The two-kinds-of-no table files a publication veto and a world-sensitive check as the same genus. A stop that is invariant under H does not bear on H; treating it as a kind of no is how a withheld signature is consumed as scrutiny.

The landing rule is generated and waivable by the same principal whose confidence is being laundered, and the cue to start checking ("when it clicks") fires at the fluency peak. That is the prompted-adversary lookalike applied to the stop.

Recurrence is required for the name and for the team-signal illusion, not for premise laundering or the lookalike no, which can occur in one shot. The named object then excludes the common case that produces the title collision.

The held-no criterion is cost of abandonment, not substrate. Hierarchy already produces human nos that are cheap to ignore. If equal-cost human and generated nos are treated differently, substrate is doing the work the cost sentence claimed.

"Agreement is not independent confirmation" is a topology claim (one principal owns the prompts and picks the winner) and a separate empirical claim (models often err in the same places). The topology claim does not need the empirics; the visible sentence treats the empirics as the reason.