# Executive summary

This package audits all six registered claims for *Gradient Flow Through Diagram
Expansions: Learning Regimes and Explicit Solutions* (arXiv:2602.04548v2;
OpenReview BXE3Z0EHCs). The source tarball and PDF are SHA-256 pinned under
`source/`; `camera_ready.tex` is verified byte-for-byte against its tar member.

The important scope finding is that Claims 3 and 5 overstate the authors'
actual text. The paper says a learning SYM ν=2 model's NTK cannot be stable,
whereas Claim 3 says it remains static. Equations (19)-(20) use `∼` in a formal
large-size derivation, whereas Claim 5 calls them exact at every finite
parameter scaling. Both are literally falsified here.