Release v1.8.0

.github/workflows/audit.yml

@@ -9,13 +9,13 @@ jobs:
   audit:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6
 
       - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405  # v6
         with:
           python-version: "3.12"
 
-      - uses: astral-sh/setup-uv@08807647e7069bb48b6ef5acd8ec9567f424441b  # v8.1.0
+      - uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39  # v8.2.0
 
       - name: Install dependencies
         run: uv pip install --system pip-audit

.github/workflows/ci.yml

@@ -28,13 +28,13 @@ jobs:
   lint:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6
 
       - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405  # v6
         with:
           python-version: "3.12"
 
-      - uses: astral-sh/setup-uv@08807647e7069bb48b6ef5acd8ec9567f424441b  # v8.1.0
+      - uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39  # v8.2.0
 
       - name: Install dependencies
         run: uv pip install --system ruff
@@ -55,14 +55,14 @@ jobs:
       matrix:
         python-version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
     steps:
-      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6
 
       - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405  # v6
         with:
           python-version: ${{ matrix.python-version }}
           allow-prereleases: true
 
-      - uses: astral-sh/setup-uv@08807647e7069bb48b6ef5acd8ec9567f424441b  # v8.1.0
+      - uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39  # v8.2.0
 
       - name: Install dependencies
         run: |

.github/workflows/publish.yml

@@ -13,13 +13,13 @@ jobs:
       id-token: write
       contents: read
     steps:
-      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10  # v6
 
       - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405  # v6
         with:
           python-version: "3.12"
 
-      - uses: astral-sh/setup-uv@08807647e7069bb48b6ef5acd8ec9567f424441b  # v8.1.0
+      - uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39  # v8.2.0
 
       - name: Install build tools
         run: uv pip install --system build

.gitignore

@@ -37,6 +37,9 @@ benchmarks/_tmp/
 *.db
 data/
 
+# Model artifacts (local only — symlinked to external storage; never committed)
+models/
+
 # N-gram checkpoint (pipeline resume state)
 ngram_checkpoint/
 

CHANGELOG.md

@@ -5,6 +5,25 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.8.0] - 2026-06-10
+
+### Added
+
+- **Aw-vowel un-mask detector (opt-in).** A pre-normalization detector that surfaces a class of Myanmar spelling errors the normalizer previously masked — flat/tall *aa* swaps in the aw-vowel rime (ော ↔ ေါ), e.g. `ခော်` → `ခေါ်`. Each violation emits one gated correction with a deterministic canonical suggestion. Opt-in via the `detect_aw_vowel_unmask` config flag or the `MSC_DETECT_AW_VOWEL_UNMASK` environment variable; default off.
+
+### Changed
+
+- **Hot-path latency reduced ~40% (default, behavior-identical).** Memoized SymSpell nasal-variant validation (instance-level cache keyed by term and level) and the syllable-span probe's per-sentence scoring (LRU). p95 latency drops from 658 ms to 401 ms and mean per-sentence time by ~43%, with byte-identical detections.
+
+### Fixed
+
+- **Concurrent batch checking.** The shared probe score cache is now lock-guarded, so concurrent `check_batch_async` workers can no longer raise a `KeyError` under load.
+- Orthographic-insertion-rescue corrections now carry an explicit confidence so a recovered correction is not silently withheld, and the aw-vowel detector defers only the overlapping span (not the whole token) to the vowel-reorder detector. Plus internal comment, dead-code, and test-coverage cleanups.
+
+### Benchmark
+
+- With the aw-vowel detector enabled, spelling composite improves `0.6520` → `0.6870` (**+0.0350**) on the v1.8.0 benchmark: +98 true positives at zero added false positives, top-1 accuracy 92% on the un-masked corrections, clean false-positive sentences within cap (91/779), p95 382 ms. The benchmark's clean/error annotations were also corrected this release (67 previously-unannotated planted typos). Default behavior (detector off) is unchanged.
+
 ## [1.7.1] - 2026-06-02
 
 ### Added

README.md

@@ -6,7 +6,7 @@
 [![Python Version](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Coverage](https://img.shields.io/badge/coverage-75%25-green)](tests/)
-[![Tests](https://img.shields.io/badge/tests-4%2C940_passed-brightgreen)](tests/)
+[![Tests](https://img.shields.io/badge/tests-5%2C182_passed-brightgreen)](tests/)
 
 ## Overview
 
@@ -34,6 +34,7 @@
 *   **Compound & Morpheme Handling**: DP-based compound resolution, ternary compound splits in morpheme correction, productive reduplication validation.
 *   **AI Semantic Checking (Optional)**: ONNX masked language model for context-aware validation.
 *   **Syllable-Span Probe (opt-in, v1.7.1)**: A frozen-encoder neural probe that improves recall on broken-compound, over-segmentation, and consonant-substitution errors. Three strategies share one small model; default-off, enabled via `use_probe_*` config flags or `MSC_USE_PROBE_*` environment variables.
+*   **Aw-Vowel Un-Mask Detector (opt-in, v1.8.0)**: Surfaces a class of Myanmar aw-vowel spelling errors — flat/tall *aa* swaps in the aw-vowel rime (ော ↔ ေါ, e.g. `ခော်` → `ခေါ်`) — that pre-normalization would otherwise silently repair before validation. Default-off, enabled via the `detect_aw_vowel_unmask` config flag or `MSC_DETECT_AW_VOWEL_UNMASK`.
 *   **Named Entity Recognition**: Heuristic and Transformer-based NER to reduce false positives on names and places.
 
 ### Dictionary Building Pipeline
@@ -65,7 +66,7 @@
 
 Full documentation is available at **[docs.myspellchecker.com](https://docs.myspellchecker.com/)**.
 
-> **What's new in v1.7.1?** See the **[Release Notes](https://docs.myspellchecker.com/reference/release-notes)** for the opt-in **syllable-span probe** — a frozen-encoder neural enhancement (three default-off strategies sharing one small model) that improves recall on broken-compound, over-segmentation, and consonant-substitution errors (+0.0125 composite when enabled). Earlier v1.7.x work added mined-confusable detection, the cross-whitespace and compound-merge probes, the skip-rule confidence gate, and benchmark-hygiene reclassification.
+> **What's new in v1.8.0?** See the **[Release Notes](https://docs.myspellchecker.com/reference/release-notes)** for the opt-in **aw-vowel un-mask detector** — it surfaces Myanmar aw-vowel spelling errors (ော ↔ ေါ) the normalizer previously masked (+0.0350 spelling composite when enabled, at zero added false positives). This release also cuts hot-path latency ~40% (p95 658 → 401 ms, with identical results) and hardens concurrent batch checking. The v1.7.1 syllable-span probe remains available behind its opt-in flags.
 
 ### Getting Started
 *   **[Introduction](https://docs.myspellchecker.com/introduction)**: Overview of the library and its architecture.