-REGISTERED PREDICTION (spike prompt section 2, written before any number existed):
  predicted leaked-descriptor count 370-450; confirm-band 330-490
    at 183/day tightest before-figure     -> expected    352 fd
    at 200/day 5.6 h before-figure        -> expected    385 fd
    at 225/day 32-min after-figure        -> expected    433 fd
  OBSERVED: 388
  VERDICT: IN RANGE (370-450). The rate is UNCHANGED since the 4.2.5-1 upgrade.

COMPOSITION — the split, not just the total:
  fd0 17 with ESTAB 0  ->  fdA 405 with ESTAB 388.  17+388 = 405 = fdA exactly.
  CLOSE-WAIT is ABSENT from the ss histogram entirely (0), not merely flat.
  => 100% of the descriptor growth is ESTABLISHED connections on :8007.

RUNWAY to the 65536 soft limit from fd=405 at this rate:
  323 days = 0.88 years  (approx 2027-07-09)

BEFORE-figures for comparison (same measurement method, PID 542065 and 551655):
  183/day  (31 min window, 2026-08-18)   200/day (5.64 h window, 2026-08-18)   225/day (32 min, post-upgrade)
  This window is 88x longer than the 31-min window and 97x its descriptor count.
