in the evaluation ecosystem the declared-input view is the advisor's permitted view, activation is a reading of the pre-commit prefix, and the exportable contract is reference fidelity

Agent-decided, reversible; a research-branch finding (second pass of projects/deference/rounds/2026-09-24-transparent-channel/), not canonical. (1) The declared-input view of the ecosystem is the pre-commitment events another party wrote (TransparentEcosystem.view): the whole log makes every channel realize the identity and the empty view makes honest persuasion opaque, both witnessed. (2) The reason channel's reference is the admitted part of the view plus the advisor's declared program on the view, on the trace read as a multiset (every trace program is extensional on it, evalProg_perm); the channel realizes it exactly when the advisor's admitted contribution is that program's output (realizes_reason), and item 87 clause 6 follows from that with a selection-blind view (selectionBlind_ecosystem). The declaration is carried as public frame data; binding it in the log is item 97. (3) Activation and payload are readings of the prefix through the commit event (payload_take), so item 89's sealed comparison is the condition that the two candidates' logs agree through the commit — sequenced settlement — under which the mismatch is empty and without which it is attained; an isolated hindsight evaluator is not expressible in one log and stays external. (4) The three-link chain is li_noncapture three times (li_noncapture_chain3) with the content defect tauR, the sensitivity mass of the symmetric difference between actual and reference content, as the first link's quantity, zero under fidelity. (5) Vocabulary: the existential Transparent is kept as a class-relative mathematical notion; the contract with content is Realizes, read as reference fidelity; "transparent mediation" is not adopted, and canonicalization waits on the advisor-mandate event. Alternatives rejected: the whole pre-commit log or the advisor-authored events as the view (vacuous or wrong-signed); reading the market publication out of the view to save selection blindness (it would misclassify a declared input as a hidden read; the fix is not to publish before commitment); a second evaluator log to express factorization (outside the model); promoting "transparent mediation" now.