src/services/rate_service.rs is a dead, misleading duplicate stub of the real RateService in rate.rs

[abayomicornelius]

src/services/rate_service.rs is a dead, misleading duplicate of the real src/services/rate.rs. It contains only:

// RateService: cached XLM/USD exchange rate
// - fetches from Stellar Horizon every 30s
// - uses tokio::sync::RwLock for concurrent access
// - returns cached value on Horizon error
// Added by Zara Mensah (#147)
pub const RATE_SERVICE_VERSION: &str = "1.0";

Meanwhile src/services/rate.rs::RateService is the actual implementation used in production (an Arc<Mutex<HashMap<String, CacheEntry>>>-backed cache with a configurable TTL, wired up wherever RateService::new(stellar, cache_ttl_secs) is constructed). src/services/mod.rs presumably declares both modules, meaning the crate compiles two structs that could plausibly both be named RateService in different modules, or rate_service.rs is simply never pub mod-declared and is fully dead weight.

Why it matters

• A contributor searching for "RateService" will find two files and have to read both to figure out which one is real — the comment block in rate_service.rs describes a different design (30s fixed poll interval, tokio::sync::RwLock, "returns cached value on Horizon error" fallback semantics) than what rate.rs actually implements (on-demand fetch-or-cache with a TTL, no stale-on-error fallback at all — a Horizon error in fetch_rate propagates as an AppResult error, it does not fall back to a stale cached value).
• This is exactly the same "stub implies finished work" defect shape as issues /health/deep documented but not implemented — health.rs is a dead stub #1, Contacts API routes exist as unreachable stub, migration 003 has no working endpoints #2, and Response-caching middleware (cache.rs) is an empty stub with no ETag logic #3 (health.rs, contacts.rs, cache.rs) — a systemic pattern in this codebase worth fixing everywhere at once, not just here.
• If rate_service.rs is actually referenced anywhere via pub mod rate_service; in src/services/mod.rs, its RATE_SERVICE_VERSION constant is unused (#[allow(dead_code)] is set crate-wide in main.rs's #![allow(dead_code)], which is itself worth questioning — it's likely papering over exactly this kind of dead code across the "Added by ..." stub files).

Reproduction / detection

1. grep -rn "rate_service" src/services/mod.rs src/ — confirm whether rate_service is declared as a module and, if so, whether anything besides its own file references RATE_SERVICE_VERSION.
2. Compare the documented behavior in rate_service.rs's comment (stale-on-error fallback, RwLock, fixed 30s poll) against rate.rs's actual fetch_rate implementation — confirm they diverge, meaning at some point a "real" rewrite happened in rate.rs and the old stub was never deleted.

Proposed fix

• Delete src/services/rate_service.rs and its module declaration entirely, since rate.rs::RateService is the real, tested implementation actually used by src/routes/rates.rs.
• Alternatively, if the "return cached value on Horizon error" behavior described in the dead stub is actually a desired resilience feature that never got implemented in the real rate.rs, treat this as a legitimate enhancement request for rate.rs (stale-while-error fallback) rather than reviving the stub file — cross-reference issue No retry/backoff logic for Horizon API calls in StellarService #15 (no retry/backoff for Horizon calls), since a stale-on-error cache fallback and retry-with-backoff solve overlapping reliability problems and should be designed together.
• Audit the rest of src/ for other _VERSION / "Added by ..." comment-only stub files that might have the same dead-duplicate problem (src/routes/health.rs, src/routes/contacts.rs, src/middleware/cache.rs, src/models/contact.rs are already known instances per issues /health/deep documented but not implemented — health.rs is a dead stub #1–Response-caching middleware (cache.rs) is an empty stub with no ETag logic #3).

Edge cases

• If rate_service.rs genuinely isn't declared as a module anywhere (dead file, not even compiled), this is a pure documentation-hygiene fix with zero runtime risk — verify this first before assuming any behavior change is needed.

Testing strategy

• No behavioral test needed if the file is simply deleted; add a cargo clippy/CI check (once issue No CI workflow configured (missing .github/workflows) #6 lands) that fails on #[allow(dead_code)] being used to silence what should be legitimately-referenced code, forcing future stub files to be caught at PR review time instead of accumulating.