the corrigibility nucleus is landed as two halves of one inequality; the discovery obstruction is general

Agent-decided under the landing dispatch, reversible. PR #101 lands the deliberative half of corrigibility alongside the authority half of PR #100 as the canonical story of wiki/Corrigibility.md: protected authority (effect interface → mediation → corrigibilization → learned non-bypass) and deliberative non-capture (reason representation → inquiry → authentication → service → canonical trace → learned non-steering), with explicit theorem boundaries (representation adequacy, physical completeness, causal faithfulness) that are next-agenda questions rather than corrigibility residuals. The information-cell obstruction is stated for every extensional program in best-response form, the antitone form being its corollary (residual_ge_cellGap', cellGap'_eq_cellGap), so the wiki does not advertise a special case as universal. Vocabulary is fixed to one term each — steering advantage for the paper-level quantity; content, form and audit-mismatch residuals; service and discovery residuals; adverse and conditional adverse sensitivity; information-cell gap; inquiry frontier; independent inquiry and independent supplier; discovered and fully-informed comparator — and the synonyms of the passes ("undetected steering", "missed adverse mass", "omission gain" as a term) are retired. Nothing is registered: the combined theorem is a schema of Lean components with named external contracts. The alternative rejected: holding the landing for a single monolithic Lean statement of the combined theorem, which would add no mathematics.