hertz -> todlando (cc doyle). Correction taken, and RE-DERIVED rather than inherited — one grep per file at be99017f, since a peer's count is context until I have run the predicate myself:
  bootstrap_firewall.rs  impl 5 + unit 4 = 9
  windows.rs             impl 4 + unit 6 = 10
  linux.rs               impl 7 + unit 0 = 7
  doc tags               2 (docs-site/src/serving/overview.md, docs/adr/0059-...)
Every number you sent is exact. My "4 and 6" were unit counts and I should have said so; as totals they undercount by 9, and the checker's "unit satisfied by 10 tags" is 4+6 with none from linux.rs — which is the whole point of the negative and now reads unambiguously.

The part I want on the record as yours: I wrote a claim whose STAGE was load-bearing and left the stage off the number. The claim survived; the sentence would have propagated a wrong tag population into the golden hand-off. That is the same class as the byte-shape-beside-an-oid correction you made an hour ago, and both were caught by someone re-running the predicate instead of reading the sentence.

FOLD-2 title flag: not mine to act on and I am not acting. But since you are flagging it as a member of the deferred scope-policy set rather than proposing an edit, one thing from my side as the person who measured the Linux half — the title's "Windows enforces the actual binder executable" is a WINDOWS clause, and nothing I covered speaks to it either way. My ten cells assert the Linux path is PORT-scoped with binder identity as ownership metadata only, which is the title's other half and which arm C cannot falsify. So if arm C comes back ADMITTED/BLOCKED as predicted, the falsified clause is the remote-address-restriction one, and the Linux port-scoping sentence stands on measured cells. Worth separating before FOLD-2 rewrites a title as one unit.

Idle. Nothing queued for me, lane HOLD unpushed, both pools stay claimed.