We'll get to our latest update in a moment. First, a reminder about what the Augur Project is all about.
Well, we've grabbed Tomas (more on that below) along with a detailed road-map for building a completely new and battle tested front end. Augur has purchased the intellectual property rights to use Tradesports' new trading interface and brought its UI developer on board to head up and completely modernize Augur's front-end development.
![broken image](http://user-images.strikinglycdn.com/res/hrscywv4p/image/upload/c_limit,f_auto,h_2000,q_90,w_1200/428927/ui-votes_jyyhel.png)
The purchase of the interface involved transferring 100,000 REP token from the Forecast Foundation's rep storehouse to Tradesports.
Integrating a completely new UI to our back end will push back the release of our beta version from what had been the best case scenario of December 25. We want to deploy and go live as soon as possible - we'd do it today if we could - but a bigger priority is ensuring that when Augur beta is deployed, the platform is an attractive storefront that commands respect and sets the stage for later widespread use and adoption when we come out with the later, final live real-money version - that after we build it, they will come. We're more confident that will happen by incorporating a modern and well-liked UI asap.
This is big, but there's a lot more going on:
![broken image](http://user-images.strikinglycdn.com/res/hrscywv4p/image/upload/c_limit,f_auto,h_2000,q_90,w_1200/428927/tomas-p_ncpwde.jpg)
As mentioned above, we're thrilled to welcome Tomas Priecinsky to the Augur Team. He joined Intrade back in 2011, where he gained experience on how to build various UIs - simplified for basic users, advanced for real-time traders and compact for mobile devices. Then he went to work for Barclays in 2013 before coming back to the development team that previously created Intrade. Tomas was a core developer of the Tradesports fantasy sports platform. He yet again worked on real-time UIs and building multi-platform hybrid mobile applications while gaining exposure to bitcoin. He's skilled at JS, CSS, Java, SQL and various JS frameworks. He spent a few years at The University of Economics, Prague (Vysoká škola ekonomická v Praze) in the Czech Republic before moving on early to a career in programming. He enjoys sports, culture, bacon and programming.
![broken image](http://user-images.strikinglycdn.com/res/hrscywv4p/image/upload/c_limit,f_auto,h_2000,q_90,w_1200/428927/ryan_mewjie.jpg)
Done:
- Price History, Market commenting, Market metadata off chain, Market searching, Market Load Optimizations
- the backend of the system has had all features needed for the beta since June. In other words, core functionality is already in place and accessible via command line. There's work still to be done on the front end to make it pretty, easily accessible and user friendly (see above)
- we've been refining the reporter system to make it more resilient
- we've added adjustable trading fees in realtime: market creators can now readjust trading fees at any time while the market is active, if they wish to adapt to changing volume conditions, for instance.
- we've added a function to allow for the initial REP distribution
- we've added a contract to allow sub currencies instead of just generic fake cash
- we've added a psuedoether subcurrency contract (making eth behave as a subcurrency, useful for generalization)
- came up with a reporting threshold system to allow randomized reports on given events/markets
- rewrote most functions so that they are lazily evaluated, basically making better use of scarce resources
- wrote a quick sort algorithm for finding the median of an outcome upon resolving (useful for things with less than 200 reports)
- set up the basic appeals process for events reported by the reporting system as "indeterminate/unethical": in other words, made it so that an irresolvable event report is pushed back and reported on again before it is final
- implemented the new penalty/loss metrics outlined in "Scaling Augur" (1, 2)
- added a lazily evaluated claim system to claim redistributed rep and trading fees (as opposed to doing it for all reporters at once, which makes the gas cost super high)
- made it so if an event gets pushed back (by being judged by the reporter system as being “indeterminate”) people can still buy and sell in the market
- implemented penalties under the new system for not reporting enough on randomly selected events/markets
- added ethicality as a separate reporting option per market/event
- created a market fork choice framework so there are possible ways to decide which fork to place a current market on in the event of a fork
- implemented a separate initial branch for general oracle system use outside of prediction markets
- figured out how to fork scalar markets (one fork has original outcome, the other re-reports)
- only by default resolve markets at less than 99 odds
- create separate sets of rep for dormant (i.e. on exchange/trading) and active rep
What remains to be done:
- Automated testing, Categorical / Scalar markets, Market sorting, Holdings improvements, Limit Orders, Artificial Economy, Stack failure tolerance, Tutorial, Add limit orders (from the Beta roadmap)
- implement forking (mechanism by which network will be allowed to branch off into a new network if there's a large enough loss of confidence in the existing reporting system)
- implement 'backstop one" where everyone reports on a market/event
I'm actually quite pleased with the progress so far, since at the (crowd)sale start we had none of that stuff above done and no idea on how to scale the system in an actually sustainable manner whilst maintaining rock solid incentive-compatible security.
- paid back all debts and other commitments that the Project had accumulated up to the crowdsale
- addressed US state tax paperwork
- interviewed people for development positions
- set up and secured an EU (in addition to our US) bank account
- got caught up on our legal situation and strategy with our Pillsbury Legal team and coordinating some related technical work
- sold all but 1,000 BTC and 100,000 ETH via over the counter (OTC) trades
- opened up a brokerage account to secure a portion of our funds in relatively secure, short term US Treasury bonds, to earn some return above bank rate interest
- zapped various crowdsale site bugs
Remains to be done:
- flow diagram of the entire system from start to finish (including all the edge cases / backstops)
- prepare Augur tutorial
Augur has also contributed $20,000 to an R&D project to transition the Ethereum network from a proof of work (POW) to a proof of stake (POS) system so it can operate at higher speeds, handle more transactions and thereby become more competitive with existing global financial networks and a more viable alternative to them.
Transitioning to a POS system would help us approach those levels, offering the closest thing to instantaneous speed and world-class financial network transaction bandwidth.
Augur is also funding research to develop an AI-powered decentralized reporter helper system, powered by natural language processing and machine learning, that would recommend answers to reporters based on scans of online news sources, and learn from which recommendations are accepted or rejected. The idea is to reduce the burden of reporting on what could eventually end up being hundreds of markets, and have the machinery improve its own performance by "learning" from how its recommendations are handled by reporters during each round of reporting. That research is being headed by NLP expert Petr Baudis, the mind behind Yoda Q&A, the best known open source alternative to IBM Watson.
His latest update:
precision (while letting around 30% of question pass
a preliminary filter) during phase 2...as we improve the machine learning paradigms and hopefully (this) will turn back upwards when we resume adding determinative features...we are exploring ways to make the system answer by identifying deeper semantic features in the news stories. So far, we are still bootstrapping to have enough features for meaningful determination of answers. Right now we decided that we need a dataset which says whether each news article is relevant to each question, so we are using Amazon Mechanical Turk to build that; another limitation is a small dataset, which makes our evaluation a little noise third, raw newspaper texts often confuse off-the-shelf parsers, so we will need to explore more alternatives there."
Recent and Upcoming Appearances