Research data for The Chemistry Performance Envelope

Run from this directory with Python 3 and the packages in requirements.txt:
  python reproduce_energy.py
  python verify_results.py
  python plot_figures.py

The scripts run offline. They do not contact a quantum provider or submit jobs.
Qiskit is optional for the independent statevector comparison; NumPy is required.
The plotting script writes regenerated figures to assets/.

PROVENANCE.json lists every selected source record, source hash and extraction rule. Extracted files are new publication records, not unmodified provider payloads. MANIFEST.json hashes the release files. Neither type of hash authenticates a provider or establishes historical timing.

The six-qubit gate sequence and Pauli operator permit direct replay of this supplied mathematical instance. Result and calibration records retain relevant scientific metadata. All 21 campaign observations and their predictions are included, including unfavorable outcomes. Repeat inputs remain partial. Benchmark and fit summaries reproduce stored summaries, not every original optimization or bootstrap run.

Conditional budgets define C=c0-E0 and D=C-delta. Limits are illustrative model-coordinate values with a=k=1, b=u=0 and target 1.6 mHa; no measured chip specification is asserted. H2 uses a stored illustrative contrast. Other contrast summaries establish their trace-to-ground convention, but are not a fresh molecular-structure validation.

Original full-width hardware inputs, complete repeat payloads and per-state sector probabilities are not available in this release. Unrelated molecular-candidate identity metadata, company/account details and internal editorial records are excluded. Selection does not remove adverse data supporting the paper's conclusions.

The seven-target prediction record omits the 1433-gate target. Its original header's pre-execution assertion is not supported for that job: the local job record describes July 9, 2026 execution, while the model record was dated August 13. The comparison is retrospective. A later retrieval timestamp alone cannot establish that an outcome was unseen.
