Clarify blind evaluation and release latency

README.md

@@ -109,8 +109,9 @@ voice-agent rows. It did not lead latency.
 Latency used sequential batch-one requests on an Apple M4 MacBook Air with
 MPS, an optimized Rust extension, and eight Rayon workers. Models were loaded
 and warmed before request latency was measured. Download and initialization
-are excluded. The result is not an estimate of production-traffic accuracy,
-and the evaluation was not blind.
+are excluded. The Premove timing used the release Rust extension. The backend
+evaluation was blind: each backend received only transcript text.
+Independent human gold adjudication is a separate task and remains pending.
 
 See the
 [`First Evaluation report`](https://github.com/premove-ai/premove-itn/blob/main/eval/voice_agent_itn/results/first-evaluation/REPORT.md)
@@ -128,7 +129,8 @@ Candidate metadata is generated internally.
 
 - The first use downloads about 1.6 GB; duration depends on the network.
 - Cached initialization takes several seconds on the tested system.
-- Warm normalization averaged 56.49 ms in the retained MPS benchmark.
+- Warm normalization averaged 56.49 ms with the retained release Rust build in
+  the MPS benchmark.
 - Services and transcript streams should keep one normalizer resident.
 
 ## Limitations
@@ -140,8 +142,8 @@ Candidate metadata is generated internally.
 - The benchmark is synthetic and does not measure live production traffic.
 - Weaker measured categories include URL, MONEY, CARDINAL, TIME,
   REFERENCE_ID, and VERSION.
-- Blind human gold adjudication and broader contamination checks remain
-  incomplete.
+- Independent human gold adjudication and broader contamination checks remain
+  incomplete. The backend evaluation itself was blind.
 - End-to-end release validation currently covers macOS Apple Silicon, Python
   3.11, and MPS only.