BRENNEN

Industries / Energy, water and waste / Oil and gas extraction / Oilfield services

Measuring the well

Measuring the well is the work of lowering instruments into a well, or carrying them behind the drill bit, to record the rock's properties against depth so that geologists can tell oil, gas and water apart.

6

Milestones achieved, from 1931 to 2020.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1931

    Telling porous rock apart

    After the first electrical log in 1927, Schlumberger's crews added the spontaneous potential (SP) log in 1931. It measures the small natural voltage that builds up in the borehole mud at the edges of permeable beds, rock that fluids can flow through, so it separates rock that can hold a reservoir from rock that cannot. A recorder introduced in 1930 let an engineer log 1,000 feet an hour, and the number of Schlumberger logging teams grew from 8 in 1933 to more than 140 just before the Second World War, by the company's count.

    slb.com

    Credit

    Conrad and Marcel Schlumberger's company (now SLB)

  2. 1942

    An equation for oil saturation

    Gus Archie of Shell Oil published in Transactions of the AIME a relationship between a rock's electrical resistivity, its porosity (the share of its volume that is open pore space) and how much of that pore space holds salt water rather than oil or gas. SLB's history calls "Archie's Law" the foundation of petrophysical log interpretation, the work of turning logs into estimates of how much oil a rock holds.

    doi.orgslb.com

    Credit

    Gus Archie, Shell Oil

  3. 1986

    Pictures of the borehole wall

    Schlumberger introduced its Formation MicroScanner in 1986 and the FMI fullbore formation microimager in 1991, tools that map small differences in resistivity around the borehole wall into what SLB describes as "seemingly photographic images." Fractures and rock layers that cut the round hole appear as sine-shaped curves when the image is unrolled flat, so geologists can measure their depth, dip and direction.

    slb.comslb.comscitepress.org

    Credit

    Schlumberger (now SLB)

  4. 1987

    Lost sources stay buried

    Density and neutron logs estimate porosity by sending gamma rays or neutrons from a small radioactive source into the rock; SLB dates its first density log to 1957. The US Nuclear Regulatory Commission's well-logging rules (10 CFR Part 39), first issued on 17 March 1987 and amended since, say that a source stuck in a well and judged irretrievable must be sealed in place with a cement plug, protected against later drilling, and marked at the surface by a metal plaque giving the source, its depth and a warning such as "DO NOT RE-ENTER THIS WELL," with a written report to the regulator within 30 days.

    Lessona measuring tool can leave a hazard that outlasts the well, so the warning has to travel with the well's records.

    law.cornell.edulaw.cornell.eduslb.com

    Credit

    US Nuclear Regulatory Commission

  5. 1989

    Logs taken while drilling

    Measurement while drilling sends the hole's direction to the surface as the well is drilled; logging while drilling (LWD) adds rock measurements from sensors just behind the bit. SLB's history says its first LWD tool came in 1989, measuring the rock before drilling fluid had time to soak into it and spoil the reading, and that it used geosteering for the first time in 1991, adjusting the path of horizontal wells from those readings to keep them inside the oil-bearing layer.

    slb.comslb.com

    Credit

    Schlumberger (now SLB)

  6. 2020

    An open lithology contest

    FORCE, a Norwegian industry forum, ran a contest to predict lithology (the type of rock) from the logs of 118 wells offshore Norway, with labels drawn by hand by geoscientists. 329 teams signed up and 148 submitted predictions; the organizers then re-ran the top teams' code on 10 hidden wells. The leaderboard leader finished 11th on the hidden wells, while the winner had worked to avoid overfitting (tuning a model so closely to known data that it fails on new data). The logs and labels are free under a CC BY 4.0 license.

    Lessona public leaderboard can reward overfitting, so only wells a model has never seen show how it will do.

    github.comzenodo.orgsodir.no

    Credit

    FORCE; organizers P. Bormann, P. Aursand, F. Dilib, P. Dischington and S. Manral

Watch for

  1. Logging without a cable

    SLB says its OnWave platform logs a well with no wireline cable, no wireline unit and no SLB crew on site, running measurements and data-quality checks on its own downhole while keeping in contact with the surface. It says the platform deploys in less than half the time of conventional wireline, and that in South Texas it cut the time to reach the bottom of a well from hours to 27 minutes, a 70% reduction.

