Open Science

RefleK's Research

Data analytics and machine learning research environment for human kinetic telemetry.

What is RefleK's Research?

The RefleK's Research repository serves as the open-science data analytics and machine learning research environment for the RefleK's platform. It houses the data processing routines, specialized datasets, and training pipelines used to model human kinetic telemetry.

The repository acts as a structural envelope for isolated, domain-specific research workspaces with their own dependencies and documentation, maintaining an open, language-agnostic structure.

Code editor showing research workspace structure

A Unique Dataset

The research repository is built on a growing archive of real human aiming data collected through RefleK's Desktop.

150,000+

Recorded runs with full telemetry metadata

125,000+

Runs with 125 Hz raw mouse traces

Research Workspaces

Each workspace is an isolated, self-contained environment with its own dependencies and documentation. Models are exported as INT8 Quantized ONNX for native inference in RefleK's Desktop.

GRU training loss curves

Mouse Deviation Analysis (GRU Engine)

A Gated Recurrent Unit neural network trained over cursor acceleration vectors to classify target overshoot and undershoot dynamics in raw mouse tracking streams.

PyTorchGRUONNXCUDA

This workspace hasn't been started yet — contribute your implementation on GitHub.

Contribute on GitHub

Scenario Recommendation (Autoencoder Engine)

A collaborative embedding network and shallow bottleneck autoencoder that projects player history into a latent space, using cosine similarity to identify weaknesses and recommend targeted training routines.

PyTorchAutoencoderONNXEmbeddings

Open Source & Open Science

RefleK's Research is MIT licensed. Models, weights, and datasets are governed by CC BY 4.0. We welcome contributions to existing tracks or new workspace proposals.

View on GitHub