Interactive Causal Sensitivity Studio
Adjust empirical assumptions live to observe dynamic counterfactual re-weighting, buffer burn, and backing ratio.
Developer Baseline Multiplier:1.00x
0.50x (Severe Deflation)1.00x (Locked Verra Benchmark)2.00x (Extreme Inflation)
150 km Spatial Durbin Leakage Rate:5.13%
0.0% (Zero Leakage)5.13% (Empirical Spatial Durbin)15.0% (Max Displacement)
Simulated Claimed VCUs
30,442,786
Simulated Verified VCUs
11,411,775
Over-Crediting Discovered
19,031,011 (62.51%)
Resulting Backing Ratio
3748 bps (37.48%)
24-Year Annual Panel Trajectory (2001–2024)Causal Synthetic Counterfactual vs Developer PDD Baseline
Shaded windows show Pre-Intervention Training (2001–2006), Out-of-Sample Holdout (2007–2010), and Crediting (2011–2024).
2001–2006: Pre-Intervention Training (Simplex Fit)
2007–2010: Out-of-Sample Holdout (Zero Treatment)
2011–2024: Post-Intervention Crediting Window
Out-of-Sample Holdout Scorecard (2007–2010)RMSPE & MAE
Evaluated strictly on pre-intervention holdout years unseen during donor weight optimization.
| Model Specification | RMSPE | MAE | OOS Error Reduction |
|---|---|---|---|
CarbonProof SCM Simplex | 0.3062 kha | 0.2770 kha | -92.55% [95% CI] |
Unconstrained OLS | 0.3280 kha | 0.2795 kha | -92.0% |
Unweighted Donor Mean | 2.4074 kha | 2.4030 kha | -41.3% |
Static PDD Linear | 4.1004 kha | 3.9673 kha | Baseline (0%) |
200-Bootstrap Resample Validation:
CarbonProof SCM achieves a mean holdout error reduction of 92.55% with 95% confidence interval [90.21%, 95.07%] relative to developer static PDD extrapolation.
30 In-Space Placebo Permutations
Post/Pre RMSPE Ratio test iteratively reassigning intervention to each donor polygon.
p = 0.03226
1/31 ≤ 0.05
Treated Project Post/Pre RMSPE Ratio:23.2757x
Max Donor Placebo Ratio (False Positive):3.4737x
Empirical p-value:(1 + 0) / (30 + 1) = 0.03226 (Statistically Significant)