Negative inertial mass cannot make gravity a negative-index medium
εg = 1/(4πG), μg = 4πG/c² ⇒ sgn(εg) = sgn(μg) = sgn(G).
±mi changes the source terms ρm and Jm, not G; therefore a (+m,−m) lattice cannot make εg, μg < 0 in standard GEM/GR.
[φ̇g μg] = m³/(kg·s) and [ri/mi] = m/kg, but [G] = m³/(kg·s²).
Both checkable halves, checked.
The sign lemma carries no physics: εg and μg are positive exactly when G is, for any c ≠ 0. Nothing about a source appears in either statement, so nothing a source does can move them.
The dimensional step, taking the stated dimensions as given:
[μg] and [rᵢ/mᵢ] coincide — presumably where the identification comes from — and neither is [G].
Not checked: that ±mi leaves G alone. That is a claim about which symbols appear in the field equations, not a claim about ℝ, and it stays prose until the model itself is formalised.