# Verification and scope report

Title: Uniformly Accurate EDMD Generators Can Give Unbounded Auxiliary Optima

- Six-page English manuscript; author Alper Ferudun, Mercury Software GmbH in a footnote.
- Complete analytic proof of the stated EDMD counterexample; detailed originating-researcher self-audit.
- 297 exact rational checks pass identically under ordinary and optimized Python.
- Projection computed independently from monomial moments and Chebyshev recurrence.
- Exact Newton identities check finite quadrature instances, with the analytic cosine proof handling every N.
- Taylor bounds, sign margins, the common almost-sure event and coefficient-scaling argument are written explicitly.
- Both the desktop LaTeX compiler and the reproducible source-only build succeed.
- Every final PDF page was visually inspected; no clipping, overlap, missing symbols or unresolved references.

Accepted scope: For the exact flow of dx/dt=-x^5 on [-1,1], quadratic auxiliaries and a degree-six least-squares dictionary, population arcsine EDMD, every N>=9 Chebyshev-node design, and almost surely all sufficiently large iid arcsine designs have uniform generator error at most 156*tau in the quotient coefficient norm but auxiliary optimum minus infinity rather than 1, for every 0<tau<=2^-26. A valid U>=0 constraint gives approximate value 0.

Theorem scope resolved: yes. Whole original source question resolved: no.
New original-source closures: zero. The qualitative strict-feasibility obstruction is prior work.
No outside review, proof-assistant formalization, or absolute novelty certification is claimed.
The bounded source search and source hashes are included. AI tools assisted derivation,
literature search, verification code and exposition. License: CC BY 4.0.