DARC supplementary package - submission-ready version

Purpose
-------
This folder supports reproducibility of the numerical manuscript without claiming new RF measurements.

Authoritative exact-benchmark files
-----------------------------------
- expected_reward_matrix.npy
- transition_tensor.npy
- policy_darc.npy
- policy_oracle.npy
- policy_greedy.npy
- policy_ucb.npy
- policy_heuristic.npy
- reproduce_exact_benchmark.py

Running reproduce_exact_benchmark.py reconstructs the exact normalized discounted costs reported in Table 5 from the frozen matrices and policies.

Current manuscript settings
---------------------------
- model_spec_current.py
- DARC_model_card_current.json
The contextual-UCB exploration coefficient is beta = 0.10, matching the policy-development code used for the supplied UCB policy.

Plot/result data
----------------
Representative CSV files for the quantitative result analyses are included (RTT/hops, goodput/loss, ablation, message-size, Pareto, weight sensitivity, dynamic workload, representative repetitions, and seed-level terminal validation summary). High-resolution final versions of all manuscript figures are supplied separately in Figures.zip.

Calibration
-----------
empirical_calibration_template.csv is intentionally blank. The manuscript does not claim empirical RF calibration until real timestamped RTT/loss traces are supplied.

Important scope note
--------------------
The frozen matrices and policy arrays are the authoritative inputs for reproducing the exact Table 5 benchmark values. The final manuscript also reports a 120,000-transition learning-stabilization diagnostic and a W=256 CVaR estimator; those current reporting parameters are recorded in model_spec_current.py and the model card.
