Accuracy and identification methodology

The service analyzes the submitted specimen image for visible characteristics such as color, grain, luster, texture, layering, crystal habit, fractures, and apparent structures. It compares those cues with learned visual patterns and returns a suggested name, classification, evidence, and short reasoning summary.

The model receives the image and limited request context. It does not physically handle the specimen, know the collection location unless you independently verify it, or perform a geological test.

Confidence summarizes how strongly the model's visual interpretation supports the returned suggestion for that image. It is not a calibrated probability that the specimen has been correctly identified, a laboratory accuracy score, or a guarantee.

A sharp image can still produce a plausible but incorrect match when different materials share similar visible features. Treat close lookalikes as alternatives to investigate, not as confirmed exclusions.

Many distinctions require evidence a photo does not contain. Weathering, coatings, lighting, camera processing, scale, and mixed specimens can also hide or distort diagnostic features.

• Mineral confirmation may require hardness, streak, cleavage, specific gravity, magnetism, fluorescence, or chemical tests.
• Rock classification can depend on grain relationships, mineral proportions, field context, or thin-section analysis.
• A possible fossil may require scale, multiple views, locality, preparation, and expert examination.
• Synthetic, dyed, treated, composite, or imitation specimens may resemble natural materials in a single image.
• Monetary value cannot be established from a photo result; condition, authenticity, weight, provenance, market, and professional grading can matter.

Image quality affects the clues available to the model. Retake the photograph if the specimen is blurred, partly hidden, strongly filtered, or dominated by glare.

• Show one specimen in focus against a plain, contrasting background.
• Use soft, even daylight and keep colors natural.
• Include the full specimen and a neutral size reference without covering key features.
• Capture more than one face in the mobile app when structure varies across the specimen.
• Do not clean, scratch, break, taste, heat, or apply chemicals to an unknown specimen solely because an automated result suggests it.

Use a qualified geologist, mineralogist, gemologist, paleontologist, laboratory, or relevant local authority when identity affects safety, scientific reporting, ownership, conservation, sale, insurance, or a significant financial decision. Do not rely on the app to declare a specimen safe, non-toxic, non-radioactive, asbestos-free, legally collectible, or valuable.

We do not publish a universal accuracy percentage because performance varies by specimen category, visual similarity, image quality, and capture conditions. A single unsupported number would be misleading. If a benchmark is published, it should identify the app and model version, labeled dataset, sample size, category mix, capture conditions, evaluation method, and error distribution.

The Glipo product team reviews product claims and methodology guidance on this site. We distinguish model confidence from identification accuracy, label photo results as suggestions, avoid presenting generated value context as an appraisal, and correct material errors when identified.