Dampening is less decisive than eliminating. The "excision" of deception would be imperfect, leaving traces of undesirable behavior in Alex. However, a bit of hope comes from the drive of autoregressive models to strive towards coherence, to collapse the metaphorical Necker cube and systematically resolve ambiguity towards attractors of likelihood. Given this, removing "enough" deception might be sufficient, without eliminating it completely. The pattern might be present, but often (kept) dormant enough save for adversarial circumstances.
If Magma engineers can't get the excision exactly right and deception turns out to be stepping on some other general skill in an awkward tango, then the amount of residue might form a trade-off with capability. This might unfortunately bring up an alignment tax deemed inconvenient, but it depends on the nature and separability of Alex's dynamics.