NVIDIA · tlshannon · Jul 20, 2026 · Jul 17, 2026 · Jul 17, 2026 · Jul 17, 2026
diff --git a/docs/CMakeLists.txt b/docs/CMakeLists.txt
@@ -104,7 +104,9 @@ configure_file(${SPHINX_CONF_IN} ${SPHINX_CONF} @ONLY)
 # TODO: set warning as error (-W flag)
 add_custom_command(
   OUTPUT ${SPHINX_INDEX_FILE}
-  COMMAND ${SPHINX_EXECUTABLE} -v -n --keep-going -b html
+  COMMAND ${CMAKE_COMMAND} -E env
+          LD_LIBRARY_PATH=${CUDAQ_LIBRARY_DIR}:$ENV{LD_LIBRARY_PATH}
+          ${SPHINX_EXECUTABLE} -v -n --keep-going -b html
           -c ${CMAKE_CURRENT_BINARY_DIR}
           -Dbreathe_projects.cudaqx=${DOXYGEN_OUTPUT_DIR}/xml
           ${SPHINX_SOURCE} ${SPHINX_BUILD}
@@ -113,6 +115,7 @@ add_custom_command(
     ${SPHINX_SOURCE}/index.rst
     ${DOXYGEN_INDEX_FILE}
     cudaqx-pymodules
+    cudaq-qec-realtime-decoding-simulation
     ${CUDAQX_PUBLIC_PYTHON}
     ${CUDAQX_PUBLIC_RST}
   MAIN_DEPENDENCY ${SPHINX_CONF_IN} ${SPHINX_CONF}

diff --git a/docs/Doxyfile.in b/docs/Doxyfile.in
@@ -28,7 +28,9 @@ ENABLE_PREPROCESSING   = YES
 MACRO_EXPANSION        = YES
 EXPAND_ONLY_PREDEF     = YES
 PREDEFINED             = "__qpu__=" \
-                         "__attribute__(x)="
+                         "__attribute__(x)=" \
+                         "CUDAQ_EXTENSION_CUSTOM_CREATOR_FUNCTION(TYPE,...)=" \
+                         "CUDAQ_EXTENSION_CUSTOM_CREATOR_FUNCTION_WITH_NAME(TYPE,NAME,...)="
 
 #---------------------------------------------------------------------------
 # Configuration options related to the HTML output

diff --git a/docs/sphinx/api/core/cpp_api.rst b/docs/sphinx/api/core/cpp_api.rst
@@ -9,6 +9,16 @@ Namespaces
     :desc-only:
 .. doxygennamespace:: cudaq::qec
     :desc-only:
+.. doxygennamespace:: cudaq::qec::decoding
+    :desc-only:
+.. doxygennamespace:: cudaq::qec::decoding::config
+    :desc-only:
+.. doxygennamespace:: cudaq::qec::detail
+    :desc-only:
+.. doxygennamespace:: cudaq::qec::realtime
+    :desc-only:
+.. doxygennamespace:: cudaq::qec::realtime::experimental
+    :desc-only:
 .. doxygennamespace:: cudaq::qec::steane
     :desc-only:
 .. doxygennamespace:: cudaq::qec::surface_code
@@ -30,6 +40,13 @@ Core
 .. doxygenclass:: cudaqx::extension_point
     :members:
 
+.. doxygendefine:: CUDAQ_EXTENSION_CREATOR_FUNCTION
+.. doxygendefine:: CUDAQ_EXTENSION_CUSTOM_CREATOR_FUNCTION
+.. doxygendefine:: CUDAQ_EXTENSION_CUSTOM_CREATOR_FUNCTION_WITH_NAME
+.. doxygendefine:: CUDAQ_EXT_PT_REGISTER_TYPE
+.. doxygendefine:: INSTANTIATE_REGISTRY_NO_ARGS
+.. doxygendefine:: INSTANTIATE_REGISTRY
+
 .. doxygenclass:: cudaqx::heterogeneous_map
     :members:
 

diff --git a/docs/sphinx/api/qec/cpp_api.rst b/docs/sphinx/api/qec/cpp_api.rst
@@ -18,6 +18,11 @@ Code
 
 .. _qec_stabilizer_grid_cpp:
 
+.. doxygenenum:: cudaq::qec::surface_code::surface_role
+.. doxygenenum:: cudaq::qec::surface_code::sc_orientation
+.. doxygenstruct:: cudaq::qec::surface_code::vec2d
+   :members:
+
 .. doxygenclass:: cudaq::qec::surface_code::stabilizer_grid
     :members:
 
