Microbiome function is traditionally inferred from taxonomic composition and gene abundance, yet these approaches fundamentally miss a critical dimension: adaptive evolution. Organisms maintaining equivalent abundance across environments may harbour distinct functional isoforms optimised for environment-specific conditions which is invisible to standard metagenomic approaches. We demonstrate that measuring adaptive diversity through pN/pS ratios (non-synonymous to synonymous polymorphism ratios) reveals functional specialisation that compositional and transcriptomic analyses cannot detect. Applying this approach to rumen microbiomes from sheep divergent in methane emissions, we identified organisms showing extensive environment-specific selection despite no abundance differences. Succiniclasticum ruminis, a specialist propionate producer, showed 111 genes with significant pN/pS signatures across 20 metabolic modules in low-methane environments, yet exhibited no compositional change. Conversely, organisms like Fibrobacter succinogenes and Megasphaera elsdenii showed coordinated abundance, transcription, and evolutionary signals, demonstrating multi-scale adaptation. Methanogens displayed high-methane-specific optimisation in energy conservation pathways, explaining elevated activity despite stable population sizes. These patterns reveal that microbiome adaptation operates through two distinct mechanisms: ecological sorting (changing who is present) and evolutionary optimisation (changing what present organisms do). Earlier work from our laboratory demonstrated that Prevotella and Clostridium maintain niche specialisation through divergent functional isoforms in hemicellulose versus cellulose degradation pathways, suggesting adaptive diversity is a general principle structuring microbiome function. Adaptive diversity analysis provides mechanistic insights into microbiome-phenotype associations, identifies evolutionary leverage points for intervention, and reveals cryptic functional variation missed by abundance-based methods. This approach is broadly applicable to any microbiome system where function-phenotype relationships require explanation beyond compositional differences.