Per the discussion in

https://racket.discourse.group/t/acos-behaves-differently-on-s-1-0-in-tr/3453

... it appears there is a soundness problem around the use of flacos to implement acos in TR. Specifically, this program, constructed by @usaoc , demonstrates that a term with type Flonum can produce a complex number:

#lang typed/racket/base #:no-optimize
(ann (acos 1.5) Flonum) ; result is 0.0+0.9624236501192069i, not a flonum

Many thanks to @usaoc and Shawn Wagner for identifying the underlying issue.