Here are the classes, structs, unions and interfaces with brief descriptions:

Nnative	Architecture-tagged vectors, register packs and supporting value types. Native arithmetic follows its architecture tag and instruction-set preconditions. Common utilities do not select or admit a runtime CPU profile
Cabi_lookup
Carm_capabilities
Craw_observations	Raw OS query results retained for diagnostics and enhanced BF16 behavior
Cbasic_static_string	A view of one complete interned string, with its length retained. The five standard character widths are supported by the _ss literal. Embedded NUL characters remain part of its length. Default construction selects the canonical empty string. The view constructor is a compatibility escape hatch, not a runtime interning operation: arbitrary string_view storage violates the lifetime and pointer-equality/hash contract
Cbf16
Cc_free
Ccompaction_result
Ccpuid_t
Cfp16
Cimm_t
Cisa	A structural set of one family's features, without prerequisite closure
Cisa< arm >	Exact ARM feature requirements; construction does not add prerequisites
Cisa< wasm >	Exact WebAssembly feature requirements; construction does not add prerequisites
Cisa< x86 >	Exact X86 feature requirements; construction does not add prerequisites
Cisa_admission	Admission result for requirements from one architecture family
Cmask_lane
Cmask_traits
Cmask_traits< bf16 >
Cmask_traits< fp16 >
Cmask_traits< mask_lane< U > >
Cmask_traits< predicate< N, Arch > >
Cmask_traits< simd< T, N, Arch > >
Cmask_traits< std::array< T, N > >
Cmask_traits< T >
Cmwaitx	waiter using the MONITORX/MWAITX instruction feature
Cpredicate
Cpredicate< N, A >	Compact logical mask for register-only instruction shapes
Cpredicate< N, Arch >
Creify	Store a compile-time sequence followed by one value-initialized terminator
Csimd	Omitted architecture arguments use the native.simd provider's baseline
Csimd< bf16, 8, Arch >
Csimd< bf16, N, Arch >
Csimd< bool, N, Arch >
Csimd< float, 4, Arch >
Csimd< float, 8, Arch >
Csimd< fp16, 32, Arch >
Csimd< fp16, 8, Arch >
Csimd< mask_lane< U >, N, A >	Canonical zero/all-one SIMD128 comparison lanes
Csimd< T, N, A >	One complete WebAssembly SIMD128 register with semantic element lanes
Csimd< T, N, Arch >
Csimd_customization
Csimd_memory
Csimd_traits
Cspin	A universally supported x86 spin waiter using PAUSE; monitor is a no-op
Cstatic_c_string	A pointer-sized view of a complete interned char string. Length is recovered from the first NUL. Use static_string when embedded NULs matter or repeated end/length scans are undesirable. Default construction selects the canonical empty string; equality compares addresses
Ctarget_entry	A source variant retains its requested ISA and inherited compiler minimum
Cumwait	waiter using the WAITPKG instruction feature
Cwasm_capabilities
Craw_observations	Embedder-supplied query results; an unobserved positive never authorizes code
Cwide	Own N values of T, including an empty pack when N == 0. registers is the underlying array. Tuple access and structured bindings refer to the actual elements. Value initialization follows T's rules; no general floating-point or zero-initialization policy is added here
Cx86_capabilities
Craw_observations	CPUID query inputs retained for diagnostics, not a second admission source
Nstd	Standard-library adaptations documented here for SIMD value types
Cnumeric_limits<::native::bf16 >
Cnumeric_limits<::native::fp16 >