    Achieved whenan operator or another outside party publishes OnWave logs beside conventional wireline logs from the same wells, with measured differences in data quality and time.

    slb.com

    Claimed by

    SLB, July 2025

  2. Promptable AI for well logs

    TGS describes a 60-million-parameter AI model trained on 1.1 million North American well logs that fills in missing log curves and, when a geologist types a formation name, picks that formation's top in a well. On 8,813 Permian Basin wells withheld from training, its median errors for six key formations ranged from 13.0 ft to 40.0 ft, by TGS's own test.

    Achieved whena party outside TGS scores the model on wells from outside TGS's collection and publishes the errors.

    tgs.comitwm.fraunhofer.de

    Claimed by

    TGS (Ben Lasscock, Altay Sansal, Keyla Gonzalez and Alejandro Valenciano), in a paper for the IMAGE 2025 conference, held in Houston on 25–28 August 2025

Open

  1. 01

    Open test for fracture picks

    A public benchmark that scores software for finding fractures in borehole image logs against human picks, built from the image logs and fracture interpretations that the Utah FORGE geothermal project has released.

    The brief
    Credit · unclaimed
  2. 02

    Blind test for log AI

    A fresh, hidden test set of public wells, labeled for lithology by two geologists and kept private, with a scoring page to which any company or researcher can submit a model, including the new well-log foundation models (large AI models trained first on raw logs and then adapted to tasks).

    The brief
    Credit · unclaimed

Open 01 · the brief

Open test for fracture picks

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public benchmark that scores software for finding fractures in borehole image logs against human picks, built from the image logs and fracture interpretations that the Utah FORGE geothermal project has released.

Why it is still open

The authors of a 2023 deep-learning fracture detector, from VTT in Finland and the company Astrock, trained on a company's own images and wrote that "there are no open-datasets available to benchmark our method." Open-source detectors posted on GitHub in 2026 ship code without labeled images; one, Borehole-Fracture-Analysis (MIT license), says its 300 training samples and model weights were left out because they carry no license to share.

scitepress.orggithub.com

Built with what exists

Utah FORGE well 16A(78)-32's FMI logs, SLB's original interpretation (as appendix graphics) and a reinterpreted spreadsheet of fracture picks with depth, dip, dip direction and type, released under CC BY 4.0; vision models; open detectors such as Borehole-Fracture-Analysis and DeepLogger as methods to score; a classic curve-fitting baseline; Python.

catalog.data.govgithub.com

  1. First step

    Download the 16A(78)-32 image log and both interpretations, and count how many of the reinterpreted fractures match SLB's original picks.

  2. Done looks like

    A benchmark covering at least two wells, with agreement between the two human interpretations measured first under a stated matching rule (for example, depth within 0.3 m and dip within 10 degrees), and at least three detection methods scored under the same rule, with code and scores published.

Open 02 · the brief

Blind test for log AI

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A fresh, hidden test set of public wells, labeled for lithology by two geologists and kept private, with a scoring page to which any company or researcher can submit a model, including the new well-log foundation models (large AI models trained first on raw logs and then adapted to tasks).

Why it is still open

FORCE 2020's hidden-test file for its 10 blind wells, labels included, is now posted in the competition's public repository, so that set no longer tests any model blind, and its organizers called a 10-well blind test "so small". TGS reports its model's accuracy only on wells withheld from its own collection. New public collections add raw data but no hidden labels: the Open Well Data catalogue lists 507 public sources of well data in 79 countries and posts standardized copies of many, and Georgia Tech's OpenSeisML, released in May 2026, gathers UK seismic and well data for generative AI.

github.comsodir.notgs.comzenodo.orgarxiv.org

Built with what exists

Norwegian well logs released under the Norwegian Licence for Open Government Data, which FORCE 2020 used; standardized logs from the Open Well Data collections; FORCE's penalty matrix and scoring code; the 2020 winners' published code as baselines; two geologists; a hosted scoring page.

zenodo.org

  1. First step

    List released Norwegian wells with full log suites that are not in the FORCE set, and have two geologists label the same well to measure how often they agree.

  2. Done looks like

    At least 20 wells not among FORCE's 118 labeled by two geologists with their agreement reported, and at least five models scored with the FORCE metric, including one foundation model or commercial tool, with results published.