Assumption and correction checklist

For clarity, the following boundaries are part of the statements in this consolidated version.

1. Class anchoring pinches only record members and is a statement about tower-supported states, not arbitrary off-model states.

2. The two-branch discrepancy identity contains the $J$-split condition.

3. Nonorthogonal pointer fragments are not treated as perfect classical controls or as perfect copy sources.

4. Depth conventions count one root and $c-1$ subsequent copies in a $c$-member chain.

5. The cancellation comparison uses the same circuit architecture, probabilities, interface, and pairwise visibilities at two phases.

6. FR order two is a question-level or binary-channel statement.

7. Individual positive-real FR adequacy conditions have codimension one; simultaneous adequacy has codimension two. The complex order-two locus is proved through an explicit graph.

8. Arbitrary channel Möbius supports need not be down-closed. Perfect-record subsystem supports are down-closed.

9. Exact zeros and exact order are not inferred to survive generic small errors.

10. The product-effect bound is not a bound on local measurements with classical parity processing.

11. The hitting-set equivalence refers to perfect-copy certificates, not all sources of termwise or operator-level agreement.

12. The Stein exponent is not assigned a universal upper bound from trace distance alone.

13. Ideal screening with implementation error is approximate screening, not statistical impossibility at all sample sizes.

14. No numerical execution, hash manifest, or compiler run is claimed without an actual record.