Drill Hole Logs and Assay Data for Mineral Exploration AI

What a usable drill hole dataset contains

A usable drill hole dataset is four or five tables joined on a hole ID and a from/to depth interval, and anything less limits which models you can train. Collars give the hole ID, easting, northing, elevation, coordinate reference system (CRS), total depth, drill type (diamond core, reverse circulation, rotary air blast) and date. Downhole surveys give depth, azimuth and dip readings, which you need to desurvey the hole into 3D; without them, an inclined hole is plotted as if vertical and every interval sits in the wrong place.

Interval logs record what a geologist saw over each from/to range: lithology codes, alteration type and intensity, mineralization style, vein percentage, weathering, recovery and RQD, plus free-text comments. Assays record sample ID, from/to, lab, method (for example fire assay or four-acid digest with ICP), element values, units and detection limits. Research on drillhole interpretation depends on exactly this pairing: classifiers trained on geological logging and assay data to suggest interpretations [2], and models that predict material type and chemistry from measure-while-drilling (MWD) signals labeled with logged material and assays at 0.1 m depth segments [1].

For broader context on the industry category, see the industry-specific operational data hub and the AI data master hub.

Which models each table supports

Each modeling task needs a different minimum set of tables, so scope your request by task rather than by "all the data." The table below maps common exploration and mining AI tasks to the inputs and labels they require.

Rock-engineering work shows the scale drilling data can reach: one tunnel study derived 23,277 labeled samples from about 500,000 drillholes in 15 hard rock tunnels [4]. That is MWD from production drilling, though, and it does not replace sparse, expensive exploration core with assays. If your target is construction-ground rather than ore, the geotechnical boring logs guide covers SPT, CPT and lab-test records instead.

Why QA/QC decides whether assays are usable

Assays are only trainable labels when you can show the lab results were in control, which means the QA/QC records must come with the assay table. Ask for certified reference materials (standards) with expected values, blanks inserted after high-grade samples, field, coarse and pulp duplicates, and umpire lab checks. Each should link to the batch or certificate number so you can drop batches that failed.

Common failure modes are specific and fixable if you see them early. Below-detection values are stored as negatives, zeros, "<0.01" strings or half the detection limit, depending on the database. Over-range values get re-assayed by a second method, and both values sit in the table. Sample intervals rarely match logging intervals, so you need a stated compositing rule.

A study on cleaning investigative drilling data makes the broader point: drilling data is not collected for machine learning and needs substantial preprocessing before use [3].

Public versus private drill data

Public drill data is real but uneven, and private company databases are usually where the dense, recent, QA/QC-controlled holes live. In the US, the USGS publishes data releases with drill core geochemistry, and state geological surveys and drill core libraries hold legacy holes, logs and sample archives [7]. Expect older analytical methods, mixed detection limits and incomplete collar surveys in legacy releases.

Listed companies reporting to the SEC file technical report summaries under Subpart 1300 of Regulation S-K, prepared by qualified persons and filed as exhibits [6]. These reports are public and describe drilling, sampling and QA/QC programs, but they are summaries. The underlying drill database with every interval, assay and survey shot is typically not filed and has to be licensed from the holder.

Location sensitivity shapes these deals. Collar coordinates and grades reveal targets, so holders may restrict or delay release around claim staking, financing or a pending resource update. Ask early whether coordinates can be delivered in full, offset into a local grid, or withheld for specific prospects. Our geospatial and location data page covers licensing location data more generally.

Core photos and paired logs

Core photos are most valuable when every tray image is registered to depth and paired with the logged codes for those intervals. Without registration, you have tray images and a separate log, and someone has to reconstruct the alignment by reading block markers. Ask for tray ID, from/to depth per tray row, wet versus dry photography, lighting setup, camera or scanner type, and whether hyperspectral scans exist.

Logging consistency is the hidden label-quality issue. Different geologists log the same unit differently, and code dictionaries change between campaigns. Request the code dictionary with version dates and, where available, the logger ID per interval so you can measure inter-logger agreement. Image licensing terms are covered on our images and inspection photos page.

Request template for a drill hole data license

A precise request names tables, fields, formats and acceptance checks, which lets a holder judge quickly whether its database fits. Use the template below as a starting structure.

Illustrative example: invented to show structure; it does not describe an available dataset.

request: drill hole dataset for porphyry Cu-Au prospectivity and core-logging models
commodity: [Cu, Au, Mo]
deposit_style: porphyry and related skarn
geography: US, any state; exact project location not required at request stage
tables:
  collar: [hole_id, easting, northing, elevation, crs_epsg, total_depth_m, drill_type, start_date]
  survey: [hole_id, depth_m, azimuth_deg, dip_deg, survey_method]
  lithology_log: [hole_id, from_m, to_m, lith_code, alt_type, alt_intensity, min_style, vein_pct, recovery_pct, rqd, comment]
  assay: [hole_id, sample_id, from_m, to_m, lab, method_code, element, value, unit, detection_limit, certificate_id]
  qaqc: [sample_id, sample_type (standard|blank|duplicate|umpire), crm_id, expected_value, batch_id]
  core_photos: [tray_id, hole_id, from_m, to_m, wet_dry, image_file]
formats: CSV or Parquet for tables; JPEG or TIFF for photos; code dictionary as CSV
minimum_quality:
  - surveys present for all inclined holes
  - assay certificates or batch IDs linkable to QA/QC results
  - below-detection convention documented
documentation: data dictionary, code dictionary with versions, provenance notes
location_handling: full coordinates, local grid offset, or withheld per prospect
intended_use: model training and internal evaluation

Ship documentation in a machine-readable form where you can; Croissant-RAI is one vocabulary for provenance and life-cycle metadata alongside the files [5]. Add a validation pass on receipt: overlapping or gapped intervals, holes with assays but no collar, and surveys deeper than total depth are the usual first findings.

Licensing questions specific to exploration data

Exploration data licenses turn on who owns the data and what the holder may share, which is often less clear than it looks. Joint ventures, earn-in agreements and option deals can give a partner rights over drill results, and data acquired with a property may carry restrictions from the prior owner. Ask the holder to confirm it can license each project's data, not only that it holds it.

Agree in writing on whether derived models, embeddings and synthetic holes are allowed, whether you may combine the data with public survey releases, and how coordinates may be shown in any publication or demo. If you also need operational context, our guide to daily drilling reports covers activity codes and time logs, and the tabular and time-series buyer's guide covers formats and schema checks across structured data.

How SourceX handles drill hole data requests

SourceX sources operational datasets from US companies on request; categories are not inventory, and a request does not guarantee a match. You describe the data you need, and SourceX looks for US businesses that hold it, with every release approved by the supplying company. Each dataset is rights-reviewed for ownership and consents and delivered under a license defining records, uses, term and delivery. The process runs Find, Assess, Agree, Transact and Manage, and nothing is contracted until a supplier agrees.

Start a request on the SourceX buyers page.

Request drill hole and assay data for your models

SourceX sources operational data such as engineering records and documents from US companies on request and manages the commercial process, including licensing agreements and ongoing purchases. Personal details are removed or replaced before delivery, and delivery runs through private, access-controlled workflows after an executed agreement and supplier approval. Describe the tables, fields and quality checks you need on the SourceX buyers page.