Co-Clearing Shared Distributed-Energy-Resource Flexibility Across TSO and DSO Markets — Verification Bundle: Evaluation Harness, Raw Results, Derived Tables and a Self-Contained Verifier

Complete verification bundle for the manuscript "Co-Clearing Shared Distributed-Energy-Resource Flexibility Across TSO and DSO Markets: A Merit-Order Algorithm with Verifiable Settlement" (revised submission to Smart Energy, Elsevier, 2026; manuscript SEGY-D-26-00175). The paper analyses what happens when one fleet of small distributed energy resources (DERs) is offered simultaneously to a transmission system operator for balancing and to a distribution system operator for congestion management, so that the two markets contend for the same physical headroom within a single 15-minute interval. This deposit contains everything needed to reproduce every number the paper claims. The bundle is self-contained. The ten experiment scripts are pure numpy/scipy generators seeded at 20260629; they read no external data file, require no network access and use no restricted input. Running verify_results.py (Python standard library only) re-derives all 112 headline claims of Section 5 and Supplementary S3-S11 from the raw results and prints a PASS/FAIL line for each; it exits 0 when every claim reproduces. Re-running the harness reproduces results_raw/ bit-for-bit, with the sole exception of the wall-clock timing fields, which are machine-dependent. New in v1.2.0. Five experiments were added in response to the referees, each with its own raw results file and derived table, and the verifier was extended from 75 to 112 claims: minlot_experiment.py: what a minimum tradeable lot costs, given that every headline number uses the divisible relaxation. Solves the indivisible problem exactly as a MILP across lot sizes from 0.10 to 2.00 kW. network_experiment.py: which network limits the structural results survive. Nested (radial) element caps keep total unimodularity and a depth-adjusted guarantee; overlapping (meshed) supports destroy integrality, demonstrated on an integral instance with a strictly fractional LP optimum. storage_experiment.py: what changes when the headroom has a memory. Lossless storage keeps the problem a min-cost flow on a time-expanded network; round-trip losses destroy integrality; and the cost of clearing quarter-by-quarter is measured against the multi-period optimum. stress_experiment.py: co-clearing under non-normal operation: asset outage, a local congestion emergency, a system balancing emergency, and a lifted price cap. scaling_impact.py: an indicative extrapolation to national scale in financial and emissions terms. This is an extrapolation rather than an experiment and is written to keep that visible, separating quantities measured by the harness from quantities sourced externally and from the participation rate, which is swept. Two inputs, the flexible share of demand and the avoided-loss coefficient of the island feeder, come from the project's network study of that feeder, which is not publicly released; the script records their provenance. It also prints a pan-European variant that the paper does not use. Contents verify_results.py: self-contained verifier; re-derives every headline claim from the raw results. harness/: the ten experiment scripts plus make_tables.py, which derives every CSV in tables/ from results_raw/ so that no table can drift from the results it summarises. results_raw/: untouched JSON output of the ten scripts; the primary record. tables/: every manuscript and supplementary table as CSV. calibration/: the published pilot aggregates the reconstruction is calibrated to. MANIFEST.sha256, LICENSE, requirements.txt, CITATION.cff. What this deposit does not contain. The pilot's raw per-offer log, participant-level records and operator network data are restricted under the General Data Protection Regulation and under confidentiality agreements with the two operators, and are available from the corresponding author only on reasonable request and with the permission of both. No result in the paper depends on them: the reconstruction is calibrated to published per-product marginals, and the paper states explicitly what that calibration does and does not fix and stress-tests its headline results against the freedom it leaves.