Loading…
Loading…
Usage Guide
PointCheck evaluates your photos (Front required; six angles recommended) plus your context notes to distinguish authentic artifacts from modern reproductions. Our AI model achieves 94% accuracy, with evidence-based verification to catch potential false positives. Clear, multi-angle photos still drive the quality of the result.
We utilize cutting-edge multimodal AI—a frontier vision model—to analyze flake patterns, Flake scar(glossary entry), notch geometry, material texture, morphological traits, in addition to the context you provide to create a detailed analysis of your arrowhead. Our highly customized prompts are specifically designed for lithic analysis, optimizing the AI's output for arrowhead assessment. Results are then verified using evidence-based detection algorithms developed from extensive testing on a high-confidence dataset.
Accuracy depends on what the model can see. Upload as many clear, well-lit angles as possible. Use neutral backgrounds, avoid oil/water, and keep lighting even so flake scars, material texture, and edge geometry are visible.
Front is required; include as many of these angles as you can for best accuracy. More angles give the AI model a truer "3D picture," reducing uncertainty.
The context (the questionnaire) you supply helps the AI weigh what it sees, but images are the primary means by which the model assesses your point. All fields are optional except acquisition method; honest context plus clear angles gives the model better evidence to work with.
Provenance
Where and how it was found can align (or conflict) with known site types and materials.
Material type
Knowing chert, flint, obsidian, or quartzite helps match expected textures and Pressure flaking(glossary entry) signatures.
Material locality
Local vs. non-local stone can support or challenge authenticity claims.
Cleaning history
Washing, oiling, or polishing can change sheen and hide Wear polish(glossary entry).
Restoration history
Re-sharpening or re-notching can mimic ancient or modern work; we look for consistency.
Associated artifacts
Flakes, pottery, or bone nearby can support context for authenticity.
Acquisition method
How you got it (found, gifted, purchased) helps interpret risk and expected wear.
Acquisition details
Seller, date, and location add provenance depth when available.
Patina description
Notes on staining or deposits help corroborate visible surface change.
The AI reviews flake patterns, notch geometry, patina clues, material consistency, and your questionnaire context. Results are then verified using evidence-based detection algorithms to catch potential false positives.
How accurate is the AI model?
Our model achieves 94% accuracy on a balanced test dataset of 168 artifacts (84 authenticated, 84 modern reproductions). Evidence-based verification helps catch potential false positives. See our Methodology page for details.
What if I don't know the provenance?
Leave it blank. The AI model still evaluates the photos. Clear, well-lit images matter more than text.
Can the system be wrong?
Yes—treat results as guidance, not certification. Accuracy still depends on your images. Re-upload clearer angles if lighting or focus was poor.
What if the point is damaged?
Photograph damage clearly. The AI model accounts for re-sharpening, re-notching, and breakage when visible.
I’m unsure of the stone type.
Leave it blank or describe what you suspect. The AI will still infer material from texture and color.
Do you store my images?
Only small thumbnails and a SHA-256 hash are stored for buyer/seller verification. With your explicit permission, your full-resolution images may be included in anonymized datasets that may be shared or licensed to academic institutions, research partners, or other third parties who contribute to advancing the study of lithics
Can I ask follow-up questions?
Yes. Use the report link or contact info on the site to share clearer images or context.
What do the verdict labels mean?
They are probability-based guidance: High confidence authentic (≥90%), Likely authentic (70-89%), Possibly authentic (50-69%), Indeterminate / mixed signals (30-49%), or Likely modern (<30%). When we detect conflicting signals (like suspicious evidence patterns), we label results 'Inconclusive' and recommend physical examination.
What does 'Inconclusive' mean?
A result is labeled 'Inconclusive' when we detect conflicting signals between the model's numerical score and its categorical verdict—for example, a high probability score but a 'Likely modern' verdict. This typically occurs with unusual artifacts, atypical materials, or items that challenge standard classification. We recommend physical examination by a specialist.
Why did my Full Analysis come back 'Inconclusive'?
The model detected contradictory signals in your artifact—its numerical confidence and categorical assessment don't align. This can happen with unusual materials, atypical construction, or edge cases that don't fit standard patterns. Consider physical examination by a COA specialist who can assess characteristics not visible in photos.