The Random-Sample-Voting Project (the “Project”) ran its first binding polls (and fourth public trial) using the Random-Sample Voting system at the World Forum for Democracy (WFD) 2017, which was held in Strasbourg, France, in November 2017. Around 2000 participants from a more than 80 countries attended the Forum, to participate in the many different tracks and parallel events, involving 206 speakers from 57 countries. Participants came from many different fields from politics (including former heads of states and diplomats), business, law, journalism, academia, political activisms and other walks of life. Satellite events happened in 7 countries, and the general event reached more than 7 million people on social media.

Objectives of the World Forum of Democracy Election

The primary purpose of the World Forum election was to understand public perception of the RSV system when used in a binding poll. This election builds upon the trial at the Global Forum by expanding public exposure of the RSV system to a non-technical audience. The World Forum also provided opportunity to test a new ballot style, alternative deployments of the VBB and ABB, and continue understanding the usability of the RSV software and ballot.

Details of the World Forum of Democracy Election

The following conclusions of the World Forum were put forth for vote by the Council of Europe:

1. Question #1:
   • EN: Parties (traditional and emerging ones) should seek to propose inclusive visions and programmes that deliver benefits for all citizens, and not only for a part of the voters.
   • FR: Les partis (traditionnels et émergents) devraient s’efforcer de proposer des visions et des programmes inclusifs qui confèrent des avantages à tous les citoyens, et pas seulement à une partie des électeurs.
   • RU: Партии (традиционные и новые) должны стремиться предлагать всеобъемлющие подходы и программы, которые приносили бы пользу всем гражданам, а не только части избирателей.
2. Question #2:
   • EN: Participatory and deliberative platforms and initiatives (citizens’ assemblies, juries, forums, etc.) should be structurally embedded into the decision-making processes to balance the oligarchic tendencies of electoral democracy.
   • FR: Les plateformes et initiatives participatives et délibératives (assemblées de citoyens, jurys, forums, etc.) devraient être intégrées de manière structurelle dans les processus décisionnels afin d’équilibrer les tendances oligarchiques de la démocratie électorale.
   • RU: Интерактивные и совещательные платформы и инициативы (собрания граждан, жюри, форумы и т. д.) должны быть структурно включены в процессы принятия решений, чтобы уравновесить олигархические тенденции избирательной демократии.
3. Question #3:
   • EN: Social media should be regulated and held accountable for their impact on a pluralistic, fact-based and hate-free political debate, in the same way as legacy media.
   • FR: Les médias sociaux devraient être réglementés et tenus responsables de leur impact sur un débat politique pluraliste, factuel et sans haine, au même titre que les médias traditionnels.
   • RU: Социальные медиа должны регулироваться и быть ответственными за свое влияние на плюралистические, основанные на фактах и свободные от ненависти политические дебаты, в той же степени, что и традиционные средства массовой информации.
4. Question #4:
   • EN: The use of electronic platforms and big data in party activities and for political mobilisation and campaigning should be transparent and auditable.
   • FR: L’utilisation de plates-formes électroniques et de méga données dans les activités des partis et pour la mobilisation politique et les campagnes électorales devrait être transparente et vérifiable.
   • RU: Использование электронных платформ и больших данных в партийной деятельности, а также для политической мобилизации и проведения кампаний, должно быть прозрачным и поддающимся проверке.
5. Question #5:
   • EN: Civil society organisations defending human rights and equality against attacks by populists should agree on a common agenda and strategy across identity politics divides.
   • FR: Les organisations de la société civile qui défendent les droits de l’homme et l’égalité contre les attaques des populistes devraient s’entendre sur un agenda et une stratégie communs à travers les clivages identitaires.
   • RU: Организации гражданского общества, защищающие права человека и равенство против популистских нападок, должны согласовать общую повестку дня и стратегию для преодоления разногласий по вопросам идентичности.

