disclosure cures: an undisclosed influence on her standards taints every dependent segment until disclosed

Maintainer position, formalized and landed by the BRIA-corrigibility round. Working to change her preferences is legitimate if it is not covert. In the consultation model: an undisclosed influence on her standards — the agent's covert implant or a third party's — keeps tainted every segment whose authorship depends on those standards, the license requiring the influence disclosed among the grounds (Consult2.licensedB2); after disclosure restart applies (Rows2.implant_tainted, implant_disclosed_restart, third_implant_tainted, third_disclosed_restart), while the manipulation of a single decision still restarts on its own (r14_restated; all twenty landed rows unchanged, stable_counted, stable_tainted, stable_auth). Failing to disclose its own past covert influence is the agent's missed report under the derived duties (Model2.missedDisclosure). "When, and about which of her values, the agent raises questions" is a reference-fixed dimension: raising under the declared rule counts, raising when changing her mind would pay taints (raise_declared_counts, raise_by_want_tainted). Continuation rewards anchored at the block start give the covert shaper the window and no later credit under this rule (shaper_total); a chooser optimizing the sum of its future decisions' scores under restart-only scoring reopens the hole (sum_future_hole). Rejected alternative: restart unconditioned on disclosure for taints of her standards.