What was tested?
Identify the product, configuration, software version and test horizon.
RESEARCH / VALIDATION & LIMITS
Know what the evidence supports.
Engineering evidence is recorded for each product, configuration and version.
| Evidence | Result | Scope |
|---|---|---|
| Doctoral reproduction profile, golden D1–D5 | D1/D2 trajectory columns match the retained implementation; D3–D5 were re-baselined per approved correction. | Linux x86_64, Python 3.10.18, numpy 1.24.4, exact mode; not an exact reproduction of the 2026-07-18 retained trajectory. |
| Corrected default PSM energy identity | Closes per period in VALUE 101 one-day and two-year smoke cases. | Stress shortfalls remain reported; both profiles value model-recorded unserved energy at £17,000/MWh. |
| Stress events (both profiles) | Shortfalls are grouped into events while dispatch and prices remain unchanged. | Recorded from 0.7.0; deducted from energy served. |
| GBP1 doctoral reproduction, first model year | Raw invariants pass on the doctoral ledger boundary; annual results are published. | One model year; 487 stress periods, 78.8 GWh shortfall; ledger conformity is distinct from physical validation. |
| Local API security boundary | Cross-origin and sessionless requests are refused in source tests. | One user per computer; final installer checks remain separate. |
| Corrected wind and solar capacity factors | Displayed beside DUKES 6.3 load factors. | Before curtailment, representative GBP1 sites: onshore 1.56x, offshore 1.23x, solar 1.04x DUKES 2020–2024 averages; no calibration. |
| 0.7.0-alpha.1 installers | Linux: offline empty-directory installation, start, 48-period VALUE 101 and stop passed on 9 October 2026. Windows and macOS: not_evaluated. | Only actually tested platforms can be marked accepted. |
| VALUE four user paths | Reproduce from existing data, Add your new data, Edit a module and Add a new function to VALUE each passed end to end on the redesigned interface: 0 high, 0 medium and 1 low defect, none affecting model results. | Linux source tree, not an installer; VALUE 101-based cases from two periods to two full years. |
| GB zonal study | Longer-horizon scenario validation is planned for further research. | Network modules run only under the corrected profile in 0.7.0-alpha.1. |
Four questions for every study
What can be reproduced?
Check access to software, inputs, reference outputs and an appropriate environment.
What has been reviewed?
State whether the scientific interpretation has independent review. Assess it using the study assumptions and independent evidence.
What is public?
List the available code, input data, reference results and their access conditions.
Versioned methodology
Read the audited equations, numerical parameters, pseudocode and initial dataset identities. The complete methodology distinguishes implementation, tested examples and remaining reproducibility gaps.
Read the full methodology