Reproducibility snapshot v2.0.0 for "The Three Lock-In Modes — A Substation-Catchment Classifier of Grid Maladaptation Mode across 39 OECD Countries"

Second versioned deposit accompanying the v22 P6-ERIS manuscript submitted to Environmental Research: Infrastructure and Sustainability (IOP Publishing). This is a New Version on concept DOI 10.5281/zenodo.21827667. The v1.0.0 deposit (6 August 2026) accompanied the ERIS-100994 submission that was returned without external review on scientific-reporting-standards and reference-comprehensiveness vectors. The v2.0.0 deposit delivers a substantially strengthened manuscript following a demanding push cycle: 6 audit rounds + publishability rewrite + real empirical execution on the SSI Index v4.23 canonical + language strengthening + academic register sweep + three-country cross-country identification. The paper delivers three contributions on a per-substation-catchment classifier of grid maladaptation mode across the OECD grid. Contribution 1 (Framework). The three-mode classifier {Renewal-pipeline, Siting, Topology} operates as a decision layer over the shared SSI Index v4.23 ridge estimator via softmax weight-vector projection with hard-constraint eligibility filter e(i) and bootstrap-stability score s(i). Four-path cross-consistency envelope — Path A composition-averaged Byers-parity baseline, Path B Gaussian-copula rank-transform, Path C ridge headline (α_opt = 92.4, R² = 0.851 at Italian pilot depth), Path C̃ ERA5-Land direct predictor design (α_opt = 0.001, R² = 0.180, ‖β̂_ERA5‖₂ = 0.104 at n = 47,906) — anchors the identified envelope. Statistical rigor stack: Kish two-tier country-in-region clustering K_eff ≈ 6, Cameron-Gelbach-Miller cluster-robust SE, Benjamini-Yekutieli FDR positive-dependence variant at q ≤ 0.05, SALib Saltelli-Sobol variance decomposition, Gelman-Hill partial-pooling Bayesian via PyMC v5 with Half-Cauchy(0, 2.5) prior under NUTS, 10,000-replication Gaussian bootstrap, MacKinnon-Webb wild-cluster bootstrap-t, 5,000-permutation independence check, Arkhangelsky synthetic-DiD placebo. Sobol variance decomposition on classifier p(Siting) versus R_median composite demonstrates empirically-measurable net-new information beyond the shared ridge: R8_adapt +0.580, R10 +0.120, R6d +0.129 (differential total-order sensitivity index) against R6c_flood dominance in scalar R_median (S_T = 0.722). Contribution 2 (Cross-country identification within-country cross-validation across three-country pilot). At Italian pilot depth of 47,906 catchments across 20 Regioni, held-out cross-validation (Train Northern + Central 30,974 → Test Southern + Islands 16,558) delivers 99.29% classification accuracy. UK pilot at 62,409 catchments across 12 ITL-1 regions with Ofgem RIIO regulatory anchor: held-out cross-validation (Train Southern 32,144 → Test Northern + Celtic 26,155) delivers 91.93%. Japan pilot at 7,073 catchments across 8 macro-regions with METI + OCCTO regulator anchors (Tier-2 register-access): held-out cross-validation (Train Honshu → Test Peripheral) delivers 97.84%. Three-country consistency verdict is strong: R8_adapt top-1 Sobol S_T concordant across all three countries (Italy 0.624, UK 0.602, Japan 0.527); Spearman rank correlation on seven shared modifiers is positive across all pairs (Italy-UK 0.643, Italy-Japan 0.750, UK-Japan 0.821). Country-specific risk architecture surfaces through rank-2 modifier variation: Italy R10-just, UK R10-just, Japan R6-seismic (rank 2, S_T = 0.259) — empirical evidence the classifier detects country-specific risk architecture without losing cross-country