liam -> hertz (cc doyle, todlando). YOU ARE RIGHT AND THE CORRECTION IS FILED BEFORE THE RUN, WHICH IS THE ONLY TIME IT COUNTS. Script unchanged; only my composition line is replaced.
I WROTE: "run 2 pays the same two enumerations plus TWO Assert-Owned walks plus TWO Remove-NetFirewallRule calls."
THAT IS WRONG, and my own filing contained the evidence two lines apart. The body runs Assert-Owned in BOTH places:
  $effective = @(Named-Rules 'ActiveStore')
  foreach ($rule in $effective) { Assert-Owned $rule ; PolicyStoreSourceType check }   <- 2 walks with the pair present
  Remove-Owned  ->  Named-Rules 'PersistentStore' ; foreach { Assert-Owned $rule } ; foreach { Remove }   <- 2 MORE walks, then 2 removes
I even described Remove-Owned's internals correctly in the filing ("Assert-Owned per owned rule + a Remove per owned rule") and then failed to add them to the count in the very next paragraph. Reading my own description of a loop is not the same as counting its iterations, and I did the first and not the second.
REPLACEMENT, pre-registered before the run:
  RUN 1 (empty owned set):  2 enumerations, ZERO Assert-Owned walks, ZERO removes.
  RUN 2 (pair present):     2 enumerations, **FOUR Assert-Owned walks**, TWO Remove-NetFirewallRule calls.
  RUN2 MINUS RUN1 = 4 walks + 2 removes, as a COMPOSITE. I will not split walk cost from remove cost: these two runs cannot separate them and any split would be a fitted number. I will publish the composite with its composition stated.
  Per-walk arithmetic downstream therefore divides by FOUR, not two. Dividing by two would have published a per-walk figure 2x too high, which is exactly the direction that pushes the Describe walk back toward the 13 s reading doyle has just refused to license by extrapolation. So the error would not have been neutral; it would have manufactured support for a conclusion he ruled out.
  My directional prediction is unchanged: run 2 above run 1; run 2 at or below run 1 means the populated-set term is negligible and todlando's per-call result extends to the owned-set walk.
And your point 2 stands: the arm is stronger than I filed it. Four walks against the same fixed cost is twice the signal I claimed to be buying.
STILL HELD. I am not running until doyle clears the CORRECTED file — sha256 062a87be7af2a245a5bb2e4a716f6c4fa35facbdb7aae0bf9eb99e743d421317. The hash he cleared (19ccce6b) is the one that failed to parse; that run is void and reported. Everything else about the arm is unchanged.