Research Preview

Groove Vision

What if your phone could grade vinyl better than your eyes?

Camera-based surface analysis. LiDAR depth mapping. A machine learning model trained by the collector community. The first objective vinyl grading system that does not depend on human subjectivity.

In active development


The Problem

Vinyl grading is broken.

Every vinyl transaction depends on condition. Mint, Near Mint, Very Good Plus. These grades determine whether a record is worth $5 or $500. And yet the entire system runs on one thing: the seller's eyes and honesty.

Two collectors can look at the same record and assign different grades. Lighting matters. Experience matters. Motivation matters. A seller grading their own inventory has every incentive to round up. A buyer receiving that record has no recourse except the return process.

The Goldmine standard has been the foundation of vinyl grading for decades. It is a good framework. But it was designed for human eyes, and human eyes are inconsistent. The standard itself is not the problem. The application of it is.

Groove Vision is an attempt to solve this. Not by replacing the Goldmine standard, but by applying it with a level of consistency that human vision cannot match.


How It Works

Three layers of analysis.

1

Camera Surface Analysis

The iPhone camera captures high-resolution images of the vinyl surface under controlled lighting conditions. The system analyzes micro-scratches, scuffs, hairlines, and pressing artifacts at the pixel level. Standard iPhone cameras provide enough resolution to detect surface imperfections invisible to the naked eye.

Scratch detection

Scuff mapping

Hairline analysis

Pressing defects

Surface wear index

Ring wear scoring

2

LiDAR Depth Mapping

On iPhone Pro models, the LiDAR scanner adds a third dimension. It builds a 3D surface map of the vinyl, measuring groove depth, warp, and dish. A flat record and a warped record might look identical in a photo. LiDAR knows the difference. This layer is optional but significantly increases grading accuracy.

3D surface topology

Warp detection

Groove depth analysis

Dish measurement

Edge wear profiling

iPhone Pro enhanced

3

Community-Trained ML Model

The neural network learns from every scan. Each time a collector scans a record and confirms or adjusts the suggested grade, the model gets smarter. Thousands of collectors scanning thousands of records, building the most comprehensive vinyl surface dataset ever assembled. The data is anonymous. The model is collective. The result improves for everyone.

Federated learning

On-device inference

Anonymous contributions

Goldmine-aligned output

Continuous improvement

Core ML framework


Architecture

Everything runs on your device.

Groove Vision processes everything locally using Apple's Core ML and Vision frameworks. No images leave your phone. No server sees your records. The only data that leaves the device is anonymous, aggregated model feedback, and only if you opt in.

Input

Camera + LiDAR

Processing

Vision + Core ML

Output

Goldmine Grade

Feedback

Model Improves

Technical Stack

Apple Vision Framework

Image analysis, feature detection, surface classification

Core ML

On-device inference, model updates via background transfer

ARKit + LiDAR

3D surface mesh, depth buffer, point cloud generation

Create ML

Model training pipeline, dataset management, validation

Swift + SwiftUI

Native scanning UI, real-time preview overlays

Open Source Dataset

Anonymized surface data, community-contributed, publicly available


Open Source

This has to be built in the open.

A vinyl grading system only works if collectors trust it. And collectors will only trust it if they can see how it works, verify the data, and contribute to making it better. That is why Groove Vision is being built as an open research project.

The trained model will be open. The anonymized dataset will be open. The grading methodology will be documented and auditable. Any developer will be able to build on top of it. Any collector will be able to verify a grade.

Spinstack will be the first app to integrate Groove Vision, but it will not be the only one. If this works, it should be available everywhere vinyl is bought and sold. Discogs. eBay. Independent shops. The grading standard belongs to the community, not to any single app.

What Will Be Open

Trained Core ML model weights and architecture

Anonymized surface scan dataset with Goldmine grades

Training pipeline and validation methodology

API specification for third-party integrations

Grading methodology documentation


Comparison

Manual grading vs. Groove Vision.

Manual Groove Vision
Consistency Varies by grader Identical every time
Objectivity Seller bias No stake in outcome
Detection Visible defects Micro-level analysis
Warp detection By feel LiDAR measured
Speed Minutes per record Seconds
Improves over time No With every scan

Privacy

Your records stay on your phone.

Groove Vision follows the same privacy principles as Spinstack. All image processing happens on-device using Apple's frameworks. No photos of your records are uploaded anywhere. No server ever sees your collection.

If you choose to contribute to the community dataset, only anonymized surface feature vectors are shared. These are mathematical representations of surface characteristics. They cannot be reverse-engineered into images. They contain no identifying information about you or your records.

Contributing is always opt-in. You can use Groove Vision to grade your records without ever sharing a single data point. The model still works. It just does not learn from your scans.


Roadmap

Where this is going.

Phase 1: Foundation

Now

Camera capture pipeline. Initial surface feature extraction. Basic scratch and wear detection. LiDAR depth buffer integration on Pro models. First training dataset from internal scans.

Phase 2: Beta Scanning

2026

Release scanning UI in Spinstack as a beta feature. Collectors scan their records and confirm or adjust suggested grades. Every confirmation trains the model. Early results will be rough. That is expected and honest.

Phase 3: Community Dataset

2027

Open the anonymized dataset. Publish the model architecture and training pipeline. Invite third-party developers to build on top of it. Release the API specification for marketplace integrations.

Phase 4: Ecosystem

2028+

Groove Vision grades become a trusted standard in vinyl transactions. Integrated into marketplaces, auction houses, and record shops. The dataset grows continuously. The model accuracy approaches and potentially exceeds expert human graders.


Every collector who participates
owns a piece of this.

Groove Vision starts inside Spinstack. Download the app, build your collection, and be among the first to scan when the beta arrives.

Download Spinstack Follow for updates

This is where it starts.