This document maps the executable path from runtime task orchestration to SYCL-enabled acceleration surfaces and highlights current implementation gaps. It is intended as an auditable bridge between architectural intent and implementation reality.

1. Runtime task submission
   • File: q_mini_wasm_v2/runtime/orchestrator.hpp
   • Entry points:
     • submit_tableau_update(...)
     • submit_moe_routing(...)
     • submit_ff_training(...)
2. Task execution loop
   • File: q_mini_wasm_v2/runtime/orchestrator.cpp
   • Worker threads pop lambda tasks from task_queue_ and execute on CPU.
3. Core subsystem invocation
   • Stabilizer: q_mini_wasm_v2/core/stabilizer/*
   • MoE: q_mini_wasm_v2/core/moe/*
   • Forward-Forward: q_mini_wasm_v2/core/learning/*
4. SYCL acceleration surfaces (declared)
   • q_mini_wasm_v2/sycl/tableau_kernels.hpp/.cpp
   • q_mini_wasm_v2/core/qgnn/message_passing_sycl.hpp/.cpp
5. Build-time SYCL enablement
   • File: q_mini_wasm_v2/CMakeLists.txt
   • Gate: -DUSE_SYCL=ON
   • Providers: IntelSYCL / AdaptiveCpp / oneAPI env fallback
6. WASM bridge interface
   • File: q_mini_wasm_v2/dll/wasm_bridge/wasm_api.cpp

• Exposes SYCL init/map/dispatch calls with extensive CPU fallback behavior.

Stage	Artifact	Current State	Evidence
Runtime orchestration	runtime/orchestrator.*	Implemented CPU thread pool	Task queue and worker loop are active in orchestrator.cpp
SYCL kernel declarations	sycl/tableau_kernels.*	Implemented but mostly isolated	Primarily referenced by tests/test_sycl_kernels.cpp
QGNN SYCL kernels	core/qgnn/message_passing_sycl.*	Implemented	Dedicated SYCL queue and kernels present
Build integration	CMakeLists.txt	Implemented	USE_SYCL option and package detection in CMake
Runtime→SYCL wiring	Orchestrator to SYCL calls	Partial / weak coupling	submit_* methods in runtime/orchestrator.cpp call CPU lambdas/core APIs directly; no direct dispatch to sycl_kernels::parallel_apply_*
WASM SYCL bridge	dll/wasm_bridge/wasm_api.cpp	Partial with CPU fallback dominance	Multiple explicit "CPU fallback" and future SYCL TODO-style notes

• RuntimeOrchestrator does not directly dispatch to sycl_kernels::parallel_apply_* paths.
• Existing SYCL kernels are validated mostly by tests and standalone modules.

• wasm_api.cpp contains many paths where SYCL is initialized but execution still uses CPU fallback logic.
• Several command handlers include explicit placeholders for future SYCL kernel dispatch.

• Some architecture docs overstate purity/completeness versus measured enforcement output.
• This traceability file is the canonical reconciliation point for runtime/SYCL execution path status.

• runtime/orchestrator.cpp
• submit_tableau_update, submit_moe_routing, and submit_ff_training enqueue CPU lambdas; no direct invocation of sycl_kernels::* APIs in the orchestrator path.
• sycl/tableau_kernels.cpp
• File includes explicit section header: Parallel Tableau Operations (CPU fallback without SYCL).
• Routing helper currently uses floating point (std::vector<double> logits), which conflicts with strict GF(3)-only core purity expectations.
• dll/wasm_bridge/wasm_api.cpp
• Contains explicit fallback notes (using CPU fallback, Future: Implement proper SYCL kernel dispatch) in command execution paths.

1. Runtime orchestrator task types map to concrete SYCL dispatch adapters in production code paths (not only tests).
2. Bridge command handlers execute SYCL kernels where feature flags indicate SYCL availability; fallback paths are explicit and intentionally gated.
3. Strict GF(3) constrained modules avoid floating-point routing/attention logic.
4. CI evidence links each traceability row to executable validation (tests and/or runtime probes).

• All recommended remediation must preserve:
  • GF(3)-only stabilizer loop semantics in core constrained paths
  • Forward-Forward only learning updates (no global gradient methods)
  • MCP-only communication for agent/gateway pathways

1. Wire orchestrator task types to explicit SYCL dispatch adapters where available.
2. Replace CPU-fallback placeholders in wasm_api.cpp command paths with real SYCL kernels or mark as intentionally disabled behind strict feature flags.
3. Keep this matrix updated with each integration milestone and CI evidence.