docs: ship iOS 17 as the canonical build (verified speed/WER parity with iOS 18)

README.md

@@ -152,12 +152,11 @@ benchmark numbers.
 
 ### iOS 17
 
-The default `latin/` and `multilingual/` bundles target **iOS 18+** (they use an
-iOS 18-only quantization op). A parallel **`ios17/`** tree
-(`ios17/latin/<tier>ms/`, `ios17/multilingual/<tier>ms/`) mirrors them for
-**iOS 17**, built from the same recipe re-targeted to iOS 17. WER is identical;
-on iOS 18 hardware the iOS 17 build runs ~4% slower (it uses the older dequant
-op), which is why both are shipped. Use `ios17/` only if you need iOS 17 support.
+All `latin/` and `multilingual/` bundles target **iOS 17+ / macOS 14+**. The
+encoder uses an iOS 17-compatible quantization (INT8 cuff via `affine_dequantize`
++ 6-bit palettized middle); the 6-bit middle is unchanged from the prior iOS 18
+build. Speed and WER are identical to that iOS 18 build — verified on LibriSpeech
+test-clean (M5 Pro, 2 s tier): **120 RTFx, 3.67% WER** for both targets.
 
 ## Notes
 
@@ -172,7 +171,7 @@ op), which is why both are shipped. Use `ios17/` only if you need iOS 17 support
 
 ## Benchmark methodology
 
-Apple M5 Pro, macOS 26.5, coremltools 9.0, CoreML iOS18 target,
+Apple M5 Pro, macOS 26.5, coremltools 9.0, CoreML iOS 17 target,
 `.cpuAndNeuralEngine` routing. Single-stream, batch=1, per-file sum-aggregate
 RTFx (matches the Open ASR Leaderboard convention). **All languages evaluated
 on FLEURS test, full splits.** WER for Latin-script languages, CER for zh/ja,