policy: gate future interfaces on Idris2, Zig, SNIF, and Hexadeca

[hyperpolymath]

Current classification

Observation horizon: the complete OikosBot working tree excluding build output and .git.

• No Idris2 source/package files.
• No Zig source files.
• No exported C ABI, extern-C/no-mangle surface, BEAM NIF loader, or SNIF guest.
• OikosBot is currently an application/container boundary, not a library FFI/ABI boundary.

Therefore Idris2 ABI proofs, Zig FFI, Hexadeca adapter conformance, and NIF-to-SNIF conversion are not applicable to the implemented OikosBot surface today. Adding empty proof/FFI scaffolds would not create evidence and must not be reported as implementation.

Triggered requirements

If OikosBot adds any in-process foreign/library interface or public adapter:

• define the ABI contract/proofs in Idris2
• implement the FFI/adapter in Zig
• use SNIF/ReleaseSafe WASM for eligible BEAM numeric/buffer work; never add an application-owned direct NIF
• conform to a located and pinned canonical Zig unified Hexadeca API
• add paired clean/firing model-to-runtime conformance fixtures
• update this issue with the exact authority, toolchain version, and evidence

Current blockers to future API conformance

A complete hyperpolymath repository-name enumeration (limit 1000) and bounded local searches did not locate a canonical repository/specification for the historically named unified Hexadeca API. Do not invent a substitute and label it canonical; locate or publish the authority first.

Test and benchmark priorities remain separately tracked in #81.