@@ -42,6 +47,11 @@ Detector Error Model
 Decoder Interfaces
 ==================
 
+.. doxygenstruct:: cudaq::qec::decoder_inputs
+    :members:
+
+.. doxygentypedef:: cudaq::qec::decoder_init
+
 .. doxygenclass:: cudaq::qec::decoder
     :members:
 
@@ -82,6 +92,14 @@ Real-Time Decoding
 Parity Check Matrix Utilities
 =============================
 
+.. doxygenenum:: cudaq::qec::sparse_binary_matrix_layout
+
+.. doxygenclass:: cudaq::qec::sparse_binary_matrix
+    :members:
+
+.. doxygenfunction:: cudaq::qec::to_parity_matrix(const std::vector<cudaq::spin_op_term> &, stabilizer_type)
+.. doxygenfunction:: cudaq::qec::to_parity_matrix(const std::vector<std::string> &, stabilizer_type)
+
 .. doxygenfunction:: cudaq::qec::dense_to_sparse(const cudaqx::tensor<uint8_t> &)
 .. doxygenfunction:: cudaq::qec::generate_random_pcm(std::size_t, std::size_t, std::size_t, int, std::mt19937_64 &&);
 .. doxygenfunction:: cudaq::qec::generate_timelike_sparse_detector_matrix(std::uint32_t num_syndromes_per_round, std::uint32_t num_rounds, bool include_first_round = false)
@@ -112,6 +130,8 @@ the ``CUDA_QEC_*`` macros exposed in ``cudaq/qec/logger.h``.
 .. doxygenstruct:: cudaq::qec::detail::forwarded_log_record
     :members:
 
+.. doxygenvariable:: cudaq::qec::detail::realtime_forwarder_default_message_capacity
+
 .. doxygenstruct:: cudaq::qec::detail::forwarder_config
     :members:
 
@@ -141,6 +161,7 @@ Common
 .. doxygentypedef:: cudaq::qec::float_t
 
 .. doxygenenum:: cudaq::qec::operation
+.. doxygenenum:: cudaq::qec::stabilizer_type
 
 .. doxygenfunction:: cudaq::qec::sample_code_capacity(const cudaqx::tensor<uint8_t> &, std::size_t, double)
 .. doxygenfunction:: cudaq::qec::sample_code_capacity(const cudaqx::tensor<uint8_t> &, std::size_t, double, unsigned)

diff --git a/docs/sphinx/api/qec/cpp_realtime_decoding_api.rst b/docs/sphinx/api/qec/cpp_realtime_decoding_api.rst
@@ -22,6 +22,17 @@ Configuration API
 
 The configuration API enables setting up decoders before circuit execution. Decoders are configured using YAML files or programmatically constructed configuration objects.
 
+.. doxygenstruct:: cudaq::qec::realtime::graph_resources
+   :members:
+
+.. doxygenenum:: cudaq::qec::decoding::config::DecoderTransport
+
+.. doxygenstruct:: cudaq::qec::decoding::config::decoder_config
+   :members:
+
+.. doxygenclass:: cudaq::qec::decoding::config::multi_decoder_config
+   :members:
+
 .. doxygenfunction:: cudaq::qec::decoding::config::configure_decoders
 .. doxygenfunction:: cudaq::qec::decoding::config::configure_decoders_from_file
 .. doxygenfunction:: cudaq::qec::decoding::config::configure_decoders_from_str

diff --git a/docs/sphinx/api/qec/python_api.rst b/docs/sphinx/api/qec/python_api.rst
@@ -2,7 +2,6 @@ CUDA-Q QEC Python API
 ******************************
 
 .. automodule:: cudaq_qec
-    :members:
 
 Code
 =============
@@ -47,6 +46,14 @@ Decoder Interfaces
 .. autoclass:: cudaq_qec.DecoderResult
     :members:
 
+.. autoclass:: cudaq_qec.BatchDecoderResult
+    :members:
+
+.. autoclass:: cudaq_qec.AsyncDecoderResult
+    :members:
+
+.. autofunction:: cudaq_qec.get_decoder
+
 Built-in Decoders
 =================
 
