Run the pipeline — tokens, loops, and the law it obeys

Tasks are tokens flowing through the delegated pipeline. Each node fires stochastically (raw success ~ σ_raw, review ~ K/N, correction ~ c); a review loop is a real back-arc — a failed token retries. Watch the empirical end-to-end success converge to the closed-form C_op the theory predicts.

Per-node: raw vs corrected (empirical bars, analytic dashes)

Simulator mso-sim.js on the analytic core mso-core.js. The empirical end-to-end success is asserted to match the closed-form C_op within 0.02 in CI (tests/test_sim_grounding.py) — the discrete Petri-net level and the mean-field MSO level agree. Loops are real cycles here (the DAG analysis sees their steady-state effect via effective catch).