The following screenshots are taken from the Flare and Songbird monitors. They highlight clear price similarities between FLR Labs, Civil FTSO and two other providers.
They exhibit similar patterns for four coins; ARB, FIL, XRP and LTC, over the span of approx. 1 week.
Typically, all of these providers have a very high hit rate for these coins, so it is often hard to spot consistencies in the prices. In each case, the initial screenshot shows a zoomed-out view to give some perspective, whereas the second one is focused a little closer on the prices themselves. The prices are that close to each other it is often difficult to even determine that there are multiple providersâ prices being displayed. Each set of screenshots also has a link to the relevant monitor, where the information can be reviewed in more detail.
At some point during each of the time windows, the providers appear to have had issues where their submitted prices jumped and were uncharacteristically higher than the rewards bands. Whilst this on its own could be a simple case of a single provider having issues with their feeds, what is telling is that in every case the issues happened to each of the providers at precisely the same time and continued on for the exact number of epochs, before returning to ânormalâ. Additionally, price movements of each provider, during these phases are almost identical.
It is hard to conceive of a scenario where this kind of issue would happen to unconnected providers in the same manner, however we would welcome some comments from the providers in question around these anomalies, and of course the view of other management group members.
To put this evidence in perspective, Aureus Ox has been dealing with degraded performance for the past 2 weeks. We periodically have had to restart and in turn re-sync our provider over this period. This restart would pull our prices outside of the reward band. The probability that some other provider running a different algorithm was doing this at the same exact time as Aureus is zero to none.
Stack that probability on top of the probability that another provider wasnât just submitting off prices but also the same off prices.
Further more, stack that probability on top of the probability that both providers recovered at the same exact time.
Not to mention, this is a total of 4 providers exhibiting the same exact behavior in the same exact. This isnât just collusion, these are clearly bad actors.
On behalf of FLR Labs, our explanation for occasionally mirroring price trends seen with other data providers lies in our use of RNN (Recurrent Neural Network) architectures. Despite these similarities, the distinctiveness of our proprietary methods is essential. Representing FLR Labs, I assure you that our results stand apart from our competitors. We also encourage data providers to appreciate our commitment to continuous development and reinvestment before drawing premature conclusions.
Labeling us as a bad actor is absurd. We are deeply invested in our development and expansion efforts. Weâve established partnerships with academic institutions like the RMIT Blockchain Innovation Hub, sharing knowledge on Flare protocols and seeking new growth opportunities. Our team is focused on drafting, testing, and developing a whitepaper for further expansion. We have also developed a dApp that compiles and incentivizes users to interact with the delegation process and have built a space with a unique 3D environment using Unity, utilizing the FTSO pricing mechanism for NFT minting. This not only drives user engagement with the FTSO system but also aligns with our broader goals.
Furthermore, weâve initiated talks with The Giving Block to integrate charity donations on our site, emphasizing our commitment to social responsibility and technological advancement for public welfare. In contrast, we notice that some data providers, possibly including your organization, are not reinvesting or demonstrating growth ambitions. This stagnation could be impeding the networkâs growth, and yet you label us as bad actors. What a joke.
The chances of two, let alone four completely isolated RNN models producing the exact same behaviors as detailed above, is incredibly unlikely. There are simply too many variables to consider. Changing the number of layers, the amount of neurons per layer, the learning rate, the batch size, the type of optimiser (to name a few) even slightly would not produce these kind of anomalies, in my opinion. Layering on top of that, the added variable of input data (data sources, prices and other features used in the model), this expands the unlikelihood considerably.
Whatâs absurd is having two separate RNN models mirroring the same results and same anomalies at the same time.
If you had took the time to understand how any of this works, you may not find yourself labeled as a bad actor. Who is your team? Is the team who developed your FTSO algorithm the same âteamâ that built your NFT minting mechanism + 3D environment? If so, maybe you should look into your âteamâ.
Whatâs a joke is that you tout partnerships instead of looking into the evidence that was presented. Did you look at it? REALLY look at it. Do you even know what the numbers mean?? Would you like to talk about jokes?
On behalf of FLR Labs, in line with best practices, our explanation of our RNN usage should be brief. Delving deeper into our algorithm would be counterintuitive, as we are all data providers, and revealing more about our algorithm methodology should be prevented. However, we can confirm that RNN is one of several methods we employ, which accounts for the scenarios depicted in the above screenshots.
First of all want to thank Ben for his hard work in finding and compiling this information.
As for FLRlabs i am not convinced that you actually use an RNN model as there are other models better suited for a system like the FTSO.
Also its poor as an excuse as especially for RNN there will be an even lower chance for something like this to happen than a linear ML model.
We will wait for the discussion to fully mature before actually deciding what we will vote for.
Thanks for doing the research on this, Ben. The results speak for themselves. The likelihood of these patterns happening to independent providers, across multiple symbols is highly unlikely, if not impossible. The justification given doesnât stack up. Iâd like to know who the other 2 providers are, they have zero presence as far is I can see, so it stands to reason theyâre clones of existing providers.
Given the evidence Bushi also supports the proposal. As others have stated, training an RNN has so many factors its hard to imagine a scenario where 4 separate models have been trained to such a similar result.
Does Civil FTSO have access to this forum?
The monitor evidence is pretty damning but i also find it funny how their website title is Civil FTSO - NFT Marketplace HTML Template.
And the content of said website is just some random ethereum mumbo jumbo.
There is so much wrong with that site, itâs difficult to know where to start. Perhaps that not a single one of the links actually does anything. Or that the NFTâs are not real. Or of course, the stock team images, as you point out.
To answer your question around Civil having access to this thread (to respond, as opposed to just view). Then that would depend, I guess.
If they are a clone of FLR Labs then the answer is yes. If not and they are using a shared algo and price feed with FLR Labs, then no, Iâm fairly sure they donât. If this is all one big misunderstanding, and itâs just âRNN being RNNâ, then perhaps we could maybe reach out to âJohnny Englishâ for further clarification.
I was hoping for a more engaged discussion around the allegations put forward here, but have seen nothing yet.