@@ -91,6 +98,8 @@ Common
 
 .. autofunction:: cudaq_qec.sample_code_capacity
 
+.. _parity_check_matrix_utilities_python:
+
 Parity Check Matrix Utilities
 =============================
 

diff --git a/docs/sphinx/api/qec/python_realtime_decoding_api.rst b/docs/sphinx/api/qec/python_realtime_decoding_api.rst
@@ -221,4 +221,4 @@ Real-time decoding requires converting matrices to sparse format for efficient d
       config.D_sparse = qec.generate_timelike_sparse_detector_matrix(
           numSyndromesPerRound, numRounds, False)
 
-See also :ref:`Parity Check Matrix Utilities <python_api:Parity Check Matrix Utilities>` for additional PCM manipulation functions.
+See also :ref:`parity_check_matrix_utilities_python` for additional PCM manipulation functions.
diff --git a/docs/sphinx/api/qec/realtime_pipeline_api.rst b/docs/sphinx/api/qec/realtime_pipeline_api.rst
@@ -64,9 +64,3 @@ Pipeline
 
 .. doxygenclass:: cudaq::qec::realtime::experimental::realtime_pipeline
    :members:
-
-.. doxygenstruct:: cudaq::qec::realtime::experimental::realtime_pipeline::Stats
-   :members:
-
-.. doxygenstruct:: cudaq::qec::realtime::experimental::realtime_pipeline::ring_buffer_bases
-   :members:
diff --git a/docs/sphinx/api/qec/tensor_network_decoder_api.rst b/docs/sphinx/api/qec/tensor_network_decoder_api.rst
@@ -1,4 +1,4 @@
-.. class:: cudaq_qec.plugin.decoders.tensor_network_decoder.TensorNetworkDecoder
+.. class:: cudaq_qec.plugins.decoders.tensor_network_decoder.TensorNetworkDecoder
 
     A general class for tensor network decoders for quantum error correction codes.
 

diff --git a/docs/sphinx/api/solvers/cpp_api.rst b/docs/sphinx/api/solvers/cpp_api.rst
@@ -4,9 +4,11 @@ CUDA-Q Solvers C++ API
 .. doxygenclass:: cudaq::solvers::operator_pool 
     :members:
 
-.. doxygenclass:: cudaq::solvers::spin_complement_gsd 
-.. doxygenclass:: cudaq::solvers::uccsd 
-.. doxygenclass:: cudaq::solvers::uccgsd 
+.. doxygenclass:: cudaq::solvers::spin_complement_gsd
+.. doxygenclass:: cudaq::solvers::uccsd
+.. doxygenclass:: cudaq::solvers::uccgsd
+.. doxygenclass:: cudaq::solvers::ceo
+.. doxygenclass:: cudaq::solvers::upccgsd
 
 .. doxygenclass:: cudaq::solvers::qaoa_pool 
 
@@ -75,8 +77,8 @@ CUDA-Q Solvers C++ API
 
 .. doxygenstruct:: cudaq::solvers::qaoa_result
     :members:
-.. doxygenfunction:: cudaq::solvers::qaoa(const cudaq::spin_op &, const cudaq::spin_op &, const optim::optimizer &, std::size_t, const std::vector<double> &, const heterogeneous_map)
-.. doxygenfunction:: cudaq::solvers::qaoa(const cudaq::spin_op &, const optim::optimizer &, std::size_t, const std::vector<double> &, const heterogeneous_map)
+.. doxygenfunction:: cudaq::solvers::qaoa(const cudaq::spin_op &, const cudaq::spin_op &, optim::optimizer &, std::size_t, const std::vector<double> &, const heterogeneous_map)
+.. doxygenfunction:: cudaq::solvers::qaoa(const cudaq::spin_op &, optim::optimizer &, std::size_t, const std::vector<double> &, const heterogeneous_map)
 .. doxygenfunction:: cudaq::solvers::qaoa(const cudaq::spin_op &, std::size_t, const std::vector<double> &, const heterogeneous_map)
 .. doxygenfunction:: cudaq::solvers::qaoa(const cudaq::spin_op &, const cudaq::spin_op &, std::size_t, const std::vector<double> &, const heterogeneous_map)
 .. doxygenfunction:: cudaq::solvers::get_num_qaoa_parameters(const cudaq::spin_op &, const cudaq::spin_op &, std::size_t, const heterogeneous_map)

diff --git a/docs/sphinx/components/qec/introduction.rst b/docs/sphinx/components/qec/introduction.rst
@@ -507,10 +507,9 @@ You must pass ``distance`` when constructing the code; there is no default.
 
