Benchmarks
Ten fixtures check the engine against exact answers, an independent NumPy reference and published studies. Every verdict carries its evidence.
Running T1-FIXED in your browser, 1 of 10.
PassFailDisagrees with publication
| Fixture | Checks | Expected | NumPy reference | Engine | Band | Verdict |
|---|---|---|---|---|---|---|
Every input fixed at the original means MMSTB, 1,000 runs | P50 208.4 Mean 208.4 | 208.4 208.4 | running | 1e−9 relative 1e−9 relative | expected pass | |
Lognormal product with an exact answer MMSTB, 200,000 runs | P90 130.6 P50 197.7 P10 299.4 Mean 208.4 | 130.4 197.7 299.3 208.4 | waiting | 99.9% interval 99.9% interval 99.9% interval 99.9% interval | expected pass | |
Stanko's all-uniform example MMSTB, 200,000 runs | Mean 311.9 | 311.9 | waiting | 99.9% interval | expected pass | |
Demo case rank correlations rank correlation, 100,000 runs | Porosity with Sw −0.500 NTG with porosity 0.400 NTG with Sw −0.300 | −0.500 0.399 −0.301 | waiting | ±0.01 ±0.01 ±0.01 | expected pass | |
Original app inputs, limits applied MMSTB, 200,000 runs | P90 114.8 P50 196.6 P10 322.1 | 112.9 197.2 318.1 | waiting | ±3.0% ±3.0% ±3.0% | expected pass | |
Demo case against the NumPy reference MMSTB, 200,000 runs | P90 105.0 P50 196.9 P10 347.5 | 105.0 196.9 347.5 | waiting | ±1.0% ±1.0% ±1.0% | expected pass | |
Condensate in place, triangular FVF million Sm³, 200,000 runs | P90 8.33 P50 9.94 P10 17.7 | 8.29 10.2 13.1 | waiting | ±3.0% ±3.0% ±3.0% | expected disagrees with publication | |
Condensate in place, fixed FVF million Sm³, 200,000 runs | P90 9.09 P50 9.93 P10 12.2 | 9.03 9.93 10.9 | waiting | ±3.0% ±3.0% ±3.0% | expected disagrees with publication | |
Oil in place MMbbl, 200,000 runs | Mean 74.4 P90 18.4 P10 150.5 | 74.7 45.1 107.1 | waiting | ±1.0% ±3.0% ±3.0% | expected disagrees with publication | |
Oil in place, in tonnes Mt, 200,000 runs | P90 2.00 P50 3.73 P10 6.98 | 1.92 3.81 6.80 | waiting | ±5.0% ±5.0% ±5.0% | expected pass |
T4-THANDER: Oil in place
Thander, Sircar and Karmakar (2014), IJLTEMAS 3(4). Values in MMbbl, 200,000 runs, seed 8, Monte Carlo.
Every input is triangular, as published, except net-to-gross: the paper has no net-to-gross term, so it is fixed at 1.
The bands allow for the paper's 10,000 runs: ±1% on the mean and ±3% on percentiles.
Evidence for the disagreement
The paper reports a minimum of −90.96 MMbbl, which strictly positive inputs cannot produce, and a standard deviation of 55.24 MMbbl against 23.9 from both implementations.