Measured, not claimed
How often Retsil is right
Every well in the recorded inventory was re-scored as if it had never been drilled: the engine saw only the wells more than 500 m away, produced a probability, and was then graded against what that well actually did. 19,579 wells were tested — 18,881 producers and 698 dry holes.
Hit rate on top-band calls
97.0%
19,225 wells scored 80%+
Ranking power (AUC)
0.776
0.5 is a coin flip, 1.0 is perfect
Brier score
0.032
Lower is better; 0 is perfect
Hit rate by score band
This is what a Retsil number means in practice. Read your score, read the row.
| Retsil score | Wells tested | Actually produced |
|---|---|---|
| 20–40% | 27 | 51.9% |
| 40–60% | 37 | 51.4% |
| 60–80% | 290 | 70.3% |
| 80% and above | 19,225 | 97.0% |
What this does not prove
- • Public well inventories under-report dry holes, so the raw success rate of the tested set (96.4%) is higher than the true field-wide rate. Scores are shrunk toward a conservative prior for that reason, and the top-band figure should be read as an upper bound.
- • The test measures how well the engine ranks ground from surrounding drilling history. It does not measure trap geometry, seal integrity, or reservoir quality at your exact location — only seismic, wireline logs and cores measure those.
- • Frontier ground with no offset wells is excluded from the test and capped at 60% in the product, because there is nothing to grade it against.
- • A screening score is not a guarantee of production and is not a substitute for review by a licensed petroleum geologist.
Back-test last recomputed 2026-09-05.