 The :ref:`stabilizer_grid <qec_stabilizer_grid_python>` helper documents how
 stabilizers and data qubits are indexed on the grid and provides helpers to
-print the layout. **Python:** :class:`cudaq_qec.stabilizer_grid` — see
-:ref:`qec_stabilizer_grid_python`. **C++:**
+print the layout. **Python:** :ref:`cudaq_qec.stabilizer_grid <qec_stabilizer_grid_python>` — **C++:**
 :cpp:class:`cudaq::qec::surface_code::stabilizer_grid` — see
-:ref:`qec_stabilizer_grid_cpp`. The header :file:`cudaq/qec/codes/surface_code.h`
+:ref:`API <qec_stabilizer_grid_cpp>`. The header :file:`cudaq/qec/codes/surface_code.h`
 contains the full declaration.
 
 **Stabilizer measurement schedule**

diff --git a/docs/sphinx/conf.py.in b/docs/sphinx/conf.py.in
@@ -93,7 +93,20 @@ master_doc = 'index'
 # List of patterns, relative to source directory, that match files and
 # directories to ignore when looking for source files.
 # This pattern also affects html_static_path and html_extra_path.
-exclude_patterns = ['_templates']
+# Fragment files pulled in via ``.. include::`` must not also be processed as
+# standalone documents — doing so causes duplicate C++ / Python domain
+# declarations in the Sphinx domain registry.
+exclude_patterns = [
+    '_templates',
+    'api/qec/nv_qldpc_decoder_api.rst',
+    'api/qec/sliding_window_api.rst',
+    'api/qec/trt_decoder_api.rst',
+    'api/qec/tensor_network_decoder_api.rst',
+    'api/qec/python_realtime_decoding_api.rst',
+    'api/qec/cpp_realtime_decoding_api.rst',
+    'api/qec/realtime_pipeline_api.rst',
+    'api/solvers/gqe_api.rst',
+]
 
 # The reST default role (used for this markup: `text`) to use for all documents.
 default_role = 'code'  # NOTE: the following may be a better choice to error on the side of flagging anything that is referenced but but not declared
@@ -156,6 +169,34 @@ def process_docstring(app, what, name, obj, options, lines):
     for i, line in enumerate(lines):
         lines[i] = fix_string(line)
 
+def _fixup_class_modules():
+    # Types bound in the qecrt nanobind submodule have __module__ set to the
+    # internal extension path.  Autodoc treats any class whose __module__ doesn't
+    # match the documenting module as an alias and emits "alias of <internal path>"
+    # instead of real documentation; stringify_annotation() also uses __module__ to
+    # build type-annotation strings in function signatures.  Fix up every public
+    # type in cudaq_qec whose module still points at the internal path so that
+    # both autoclass and autofunction produce clean output automatically.
+    try:
+        import cudaq_qec
+        _internal_prefix = 'cudaq_qec._pycudaqx'
+        for _name in dir(cudaq_qec):
+            if _name.startswith('_'):
+                continue
+            _obj = getattr(cudaq_qec, _name, None)
+            if isinstance(_obj, type) and \
+                    getattr(_obj, '__module__', '').startswith(_internal_prefix):
+                try:
+                    _obj.__module__ = 'cudaq_qec'
+                except (AttributeError, TypeError):
+                    pass
+    except ImportError:
+        pass
+
+_fixup_class_modules()
+del _fixup_class_modules
+
+
 def setup(app):
     app.connect('autodoc-process-signature', process_signature)
     app.connect('autodoc-process-docstring', process_docstring)
@@ -183,6 +224,16 @@ redirects = {"versions": "../latest/releases.html"}
 nitpick_ignore = [
     ('cpp:identifier', 'pid_t'),
     ('cpp:identifier', 'uint8_t'),
+    ('cpp:identifier', 'uint32_t'),
+    ('cpp:identifier', 'int64_t'),
+    ('cpp:identifier', 'index_type'),
+    ('ref', 'graph__resources_8h'),
+    ('ref', 'decoder__config__schema_8h'),
+    ('cpp:identifier', 'cudaq::pauli_word'),
+    ('cpp:identifier', 'cudaGraphExec_t'),
+    ('cpp:identifier', 'cudaStream_t'),
+    ('cpp:identifier', 'SIZE_MAX'),
+    ('cpp:identifier', 'uint64_t'),
     ('cpp:identifier', 'details::tensor_impl<Scalar>::scalar_type'),
     ('cpp:identifier', 'cudaqx'),
     ('cpp:identifier', 'size_t'),
@@ -199,9 +250,28 @@ nitpick_ignore = [
     ('cpp:identifier', 'cudaq::qview<>'),
     ('cpp:identifier', 'cudaq::measure_result'),
     ('py:class', 'SpinOperator'),
+    ('py:class', 'cudaq.SpinOperator'),
+    ('py:class', 'cudaq::sum_op<cudaq::spin_handler>'),
+    ('py:class', 'list[cudaq::sum_op<cudaq::spin_handler>]'),
+    ('py:class', 'tuple[float, list[float], list[cudaq::sum_op<cudaq::spin_handler>]]'),
+    ('py:class', 'pauli_word'),
+    ('py:class', 'PauliWord'),
     ('py:class', 'numpy.int32'),
     ('py:class', 'numpy.uint8'),
-    ('py:class', 'cudaq.mlir._mlir_libs._quakeDialects.cudaq_runtime.qview')
+    ('py:class', 'cudaq.mlir._mlir_libs._quakeDialects.cudaq_runtime.qview'),
+    ('py:class', 'cudaq_qec.surface_role'),
+    ('py:class', 'cudaq_qec.operation'),
+    ('py:class', 'sc_orientation'),
+    ('py:class', 'vec2d'),
+]
+
+# Suppress cross-reference warnings for types whose fully-qualified path still
+# contains the internal pybind11 module name.  This happens when Sphinx
+# extracts the cross-reference from the raw nanobind-generated signature before
+# the process_docstring / process_signature hooks have a chance to strip the
+# prefix via doc_replacements.
+nitpick_ignore_regex = [
+    ('py:class', r'cudaq_qec\._pycudaqx_qec_the_suffix_matters_cudaq_qec\..*'),
 ]
 
