What problem does it solve? Writing a professional, high-coverage hardware testbench by hand is slow and error-prone, and a hand-waved skeleton can silently pass without verifying anything. This Skill derives a real functional-verification environment directly from a design's Phase-1 L-docs, so every design gets a scoreboard, functional coverage, and assertions instead of a vacuous pass. ## Core Features & Use Cases - Deterministic testbench generation: The program professional_tb_gen.py (MCP tool eda_professional_tb) reads phase1/generated_docs/L*.json and emits a cocotb testbench with clock/reset from L8/L9, a scoreboard, cocotb-coverage covergroups, an L28 coverage model, an L29 SVA assertion set, and a verification plan. - Three-tier reference-model strategy: closed-form arithmetic oracles, auto-derived bounded-latency streaming scoreboards for serial datapaths, and a clearly-marked expert reference hook for CPU/SoC/DSP classes that TestSkips until filled — never a silent vacuous pass. - Coverage review loop: Read coverage_<top>.xml and the L28 model, add directed vectors or tighten constrained-random constraints until functional bins hit 100%. - Use Case: After Phase-2 RTL is authored for a bit-serial multiplier, run Step 4 to auto-derive latency and bit order from a calibration vector and verify 208/208 vectors against the closed-form reference. ## Quick Start Ask the AI to generate a professional testbench for the current project from its Phase-1 L-docs and run the cocotb simulation.
