docs: record FunASR v1.4.14 weekly audit (#3643)

Record the verified v1.4.14 weekly release audit and the decision to skip an empty v1.4.15.

Signed-off-by: LauraGPT <[email protected]>

docs/community_growth_20k.md

@@ -43,6 +43,14 @@ As of 2026-09-03 09:22 UTC, the ecosystem has 37,080 combined GitHub stars, or 5
 | `QwenAudio/SenseVoice` | 9,211 | 817 | 7 | 0 | 2026-08-31 |
 | `modelscope/FunClip` | 6,211 | 742 | 5 | 0 | 2026-09-01 |
 
+### 2026-09-04 FunASR v1.4.14 weekly release audit
+
+- The signed `v1.4.14` tag still peels exactly to `a23946a8b3e9146f9bf80e18f768298f0cd000b7`, and `main` initially had no commit or source diff after that tag, so the weekly run correctly skipped an empty `v1.4.15`. The public PyPI wheel SHA-256 is `8f1bceeb183359fb44402dea16200415a838da314742aa0c80752efe0e229e6c`; the sdist SHA-256 is `5dd3ab30f6f2f26207992173a1da54d2f0f3f74f64d0823a05a9eab9f93a704f`. Isolated installation, metadata, entry points, package data, compileall, PEP 517, Twine, and source-to-artifact content checks passed.
+- The unique public [v1.4.14 GitHub Release](https://github.com/modelscope/FunASR/releases/tag/v1.4.14) now has 12 uploaded assets: two Python distributions, nine mirrored `runtime-llamacpp-v0.2.6` archives, and `SHA256SUMS`. The manifest SHA-256 is `6910be0401e6315f6260f8d8cf2f01274ec123ad96231ef2647ec0250fa5941d` and exactly matches all 11 other attached asset digests. The separate [Blackwell CUDA runtime](https://github.com/modelscope/FunASR/releases/download/runtime-llamacpp-v0.2.6/funasr-llamacpp-windows-x64-cuda-blackwell.zip) remains on the runtime release. The release body was repaired to restore the generated runtime download section without changing its highlights.
+- FunASR PR [#3640](https://github.com/modelscope/FunASR/pull/3640) merged as verified commit `85106d6151bb964f830e4f54d6b2960f02576072`, updating the stale release-contract tests from 1.4.13 to 1.4.14. The release/documentation audit passed 210 tests; pre-existing historical commits without DCO trailers were recorded but immutable history was not rewritten.
+- FunASR PR [#3641](https://github.com/modelscope/FunASR/pull/3641) merged as verified commit `2e2f18bce1201651335b689a20ad6f154b873f94`, adding bilingual release guides, exact 12-asset/9-runtime contracts, sitemap and manifest coverage, and a 390 px long-hash overflow regression. Validation passed 106 Python site tests, 29 Playwright tests, and two identical exact-main builds of 112 HTML pages. Production release `20260904T041224Z` was atomically deployed to the [Chinese guide](https://www.funasr.com/blog/funasr-v1-4-14-portable-source-release.html) and [English guide](https://www.funasr.com/en/blog/funasr-v1-4-14-portable-source-release.html); archive SHA-256 is `7477817fe3faf239e746b9820b9c5ec1ad9423ae983c7f1353f9fd037bc77d6a`, and `20260903T172000Z` remains available for rollback. The pre-switch live-site backup SHA-256 is `07b5529d71717fc1a80df9f9999f24ba104a98d4a666c1f6e474807defcaf61e`.
+- Bilingual announcement [#3642](https://github.com/modelscope/FunASR/discussions/3642) publishes the pinned install command, exact artifact hashes, attached-versus-Blackwell runtime boundary, verification evidence, and both production guides. Current model-card install guidance is version-agnostic (`funasr>=1.3.26`), so no unrelated content-only rewrite was made.
+
 ### 2026-08-21 FunASR v1.4.3 stable release
 
 - FunASR PR [#3519](https://github.com/modelscope/FunASR/pull/3519) merged as `eedd4e22d10dc2e81d9c2bb321edb3750253964b`. The release adds the optional `AutoModel(vad_model="silero-vad")` adapter with millisecond segments, threshold controls, 8/16 kHz input, ONNX mode, and bounded segment lengths. Large embedding sets with a known speaker count now use fixed-K clustering instead of dense spectral clustering.