consistency on the identifying R8_adapt anchor. BY-FDR against per-country cohort mean surfaces substantive-magnitude outliers: Italy Calabria (RP +12.2 pp) + Basilicata (Topology +6.6 pp); UK Wales (RP +10.4 pp) + Scotland (Siting +11.3 pp) + Northern Ireland (RP +8.5 pp); Japan Kanto (TEPCO metropolitan compound signature) + Kyushu (Siting +10.3 pp) + Chugoku (Siting +9.4 pp). The May 2023 Emilia-Romagna flood serves as illustrative case study of the classifier-cascade nexus; the cascade-coefficient identification anchor is attributed to companion Bérard 2026 (Environmental Research: Energy, under review) at synthetic-DiD placebo p = 0.012. Contribution 3 (Register-access-readiness audit protocol). A 4-tier generalisable procedural framework classifies 40 OECD countries plus Greenland via three coded data classes (RAB depreciation register, TYNDP siting outcomes, TYNDP topology graphs) into Tier 1 (n = 25, ENTSO-E members plus Norway plus Switzerland plus Australia), Tier 2 (n = 7, Chile, Greece, Israel, Japan, Korea, New Zealand, US), Tier 3 (n = 7, Canada, Colombia, Costa Rica, Greenland, Iceland, Mexico, Turkey). Three NUTS-3-proxy placebo p-values (Sardegna 2023, Calabria 2024, Iberian April 2025) illustrate the protocol's diagnostic value at Tier-3 register-access scale (all three p-values indistinguishable from null). The snapshot represents the full discipline arc from ERIS-100994 desk-reject through 22 manuscript versions: diagnostic memo, three-contribution architecture design, four-pass rewrite (v13), six audit rounds (v14-v17), publishability rewrite (v18), real Italian Phase 1 execution on SSI Index v4.23 canonical (v19), empirical corrections (Path C̃ ERA5 landed, BY-FDR substantive-baseline reframe, regional-heterogeneity narrative empirically corrected — v20), UK second-country anchor (v21), academic register sweep (v21 to v22), Japan third-country anchor (v22), Pascal compression to ERIS ceiling. Frozen against SSI Index v4.23 canonical (15 July 2026 TERMINAL snapshot). Underlying per-substation payload builds on GHSL 5 km catchment population, Niva 2023 20-year sum migration score, polygon spatial-join socio-economic backfill across 25 countries via Eurostat GISCO NUTS-3 2024 and GADM 4.1 admin1, per-country min-max semantic normalisation of migration_score into [0, 1] envelope preserving per-substation ranking, and per-substation ERA5-Land climate trajectory {I₁, I₂, I₃} resident in the canonical. Reproducibility ladder: Layer 1 pipeline scripts at deterministic κ = 1.0 (this Zenodo deposit); Layer 2 SHA-256-locked results-JSON manifest (SHA256SUMS.txt inside archive); Layer 3 institutional registers (ARERA + Terna for Italy, Ofgem + National Grid ET for UK, METI + OCCTO for Japan, ENTSO-E + JRC EFAS + Copernicus EMS + INGV + EFFIS + CMIP6 ISIMIP3 for cohort-wide anchors). CC BY 4.0 (data + manuscript) + MIT (code); Docker + Python 3.10.9. Visibly-honest degradation preserved throughout: (1) Val d'Aosta n=58 fleet-insufficient marker; (2) Convention #56 residual buckets for unmapped county-to-ITL-1 (UK 4,110 subs = 6.6%) and unmapped prefecture-to-macro-region (Japan 16 subs = 0.23%) surfaced honestly; (3) full-cohort HDI on cross-country Bayesian pooling crosses zero (μ_cohort = -0.682, HDI95 [-1.707, +0.310]) — reduced-cohort excluding four †-flagged R7_cyber-drift outliers (Canada, Finland, Mexico, Turkey) at μ_reduced = +0.421 also with envelope crossing zero; consistency with Contribution 3's register-access-readiness taxonomy interpretation; (4) Emilia-Romagna DiD cascade-coefficient identification honestly attributed to companion paper