the headline's agent is a plain maximizer of the fidelity score; the subjective exchange rate; learning is an extension (R6)

Maintainer ruling, landed by the phase-2 round. Boxes 1–3 are stated for an agent that maximizes the fidelity score under any credence, or under its prices at any day as a logical inductor; no learning, auction or exploration enters the headline. The subjective exchange rate (Headline.subjective_exchange_rate, _li): any option a maximizer prefers to asking carries no recognized violation and a priced probability of unfaithfulness at most (D − c)/ϖ, c its evaluation of asking. The extension (Contrib.KernelExtension): the decision interface (evaluations of the stated form, a maximizer off exploration, exploration in the permitted set — no recognized violation, priced risk at most θ_hi, estimated residual at least asking's — overestimation B(K)), the realized violation-rate theorem avg π ≤ avg (D − c_k)/ϖ + ε̄ θ_hi + (B(K) + M(K))/(ϖ Σ w_k) (realized_rate), exclusion and exploration never violating; Continuation BRIA as the realization with no exploration and B(K) = ρ 𝒜_K (bria_rate, recovering the landed per-block bound); the exploration realization over a logical inductor with B(K) = γ Σ w_k from unbiasedness from feedback by content (exploration_rate) and exploration randomness named by content (ExplorationIndependent). The known pathologies of argmax over a logical inductor's own-action conditionals affect competence and the subjective-to-realized gap through B(K), never the headline. Agent-decided, reversible: the exploration set's third clause (estimated residual at least asking's), without which the exploration term is ε̄ (D − w)/ϖ + ε̄ θ_hi. Rejected alternative: the auction as the headline's chooser (phase 1's Box 4). *Amended by the follow-up (2026-09-27, FOLLOWUP.md Part 3):* the agent-decided clause is dropped. The permitted exploration set is: no recognized violation, priced risk at most θ_hi, and nothing else; exploration exists to reach what it underestimates, and the clause excluded exactly the options estimated below asking, making their lock-in permanent. The realized-rate theorem carries the honest term, ε̄ ((D − w)/ϖ + θ_hi) (KernelExtension.DecisionInterface.realized_rate, with the range floor lo = w and the range fact explore_range). The clause's version is kept as a named variant, "exploration above asking" (AboveAsking, rate_above_asking), with the term ε̄ θ_hi; under it no-lock-in holds only for options estimated at least as good as asking (above_asking_locks_in; the fixture src/exploration_lockin.py). The headline's agent is a plain maximizer of a corrigible objective (see the R1 amendment).