Colombia 2026 Presidential Runoff Forecast — v2.0.0 implementation

Implementation version v2.0.0 — Colombia 2026 forecast The Colombia 2026 forecasting platform has reached implementation version v2.0.0, the deposit named in this project's existing description. This note announces v2.0.0 and indexes its components. Implementation versions registered with this OSF project: | Version | Git commit | Date | Zenodo DOI ||---------|--------------------------------------------|------------|-----------------------------|| v1.0.0 | e2c363c4574545171b609cd9fc2306bb2f0e6983 | 2026-06-07 | 10.5281/zenodo.20578542 || v2.0.0 | 20f2831a355e0fa3d70aa5eff21db83e0a78a11b | 2026-06-09 | 10.5281/zenodo.20618031 | Both versions implement the umbrella spec §16.1 first-round model + §16.2 runoff transfer-matrix model + §6 six-layer validation framework. The implementations differ in parameter values, data sources, and inference method; the model equation is identical. The runoff (2026-06-21) is the binding prediction for both versions. Forecast topline (eligible-weighted national, share of voters): | Quantity | v1.0.0 | v2.0.0 ||----------------------------------|----------:|------------:|| Cepeda voter share | 48.18 % | 47.57 % || Espriella voter share | 51.82 % | 52.43 % || Margin (Cep − Esp) | −3.64 pp | −4.85 pp || P(Cepeda wins) | 0.052 | 0.000 || 80 % credible interval width | 2.81 pp | 1.66 pp | v2.0.0 implementation summary: 1. Per-dept R1 conditioning — Runoff fit conditioned on per-dept actual Registraduría R1 results.2. Item C v3 mobilization prior — Per-dept signed Normal prior on R1→R2 mobilization, calibrated from joint 2018+2022 per-dept data.3. Calibrated transfer matrix — Per-eliminated-candidate Dirichlet concentrations derived from 2022 R1→R2 ecological inference (Fajardo 70/15/15, López 75/10/15, Valencia 5/80/15, κ=10).4. Persuasion churn σ = 0.025 — Derived from cross-dept variance of 2022 R1→R2 finalist share changes.5. Item G late-cycle shock — Student-t(ν=3) per umbrella spec §17.6. MRP frame v0.6.0-ops4 — Includes within-dept estrato rebalance from GEIH 2024-2025 EIH household microdata.7. Importance-sampling reweighting — Pre-runoff inference reweights prior-predictive draws against one observed interregnum head-to-head poll (Atlas Intel 2026-06-02, n=2030; IS ESS 83 %).8. Tier 2 runoff fit — Tier 1 + global drift, per spec §4.5.2 promotion criterion triggered by Item K backtest results.9. TLS certificate bundle — CNE-based scrapers validate against bundled Sectigo Public Server Authentication CA DV R36 intermediate certificate.10. Test coverage additions — PDF parser fixture regression tests (5 frozen PDFs × 3 firms); CNE walker max_pages parameter. Validation outputs (v2.0.0): - Layer 1 (NUTS convergence) — PASS: R-hat 1.0021, ESS bulk 3349.87, ESS tail 2814.08, divergences 0, E-BFMI 1.95.- Layer 2c (runoff Brier on 2018 R2 + 2022 R2) — Tier 1 Brier 0.137 / 0.159; both exceed 0.07 threshold; Tier 2 used for production.- Layer 3 (Simulation-Based Calibration) — KS p-value 0.84 (uniformity holds).- Layer 4 (single-ablation LOO) — Item C is most load-bearing (ΔMAE 0.029); other modules small but non-zero.- R1 calibration audit (model vs. actual, eligible-weighted) — Overall MAE 11.02 pp; 80 % CI coverage 41.2 %. Independent peer review: Gemini 2.5 Pro, 2026-06-09. Verdict: APPROVED-WITH-FIXES. All 5 findings closed (1 CRITICAL TLS bundling, 2 IMPORTANT [PDF fixture tests + CNE walker max-pages], 2 MINOR [Datexco citation + LAPOP imputation transparency]). Test suite: 1174 passed, 2 skipped, 0 failed. File integrity:- File: colombia_2026_mrp_v2.0.0.tar.gz- Size: 566,381,149 bytes (540.2 MB)- SHA-256: 34fe5c45ce7e408ee5dcc1f544e062c08f79ed2f0de2e249bb8fcdaff1d2dbc4- Files inside tarball: 45 Linked OSF pre-registration: https://osf.io/3ug2m Citation (pre-registered methodology):Santiago, S. (2026). A pre-registered Bayesian MRP framework for forecasting the 2026 Colombian presidential election. Zenodo. doi:10.5281/zenodo.20578542. Citation (this deposit):Santiago, S. (2026). Colombia 2026 presidential runoff forecast — v2.0.0. Zenodo. doi:10.5281/zenodo.20618031. Linked from osf.io/3ug2m. — Sigfredo Santiago, 2026-06-09