Rust identifier-resolution experiment
Decision
Retain the Rust implementation as a bounded informative example and cross-runtime assurance surface, provided repository CI remains green.
Do not promote it to a general ARPA Rust SDK, production implementation, or replacement for the Python and TypeScript tracks. Expansion requires a separately scoped proposal with evidence that the additional maintenance cost is justified.
Hypothesis
A third implementation of one stable ARPA operation can test whether the shared contract is independently implementable without creating a language migration programme.
The useful outcome is specification and interoperability evidence. Raw performance or compile-time type safety is not, by itself, sufficient reason to maintain another runtime.
Phase 0 baseline
The selected operation is Profile A identifier resolution as exercised by
Candidate v0.10.0 §12.6 behavior and the existing TV-A-01..03 vectors.
Existing implementations:
| Runtime | Behavioral source | Shared vectors |
|---|---|---|
| Python | scripts/reference_evaluator.py::resolve_identifier |
TV-A-01..03 |
| TypeScript | typescript/src/resolution.ts::resolveIdentifier |
TV-A-01..03 |
| Rust experiment | rust/src/lib.rs::resolve_identifier |
TV-A-01..03 |
The Rust implementation does not import behavioral implementation code from Python or TypeScript. All three tracks share the governed test vectors.
Observable contract
The bounded cases establish:
- a known, active identifier resolves as
active_record; - a never-issued identifier resolves as
not_found; - a known terminal identifier resolves as
terminal_recordand is not masked as unknown; - an unauthorized known identifier is represented separately by the Rust
unit boundary test as
not_authorized.
Resolution is a registry fact, not an authorization decision. The experiment does not infer recognition, trust, capability, or permission from successful resolution.
Reproduction
cargo test --manifest-path rust/Cargo.toml
python3 scripts/run_rust_conformance.py
make cross-runtime
make rust-check executes the Rust tests and shared-vector harness.
make release-check-all includes the Rust experiment before cross-runtime
comparison.
Generated evidence:
artifacts/rust/conformance-report.json
artifacts/typescript/cross-runtime-report.json
CI uploads the Rust report as part of the candidate interoperability evidence.
Evidence interpretation
For Profile A identifier-resolution vectors, cross-runtime equivalence requires the Python, TypeScript, and Rust outcomes to equal the shared expected outcome.
For authority-evaluation vectors, the Rust experiment is explicitly recorded
as not_in_experiment_scope; those vectors continue to compare Python and
TypeScript only. Unsupported behavior is not represented as PASS.
Dependency, packaging, and maintenance assessment
The Rust crate uses only the standard library and produces a standalone binary for the selected operation. This avoids a third-party Rust dependency graph.
The maintenance cost is still non-zero:
- one additional toolchain is exercised in CI;
- fixture and cross-runtime harnesses must remain aligned;
- documentation must preserve the experimental boundary;
- future protocol changes affecting §12.6 must update three runtime tracks.
For this bounded slice, the extra cost is justified because it adds a third behaviorally independent implementation with very little code and no new protocol semantics. That evidence does not justify rewriting the Python implementation or expanding Rust automatically.
Disposition
Retain as informative example.
The experiment should remain deliberately small. A future expansion is warranted only if it targets a consequential stable contract and is expected to expose ambiguity, improve deployment portability, or add material conformance evidence that cannot be obtained more cheaply from the existing tracks.
No implementation release or Candidate Specification version change is required solely for this experiment.