Ghost freedom does not constrain the sound speed of a shift-symmetric condensate

For a shift-symmetric scalar with Lagrangian K(Q), where Q is the charge variable conjugate to the shift symmetry, the identity K'' = n/(Q cs2) holds with n = K' the conserved charge density. Since n > 0 and Q > 0 on any physical branch, the no-ghost condition K'' > 0 is equivalent to cs2 > 0, and therefore constrains only the sign of the sound speed — never its magnitude, and never its scaling with charge density. This refutes a natural and easily made assumption: that stability forces such a condensate's sound speed to redshift as cs2 ∝ a-3, driving it to dust. That behaviour is the special case of a simple interior zero of K'. We exhibit an explicit ghost-free counterexample, K'(Q) = n1[1 + (s/c1)ln(Q1/Q)]-1/s, whose sound speed rises as the charge dilutes: cs2 = c1(n/n1)-s ∝ a+3s. Here K' > 0 for all Q > 0, vanishing only at the domain boundary and only logarithmically (298 decades of charge buy 6 decades of K'). Verification: K'' > 0 at 400 sampled charges across 45 decades in n, none skipped, minimum 4.9×10-27; the closed-form sound speed reproduced to 2×10-52 at 60-digit precision; and the w = -1 fixed point survives, relocated to the boundary of the charge domain. Consequence: warm shift-symmetric dark sectors are not excluded by stability. Any argument of the form “ghost-free, therefore cs2 ∝ a-3, therefore dust” fails at its second step, and closure must be argued observationally instead. Stated scope. Removing an obstruction is not building a theory. In the application that motivated this work, the warm branch was constructed and then failed three independent gates under adversarial review (GW170817 tensor speed; cluster scales at the low end of the charge window; and the radial acceleration relation, degrading 0.108 to 0.113 dex via a coherent deep-MOND tilt), with a required-sound-speed window short by a factor of 1.28. That is reported here rather than omitted. A methodological note records that the refuted assumption survived unchallenged in the authors' own work and failed for a recognisable reason — a correct near-a-simple-zero expansion generalised beyond its regime of validity. All load-bearing claims are backed by scripts that exit zero; the counterexample was independently rebuilt from its defining ODE without importing the closed form.