[vedika-saravanan]

Description

Adds a Decoding Server User Guide for using the prebuilt Docker package
without building the server.

• Explains the configuration, measurement and correction workflow with a diagram.
• Uses one reusable container and a shared host directory throughout the guide.
• Covers generated surface-code A/B configurations, mapped BULK/FINAL inputs,
  software playback, CUDA-Q clients and third-party runtimes.
• Links the color-code AI example and the task-graph and playback reference pages.
• Explains custom decoder implementation, SDK extraction, compilation outside
  the server container, plugin deployment and startup verification.
• Documents how to confirm external-server routing and troubleshoot common failures.
• Corrects mapped-input handling in MESSAGE_FORMAT.md and updates the existing
  color-code example for the Python playback API.

No library or server runtime changes and no new dependencies.

Validation

• Strict full-site Doxygen/Sphinx build with warnings treated as errors.
• Generated memory configuration, server startup and UDP playback with the
  expected correction.
• Compiled CUDA-Q client checks with server-side decode/error counters.
• Custom plugin SDK extraction, external compilation, deployment and startup;
  missing-plugin negative control.
• Full tracked-tree formatting: clang-format 22 and YAPF 0.43.0, plus
  git diff --check.

These are functional checks, not hardware or performance qualification.

Release image

The public decoding-server registry path remains a clearly marked placeholder
pending confirmation of the final release image name.

[eliotheinrich]

Thanks, Vedika; I left a few comments.

Also, do you think this should live in the Libraries section of the docs or the Examples? Right now it reads more like an example, and the new file decoding_server.rst lives in examples_rst, but it is only linked to from components/qec/index.rst, and not the Examples toctree.

[vedika-saravanan]

Regarding documentation placement:

Thanks, I agree the walkthrough fits better under Examples. It is now listed under Examples -> Realtime Decoding, with a cross-link from Libraries. Updated in f6d6108f.

[eliotheinrich]

Since this path should be established by the release, can we put it here? I believe it is already named: realtime_decoding_demo.rst:11-15 gives nvcr.io/nv-quantum/cudaq-qec/cudaq-decoding:cu13-latest

[vedika-saravanan]

I synced with @anjbur earlier, and we agreed to keep the placeholder until she finalizes the release image path. She’ll update it in a follow-up PR

[bmhowe23]

@vedika-saravanan - can you please be sure to add a section on how users can write their own decoders and drop in the resulting .so file into the decoder-plugins directory, too? While users may develop the decoders in our decoding server Docker container for convenience, I don't think that is the typical use-case expected in the long term. That is - in the long term, we expect the Decoding server to not necessarily have C++ compilers in it.

[vedika-saravanan]

@vedika-saravanan - can you please be sure to add a section on how users can write their own decoders and drop in the resulting .so file into the decoder-plugins directory, too? While users may develop the decoders in our decoding server Docker container for convenience, I don't think that is the typical use-case expected in the long term. That is - in the long term, we expect the Decoding server to not necessarily have C++ compilers in it.

Addressed in e360ddb. The guide now covers obtaining the matching SDK and building and deploying a custom decoder plugin without requiring a compiler in the server image.