The diamond drill core shed is one of the last meaningful manual data-entry environments in modern mineral exploration. A geologist with a marker, a tape measure, and a laptop spends hours per shift describing lithology, alteration, structure, and mineralization on cores that may total tens to hundreds of meters per day. The output is a textual log that becomes the primary geological dataset for everything downstream — resource estimation, deposit modeling, technical reporting. The bottleneck is the geologist's time, and the variability between geologists is a chronic source of inconsistency that compounds across long programs.
Automated core logging — using high-resolution imagery of the core and machine learning models trained to classify lithology, alteration, mineralization, and structural features — has been emerging for about a decade and reached operational maturity in the early 2020s. By 2026, several firms operate this as a service at industrial scale, and a few major mining companies have integrated automated logging into their standard workflow. The technology is real, the value proposition is concrete, and the limitations are specific enough to plan around.
What the Technology Actually Does
The core workflow is straightforward. Drill core is scanned through a high-resolution imaging system that captures consistent-illumination photography of the full length, often supplemented with structured-light scanning to capture surface topography. Some systems add hyperspectral SWIR scanning to capture mineralogy data, and a few add fluorescence or LIBS for elemental data. The raw outputs are a registered set of imagery, depth-tagged and ready for ML processing.
The ML models, typically CNNs trained on labeled core imagery, then classify the core at the centimeter-to-meter scale. The standard outputs include lithology classification, alteration intensity and type, structural features (foliations, veins, fractures), and mineralogy where hyperspectral data is included. The model output is reviewed by a geologist — usually faster than logging from scratch — and the validated log becomes the project's core record.
The leading commercial providers — Datarock (Australia), GeologicAI (Canada), Corescan (Australia, hyperspectral focus), Minalyze (Sweden, XRF and visual integration) — each offer somewhat different configurations of imagery and analytical capability. The competitive landscape is moving fast, and any specific recommendation here will be partially outdated within a year. What matters for a junior is the general capability category and the fact that there are now multiple competent providers, not the specific market positions of any one of them.
Where It Earns Its Keep
Three distinct value propositions show up in production use. First, throughput. A geologist logging core manually produces something in the range of 30 to 100 meters of logged core per day depending on geological complexity. An automated system processes hundreds to thousands of meters per day, with the geologist's role becoming review and validation rather than first-pass logging. For a 10,000-meter program, this compresses the total logging time from quarters to weeks.
Second, consistency. Manual logging has irreducible inter-geologist variation. Two competent geologists describing the same core will produce logs that disagree on lithology contacts, alteration intensity, and structural features, especially when the geologists rotate over a multi-month program. Automated logging applies the same model to all the core, eliminating inter-geologist variation. The model has its own biases, but those biases are consistent across the dataset, which makes the data more usable for resource estimation downstream.
Third, structured output. Manual logs are textual or coded, and converting them to the structured format that resource estimation software needs is itself a workflow with its own error sources. Automated logs are structured from the start — every meter of core has a quantitative classification, with confidence scores, ready for direct ingestion into the project database. The downstream data processing burden drops substantially.
What the Models Are Actually Good At
Different aspects of core logging are different ML problems with different success rates. Lithology classification at the broad-class level — sediment vs. intrusive vs. extrusive, broad alteration types — is highly accurate, often comparable to or better than a competent contract geologist's logging. Specific lithology discrimination within a class — distinguishing similar intrusive phases, fine-grained sediments, hydrothermally altered rocks — is harder and depends on the training data being well matched to the project's geology.
Structural feature detection from imagery is mature for the obvious things — veins, fractures, lithological contacts — and more limited for subtle features like foliations or relict primary structures. The standard approach is to combine imagery-based detection with surface-topology data from structured-light scanning, which captures fracture orientations more reliably than imagery alone.
Alteration mineralogy from hyperspectral SWIR data is mature for the alteration mineral suite that has diagnostic SWIR absorption features — clay minerals, micas, chlorites, carbonates, some sulfates. Iron oxides, silica polymorphs, and minerals without SWIR features are not well characterized by this approach and require complementary analytical methods if their abundance matters.
Sulfide and oxide mineralogy at the micrometer scale — distinguishing chalcopyrite from bornite, identifying specific sulfosalts — is beyond what optical imagery can do reliably. For projects where this discrimination matters, automated logging is the first pass and electron microprobe or QEMSCAN-style analysis on selected intervals is the follow-up. The two are complementary rather than competing.
