Skip to content
Taoshi Documentation

Plagiarism Microservice

Previously, we have implemented a plagiarism detection system (PDS) that runs in the background on PTN and tracks recent plagiarism scores. We released a white paper providing technical details and motivations that can be read here: Plagiarism White Paper.

Problem Statement

The existing plagiarism detection system, though useful for monitoring the prevalence of plagiarism on the network, has a few limitations with its current structure:

  1. The established protocol for addressing plagiarism events lacks structure.
  2. Running PDS on validators limits compute resources available for our detection algorithms.
  3. Further development of PDS is slowed down by the additional testing required to ensure that updates don’t disrupt other validator processes.

Proposed Solution

We propose removing the existing plagiarism detection system (PDS) from PTN and running detection on a separate server in a private repository. Information related to plagiarism will be queried by validators and will also be visible on individual miner dashboards.

Removing PDS from validators offloads unnecessary computation and modularizes PTN by isolating processes that were already largely independent. Development time will be reduced as updates to PTN that may have impacted PDS will now be independent. Similarly, updates to PDS can be more rapid since we will not have to test for any disruption to other validator processes. Developing PDS in a private repository inhibits malicious miners from exploiting the plagiarism logic or fine-tuning trading behavior to remain undetected.

Addressing Plagiarism Events

When plagiarism scores surpass a set threshold of similarity, the plagiarizing miner will be locked in probation for 10 days. During this period, the accused miner will have the chance to offer a defense, which we will evaluate. If their explanation or defense is inadequate, the miner will be eliminated. Note that miners in probation are still able to submit orders.

The status of the miner, the cause of demotion, and plagiarism scores will be displayed on the miner dashboard. We will also publicly display a compilation of evidence for cases of plagiarism while obfuscating any sensitive trading signal data. One piece of evidence is the correlation calculated from rasterized net leverages of two miners on particular trade pairs as shown in the plot below:

Example Plagiarism

Timeline

Proposal delivery: Aug 15, 2025

Integration: Aug 22, 2025