The method is public and the facility-resolved results are not. That is deliberate: a platform that hid its method would be asking to be trusted, and this one can be checked. Everything on this page is readable without an account.
Monte-Carlo ensemble mean
Every emission mass on the platform is the ensemble mean over 40,000 parameter draws, not the deterministic run at central parameters. Several loss terms are multiplicative and right-skewed, so the mean of the distribution exceeds the function of the means; nationally the two differ by about 15%. The deterministic run is reported as a diagnostic only.
29.8
IPCC AR6, fossil methane, 100-year basis
40,000
Every documented parameter range, propagated to a 95% interval.
1.0.0
Stamped into every stored result, so any figure names the code that produced it.
These are checks of internal consistency, not verification against direct field measurement. The distinction is the most over-read one in the whole product, so each gate states both sides of it.
15 % divergence on the field total · threshold 15
Different factors, software and analyst; no shared code.
0.2 % maximum relative error · threshold 0.2
17.2 % mean absolute error · threshold 25
oil 16.7%, gas 17.2%
0.979 R^2, out-of-sample · threshold 0.95
CI R^2 0.979, methane R^2 0.9998
Inputs, their provenance and quality
The twelve sources and every uncertain parameter
The tier of each source and Scope 1 coverage
Driver model, hold-out validation and bands
Anomaly, reconciliation and triage, with their guards
What it answers, what it refuses, how it is tested
What the figures do not establish