Devon Energy and HXMX will present joint research at IMAGE 2026 showing that an automated interpretation platform can pick formation tops as reliably as expert geoscientists across a large Williston Basin dataset. The study also introduces a new method for measuring how far a single reference well can carry an accurate correlation. The presentation is scheduled for August 19, 2026, at 8:00 a.m. in Room 372A of the George R. Brown Convention Center in Houston.

Authored by Drew Kreman and Kevin Pelton of Devon Energy and Alan Lindsey of HXMX, the study benchmarks HXMX’s automated formation-top picking against traditional manual correlation. The team calibrated the platform on a multi-well calibration set with interpreter-verified tops, then deployed it on an extensive multi-county test set. Automated picks matched expert picks with strong accuracy across most intervals.

At the center of the work is Triangulated Cross-Validation (TCV), a method that checks each automated pick against every other well in a set and reports consistency in feet. Across both large and tightly spaced study areas, key formations picked with sub-foot spreads from a single reference well. TCV gives geoscientists a direct, quantitative measure of uncertainty — and defines the practical range over which a verified reference well can be trusted to guide automated picking.

Manual well-log correlation is accurate but slow, and it scales poorly across the thousands of wells required for basin-wide analysis. The HXMX platform lets geoscientists calibrate the system on a small set of verified wells, then extend consistent picks across a basin while staying in the role of reviewer and decision-maker. Because the approach uses pattern recognition anchored to expert-verified wells rather than a black-box model, every pick carries a measurable confidence.

“Geoscientists typically pick on cross sections and loop-tie back to verify that their picks have stayed consistent,” said Alan Lindsey, CEO of HXMX and a co-author of the study. “This work shows we can expand that process dramatically. We start with a geoscientist’s reference picks. Instead of checking only a handful of loop ties, every well can be checked against every other well in the area, homing in on the best pick for each marker and measuring uncertainty in feet. The geoscientist stays in control, and the platform handles the repetitive work.”

“Testing this across a multi-county dataset gave us a quantitative basis for evaluating where automated correlation is performing well and where additional review is warranted.

Expressing uncertainty directly in feet better represents the geological reality of top prediction error — it’s easy to generate, simple to track across multiple tops, and makes it straightforward to flag wells that need a closer look,” said Drew Kreman, Devon Energy.

IMAGE, organized by the Society of Exploration Geophysicists, the American Association of Petroleum Geologists, and the Society for Sedimentary Geology, is the largest annual gathering in applied geoscience and energy, drawing thousands of professionals to Houston each August. AI and scalable interpretation are central themes of the 2026 program.

The full abstract is available through the IMAGE 2026 technical program.

About HXMX

HXMX is transforming subsurface interpretation through AI-powered signal intelligence. The platform captures expert geoscience judgment and applies it at basin scale, helping operators interpret more wells faster while providing a measurable confidence metric for every pick. The same technology can be applied wherever expert pattern recognition needs to be scaled across large datasets. Learn more at hxmx.ai.

Media gallery

About The Author
Business Wire