 napoleon_google_docstring = True

diff --git a/docs/sphinx/examples_rst/qec/decoders.rst b/docs/sphinx/examples_rst/qec/decoders.rst
@@ -132,7 +132,7 @@ Exact Maximum Likelihood Decoding with NVIDIA Tensor Network Decoder
 Starting with CUDA-Q QEC v0.4.0, a GPU-accelerated Maximum Likelihood Decoder is included with the
 CUDA-Q QEC library. The library follows the CUDA-Q decoder Python interface, namely :class:`cudaq_qec.Decoder`.
 At this time, we only support the Python interface for the decoder, which is
-available at :class:`cudaq_qec.plugins.decoders.tensor_network_decoder.TensorNetworkDecoder`.
+available at :ref:`TensorNetworkDecoder <tensor_network_decoder_api_python>`.
 As documented in the API sections :ref:`tensor_network_decoder_api_python`, there are many configuration options
 that can be passed to the constructor. The decoder requires Python 3.11 or higher.
 
@@ -170,7 +170,7 @@ to decode surface code syndromes using PyTorch and Stim, exporting the model to
 deploying it with the TensorRT decoder for accelerated inference.
 
 Overview of the Training-to-Deployment Pipeline
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 The workflow consists of three main stages:
 
@@ -186,7 +186,7 @@ The workflow consists of three main stages:
    be loaded by the TensorRT decoder for optimized GPU inference in production QEC workflows.
 
 Training a Neural Network Decoder with PyTorch and Stim
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 The following example shows how to generate training data using Stim's built-in surface code
 generator, train an MLP decoder with PyTorch, and export the model to ONNX format.
@@ -197,7 +197,7 @@ For instructions on installing PyTorch, see :ref:`Installing PyTorch <installing
    :start-after: [Begin Documentation]
 
 Using the TensorRT Decoder in CUDA-Q QEC
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 Once you have a trained ONNX model, you can load it with the TensorRT decoder for accelerated
 inference. The decoder can be used in both C++ and Python workflows.
@@ -279,7 +279,7 @@ you can load it directly:
                                 engine_load_path="surface_code_decoder.trt")
 
 Converting ONNX Models to TensorRT Engines
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 For production deployments where initialization time is critical, you can pre-build a TensorRT
 engine from your ONNX model using the ``trtexec`` command-line tool that comes with TensorRT:
@@ -310,7 +310,7 @@ Pre-built engines offer several advantages:
 
 
 Dependencies and Requirements
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 The TensorRT decoder requires:
 
@@ -324,7 +324,7 @@ For training:
 - **Stim**: For quantum circuit simulation and data generation
 
 See Also
-^^^^^^^^
+~~~~~~~~
 
 - :class:`cudaq_qec.Decoder` - Base decoder interface
 - `ONNX <https://onnx.ai/>`_ - Open Neural Network Exchange format