Scientific Reports Supplementary Data 1 — Core reproducibility

Manuscript: Continuation-specific reliability assessment of receding-horizon wind–battery–reserve dispatch
Prepared: 2026-10-03

CONTENTS
- core_reproducibility/: frozen causal inputs, model, controller outputs, analysis code, independent audit, and reproduce_round3.py. Raw AEMO daily ZIP archives are intentionally excluded to keep this file below the journal's 50 MB supplementary-file limit.
- timing_provenance/: compact source-recovery and AEMO timing-audit outputs.
- robustness/: frozen latency, resource, block-length, gate, leave-one-day-out, scoring-definition and obligation-level diagnostics plus reproducible scripts.
- auxiliary_scripts/: provenance scripts for timing audit, manuscript construction and figure regeneration.
- MANIFEST_SHA256.csv: SHA-256 manifest for every staged file except the manifest itself.

CORE ANALYSIS REPRODUCTION (NO NETWORK)
From core_reproducibility/:
  python reproduce_round3.py --analysis-only --output <output_dir>
recomputes the principal analysis tables from the saved controller outputs.

FULL OFFLINE CONTROLLER RECOMPUTATION
Supplementary Data 2 contains the 30 official AEMO daily DISPATCH_SCADA ZIP archives. Before using:
  python reproduce_round3.py --recompute --workers 4 --output <output_dir>
copy Supplementary Data 2's raw/ directory to:
  core_reproducibility/confirmation/raw/
No network access is then required.

The primary 20-min analysis is frozen. Sensitivity outputs are diagnostics and were not used to select the headline result. Public archive timestamps do not establish field telemetry packet receipt or external NEMWeb client receipt time.

INTERNAL DIRECTORY LABEL NOTE
The historical internal directory name `confirmation` is retained to preserve the frozen file structure and reproducibility hashes. In the final manuscript, that cohort is described more precisely as the Australian external stress test. The two labels refer to the same 11-28 September 2026 AEMO cohort.

FINAL POST-FREEZE OBLIGATION-LEVEL SENSITIVITY
robustness/obligation_level_sensitivity.csv reports a full DMPC rerun after multiplying the already-defined synthetic obligation by 0.9 or 1.1 and clipping to [0,1], with the frozen primary run as scale 1.0. The reference-dependent sign reversal persists in both cohorts and both 7-day paired percentile ranges exclude zero at every tested scale. This is a level sensitivity only; it is not validation of a different commercial product or issuance rule. The reproducible driver is robustness/obligation_level_sensitivity.py.

PORTABILITY REPAIR AND VERIFICATION 2026-10-06
The latency and resource sensitivity drivers now resolve core_reproducibility relative to the package. This changes only historical filesystem paths, with no scientific parameter or algorithm changes. The original supplied archive remains unchanged; this folder is a repaired working copy. verification_20261006 records original/new hashes and fresh verification results. The repaired 20-min builder matches archived input arrays to maximum absolute difference 3.50e-15. The repaired resource driver reproduces all Australian baseline controller arrays except measured solve time. Independent physical replay, causal-input audit, raw AEMO reconstruction, target-matched scoring analysis, scoring-definition sensitivities, and obligation-level sensitivities were rechecked. Obligation CSV values match at their archived 3-decimal precision.
The four files under auxiliary_scripts retain historical construction/provenance paths and are not turnkey final manuscript/figure reconstruction entry points. Use core_reproducibility/reproduce_round3.py for scientific result reconstruction. The complete altered latency/resource grid was not independently rerun in this verification.
