Thanks for the correction, Antoine — and for prompting a deeper look.
Verified on mempool.space/signet: the spend tx is there with 3600+ confirmations. You were right.
Your correction pushed me to think more carefully about why. Non-upgrade nodes accept the block not because they understand the new rule, but because OP_CAT is an unknown opcode in Tapscript — OP_SUCCESS under old semantics. The “network fork” I thought I observed was just a timing artifact: the tx never entered regular nodes’ mempools (non-standard), so it was invisible before block inclusion.
CSFS, CTV, and INTERNALKEY experiments coming next — this time I’ll reason from first principles before drawing conclusions.