SUPPLEMENTARY DATA 1: CODE, PROCESSED INPUTS AND REPRODUCIBILITY
Continuation-specific reliability assessment of receding-horizon wind-battery-reserve dispatch
Journal-supplement release: 6 October 2026

START HERE / 使用说明
This is the canonical Supplementary Data 1 for the paper. The companion
Supplementary_Data_2_AEMO_Raw_Sources.zip contains all 30 downloaded AEMO daily
archives. The journal manuscript, Supplementary Information PDF and four
publication figures are separate submission files. Transport ZIP parts are
only a download mechanism; do not submit them instead of the restored files.
本包为论文的正式补充数据1。补充数据2包含30个已下载的AEMO原始日包。
下载分卷仅用于传输；期刊投稿应上传还原后的完整ZIP，不上传传输分卷。

REPRODUCIBILITY BOUNDARY
All numerical inputs used by the supported normalized-input workflow, the
frozen model, three controllers, saved controller arrays, analysis tables,
sensitivity drivers/tables, source timing records and scientific verification
records are included. This does not claim that all upstream raw Kelmarsh
SCADA files or the historical original forecast-generation source are included
or that the raw-turbine-to-aggregate chain was independently reconstructed.
See DATA_SOURCES_AND_RIGHTS.txt, DATA_INVENTORY.csv and NPZ_ARRAY_SCHEMA.csv.

REQUIREMENTS
Verified runtime: Python 3.12.14; NumPy 2.3.5; SciPy 1.17.0; pandas 2.2.3.
HiGHS is included with SciPy; no commercial optimizer is needed.
In an isolated Python environment:
  python -m pip install -r requirements.txt
Dependency installation may need Internet access; the supplied numerical
workflows are offline after installation. Python is needed only to recompute
science, not to restore the downloadable files on Windows.

1. CHECK THE EXTRACTED PAYLOAD
From this extracted directory:
  python verify_manifest.py
MANIFEST_SHA256.csv is the current payload manifest. Historical manifests in
verification_20261006/ and package_manifest.json are provenance snapshots,
not the current file list. Keep newly generated output outside this directory.

2. RECOMPUTE THE PRIMARY STATISTICS
  python core_reproducibility/reproduce_round3.py --analysis-only --output ../reproduced_analysis
This recomputes three cohorts' statistics from saved controller results,
including paired event gaps, 2,000-replicate block bootstrap, resources and
target-matched scores. Expected: all three groups PASS in
comparison_to_archived.json. Choose a new empty output directory.

3. OPTIONAL FULL CONTROLLER RECOMPUTATION
  python core_reproducibility/reproduce_round3.py --recompute --dry-run --output ../recompute_check
  python core_reproducibility/reproduce_round3.py --recompute --workers 4 --output ../recomputed_full
This uses bundled processed UK input, frozen model and Australian normalized
CSV; Supplementary Data 2 is not needed for this entry point. The full controller
workflow was dry-run checked, not rerun end-to-end for this packaging release.
The saved-output statistics were freshly recomputed after fresh ZIP extraction.
See PACKAGING_VALIDATION.json for the exact checks and limits.

4. OPTIONAL RAW AEMO REPARSE
Unzip Supplementary_Data_2_AEMO_Raw_Sources.zip separately. Copy the complete
core_reproducibility/confirmation directory to a NEW working directory. Copy
the raw directory from Data 2 into that working copy. From that copied
confirmation directory run:
  python acquire_scada_structure.py --reference-mw 453 --reference-description "Fixed project nameplate normalization reference; not current export capacity"
All 30 bundled daily archives make this step offline. Do not overwrite the
frozen source records. The archived October 6 reparse compared normalized
power, proxy MW and provenance byte-for-byte.

5. OTHER CHECKS AND FIGURES
CODE_ENTRYPOINTS.txt lists primary and optional scripts. Figure reconstruction
uses publication_figures/build_figures.py and its own pinned requirements.
The same assets are also delivered in a separate convenience figure ZIP. No figure or manuscript-generation program is required to
recompute the numerical results in this Data 1 package.

STRUCTURE
core_reproducibility/inputs: processed 2021 UK source input
core_reproducibility/causal_inputs: model, causal input generation and ledgers
core_reproducibility/confirmation: Australian processed input/source records
core_reproducibility/baselines: controllers and all saved results
core_reproducibility/analysis: analysis code and frozen statistical outputs
core_reproducibility/review: independent scientific checking code and records
core_reproducibility/upstream/provenance: historical upstream/timing evidence
robustness: archived sensitivity tables and relevant drivers
timing_provenance: interpretation and archive timestamp evidence
verification_20261006: documented scientific/portability checks and old manifests
publication_figures: current four figures, build script, source data and QA

SCOPE AND INTERPRETATION
The historical name confirmation means the 11-28 September 2026 Australian
external stress-test cohort, not pristine prospective blinding. UK data were
previously exposed development data. The results are conditional on the model,
fixed cohorts, synthetic obligation and declared timing assumptions. They do
not establish field safety, economic dominance or universal calibration.
Complete altered latency/resource grids and all controller optimizations were
not newly rerun at packaging time. See the scientific verification ledger.

CHANGES AND RIGHTS
RELEASE_CHANGES.json identifies every retained, moved, removed, changed and
added file relative to the preceding verified package. All original scientific and historical scripts, source, model, saved arrays
and result tables retain their previous bytes. Historical mixed analysis,
manuscript and figure construction scripts are retained for provenance; they
are not advertised as a turnkey or portable regeneration route. The current
portable numerical entry point and publication_figures are described above.
DATA_SOURCES_AND_RIGHTS.txt supplies source attribution and known rights.
No new open-source software licence has been invented for the authors' code.
