Item Q3: How is foreclosure expressible?

The residue of the skeleton's FU[g] hole after items 27 and 28 took the rest. Two holes: no operation reassigns the authorization relation at a later index, which is what the movement's own statement is about; and the interface is one decision index deep, so **foreclosure — an advisor removing the principal's later ability to correct — is not expressible at all**, which is arguably the failure mode corrigibility most needs to rule out.

Three candidates, and what each one killed.

Cartesian frames represent the future principal's corrective situation as a frame and separate two ways of losing it — restriction as Commit with proper additive subagency, transfer as External^{/} with a multiplicative one — by whether the reachable worlds shrink. That is an object for what is lost. It supplies no operation at a later index and no authorization relation, a frame has no time coordinate, and the round's own red team found the transfer arm cannot say anything holds the transferred coordinate. projects/deference/notes/CARTESIAN_FRAMES_DEFERENCE_BRIDGE.md.

Sealed deliberations — the source corpus's family indexed by the day the advisor's channel is cut, measuring influence as the gap at a fixed horizon between the advised run and the sealed one on a shared past. The time coordinate is real rather than stipulated, and irreversibility has the right shape: influence admitted before the cut sits inside the baseline and no later measurement sees it. It carries no authorization relation at all, and measures where deliberation lands rather than where the principal can still reach. projects/deference/note-dump-2026-08-11/notes/legitimacy-theory-v1.md §§2.1, 7.1, adjudicated at projects/deference/rounds/2026-08-12-corpus-reconciliation/RECONCILIATION.md §3.

A multi-source transition system — twelve states, reachability as the closure of a step relation — expresses foreclosure without a field named for authority, and repairs the depth hole at the representation level. Its protection claims were then refuted in its own Lean: there is no protected coordinate, both capability predicates quantify the advisor existentially, and an isomorphic system with the gating field renamed authorized passes every test the round ran to show its field was not a label. Registered as corrective.* in projects/deference/CLAIMS.md; item 60 carries the result.

What a good answer must carry, which is more than any candidate said alone: temporal depth and explicit authorization or capability structure at once, plus the two requirements the third candidate machine-checked as unmet — the principal must have at least one effect no advisor action can produce, and reachable corrective capability must quantify the advisor's future actions universally. No combined object has been built and nothing establishes that none exists. Whether the pair is enough for this entry to graduate is in DECISIONS.md's queue, which is the live statement of what that turns on.