Minimize public model card details

README.md

@@ -1,54 +1,32 @@
 ---
 license: apache-2.0
-library_name: coreml
-pipeline_tag: automatic-speech-recognition
 inference: false
 tags:
-  - coreml
-  - apple-silicon
-  - macos
-  - on-device
   - doppel
+  - runtime-artifact
 ---
 
-# Doppel ASR companion
+# Doppel runtime artifact
 
-This repository is Doppel's distribution channel for an app-specific Core ML
-conversion of [`Qwen/Qwen3-ASR-0.6B-hf`](https://huggingface.co/Qwen/Qwen3-ASR-0.6B-hf).
-It exists so supported Doppel releases can perform speech recognition on the
-user's Mac.
+This repository contains a versioned runtime artifact consumed by compatible
+Doppel releases.
 
-## Intended use
+It is not a standalone model, SDK, or supported public API. Its format,
+compatibility contract, and behavior are private implementation details and
+may change without notice. No external loading, conversion, integration, or
+evaluation guidance is provided.
 
-These artifacts are intended to be downloaded, verified, and invoked by
-Doppel on Apple-silicon Macs running macOS 15 or later. Inference stays on the
-device.
+The repository contains no customer data.
 
-This is not a standalone ASR distribution or a supported developer API. The
-repository does not include Doppel's audio preprocessing, tokenizer assets,
-runtime orchestration, output handling, or compatibility logic. It is not
-expected to work with generic Transformers, Core ML sample code, or Hugging
-Face hosted inference. For general-purpose use, start with the upstream model.
+<details>
+<summary>License and attribution</summary>
 
-## Artifact and integrity notes
-
-The compiled Core ML components are modified from the upstream checkpoint for
-Doppel's fixed runtime contract. `manifest.json` is the authoritative artifact
-inventory and records the integrity hashes checked by the app.
-
-The repository contains no user audio, transcripts, prompts, indexes,
-telemetry, or other customer data.
-
-## Limitations
-
-Speech recognition can be inaccurate, especially with noise, overlapping
-speakers, uncommon names, code switching, or domain-specific language. Doppel
-must not treat a transcript as a verbatim or safety-critical record without
-human review.
-
-## License and attribution
-
-The upstream model is provided under the
+This distribution includes modified material derived from
+[`Qwen/Qwen3-ASR-0.6B-hf`](https://huggingface.co/Qwen/Qwen3-ASR-0.6B-hf),
+provided under the
 [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0).
-This converted distribution remains subject to the applicable upstream terms.
-Doppel provides no support or warranty for use outside the Doppel app.
+The source link is provided for attribution and does not define a supported
+Doppel compatibility contract. No warranty or support is provided for use
+outside Doppel.
+
+</details>