Where the Limitations Bite
The honest limitations are worth knowing before committing to a program. The training data for any specific automated logging service tends to skew toward the deposit types and geological settings the provider has historically worked on. For a porphyry copper project in a province where the provider has dozens of past projects, the model performs well. For an unusual deposit type or an unconventional setting, the model may need substantial retraining on project-specific labeled data, which adds cost and timeline to early implementation.
Edge cases — broken core, missing intervals, RQD-zero zones, post-drilling alteration during storage — confuse models that haven't been trained on representative examples. The geologist's review step catches most of these, but the automation efficiency degrades when the core quality is poor. Programs with consistently good core recovery benefit most; programs with chronic recovery issues see less time savings.
Calibration to a project's specific geology takes time. The first few hundred meters of automated logging on a new project often require iterative model refinement, with the geologist correcting outputs and the provider retraining or fine-tuning the model. This calibration phase adds setup cost and means the value proposition is strongest for larger programs that amortize the calibration across many meters of core.
And finally, automated logging doesn't replace the geologist's role in deposit interpretation. The logged data is an input to interpretation, not the interpretation itself. Decisions about deposit-type model, structural framework, ore controls, and exploration vectoring still come from geologists synthesizing the logged data with everything else they know about the project. The automation reduces the labor cost of producing the data; it doesn't compress the interpretive work.
Integration with Downstream Workflows
The value of automated logging is highest when its output flows cleanly into the project's downstream systems. A logged dataset that lives in the provider's proprietary format and requires manual export to the drillhole database is partially wasted; one that ingests directly into the project's database, becomes immediately available for resource estimation, and updates with each new batch of scanned core is fully realized.
The major providers have improved their integration substantially over the past few years. APIs and structured data exports to common drillhole database formats (acQuire, Maxwell GeoServices, internal SQL databases) are now standard. For projects using Leapfrog Geo for modeling, direct ingestion of automated logging outputs into Leapfrog through the standard import paths works reliably. The integration friction is meaningfully lower than it was in 2020, and continues to improve.
For a junior committing to automated logging, the right pattern is to specify the database integration requirements upfront as part of the contract scope. A provider that delivers structured data in the exact format the project database expects is more valuable than one that delivers higher-accuracy logs in a format that requires manual conversion.
Cost and Procurement
Automated core logging contracts run roughly $20 to $50 per meter of core for the standard imagery-plus-classification offering, with hyperspectral mineralogy adding another $15 to $40 per meter depending on the provider. For a 5,000-meter program, the total cost is on the order of $100,000 to $400,000. Compared against the avoided contract logger time (typically $400-$800 per day for experienced contractors, producing 30-100 meters per day, so $5 to $25 per meter of logging cost), the automated logging is more expensive per meter but produces a more uniform and more structured dataset, plus much faster turnaround.
The cost calculus favors automated logging more strongly for: larger programs that amortize setup costs, programs where contract logger availability is a constraint, programs where consistency matters for resource estimation, and projects in jurisdictions where geologist labor costs are high. It favors manual logging more strongly for: small one-off programs, programs with unusual deposit types requiring substantial model calibration, and projects where the geologist's interpretive engagement during logging is itself a strategic value.
The procurement decision is also a multi-year decision. Switching providers mid-program is expensive — the second provider has to recalibrate to the project's geology and the dataset gets a methodology break. Picking the right provider for the project's life expectancy is more important than negotiating the lowest per-meter price.
A Reasonable Pilot
If your project is actively drilling and has never used automated logging, the right first step is a pilot on one or two completed holes. Most providers will price pilots at modest discounts to give projects a chance to evaluate the deliverable on their specific geology. The pilot output gets compared against the manual logs, the discrepancies get reviewed by the project geologist, and the team gets a concrete basis for deciding whether to scale to the full program.
The pilot also surfaces integration questions early: how the data flows from provider to database, how the QP signs off on automated vs. manual logs, how the model handles project-specific geology. Those questions get answered cheaply on a pilot and expensively on a committed full-program scale.
For projects that are not actively drilling but anticipate a future program, the right move is to keep automated logging in the budget plan but defer commitment until the program is funded and scheduled. The capability is mature enough that a pilot can be procured on relatively short notice when needed.
For an outside view on whether automated core logging would pay back on your specific project, our free workflow audit covers exploration data workflows, or contact us to discuss how to scope a pilot.