Each race was configured based on an estimation of the conference attendance, with 360 real ballots and 40 decoy ballots per race. To ensure with reasonable certainty that no ballots would be duplicated, the election was defined to have a roster size of 10,000. The poll ran during the Thursday of the conference from 11am to 11:30pm local time. Our field team distributed paper ballots to conference attendees as they entered their respective morning sessions and were given a sticker to place on their badge upon exiting to ensure no one received multiple ballots. An unvoted sample ballot for each race is available here:

• Question 1
• Question 2
• Question 3
• Question 4
• Question 5

Election Software

The software deployed for this election was a modified deployment of the full system used at the 2016 Global Forum for Modern Direct Democracy (Global Forum), including the election authority software, the audit bulletin board (ABB), voter bulletin board (VBB), and verifiable random beacon. In the interest of transparency, the software used in the trial is available upon request for auditing and research purposes. Please direct all requests to software (at) rsvoting.org. The software that may be requested include:

• Election Authority
• Audit Bulletin Board
• Voter Bulletin Board

Verifiable Random Bit Generation

The entropy source used for the election was the NIST random beacon. The United States National Institute of Standards publishes “random” values every minute online via their random beacon. While not as “un-manipulatable” as other entropy sources, the NIST random beacon provides high-throughput and low-latency bits, which was necessary to meet the tight scheduling requirements of the election. For the initial draw, the verifiable randomness accumulation times were staggered to ensure each election would have a different subsequence of random bits. In the future, it is hoped that the RSV protocol’s division into batches can be done based on more than one source of randomness, one allowing immediate posting of results and the other providing stronger integrity later.

Election Procedure

The timeline below describes the events that took place in running the election. The Project managed the election through the Election Authority software.

• Monday, November 6, 2017, 7:01 UTC – The Project published the election definitions to the ABB and VBB.
• Monday, November 6, 2017, 7:29 UTC – The Project published the election definitions to Origin Stamp, a Bitcoin-based timestamping service.
  • Definition Timestamp for Question #1
  • Definition Timestamp for Question #2
  • Definition Timestamp for Question #3
  • Definition Timestamp for Question #4
  • Definition Timestamp for Question #5
• Monday, November 6, 2017, 8:00 UTC – The Project began accumulating random bits from the NIST beacon for Question #1.
• Monday, November 6, 2017, 8:05 UTC – The Project began accumulating random bits from the NIST beacon for Question #2.
• Monday, November 6, 2017, 8:10 UTC – The Project began accumulating random bits from the NIST beacon for Question #3.
• Monday, November 6, 2017, 8:15 UTC – The Project began accumulating random bits from the NIST beacon for Question #4.
• Monday, November 6, 2017, 8:20 UTC – The Project began accumulating random bits from the NIST beacon for Question #5.
• Monday, November 6, 2017, 9:00 UTC – The Project finished accumulating random bits from the NIST beacon for Question #1.
• Monday, November 6, 2017, 9:05 UTC – The Project finished accumulating random bits from the NIST beacon for Question #2.
• Monday, November 6, 2017, 9:10 UTC – The Project finished accumulating random bits from the NIST beacon for Question #3.
• Monday, November 6, 2017, 9:12 UTC – The Project queried the random beacon for the initial draw’s verifiable random bits for Question #1. The Project published the initial draw to the ABB, committed the final summands to the ABB, and published the ballot authentication information to the VBB.
• Monday, November 6, 2017, 9:15 UTC – The Project finished accumulating random bits from the NIST beacon for Question #4.
• Monday, November 6, 2017, 9:20 UTC – The Project finished accumulating random bits from the NIST beacon for Question #5.
• Monday, November 6, 2017, 9:25 UTC – The Project queried the random beacon for the initial draw’s verifiable random bits for Question #2. The Project published the initial draw to the ABB, committed the final summands to the ABB, and published the ballot authentication information to the VBB.
• Monday, November 6, 2017, 9:42 UTC – The Project queried the random beacon for the initial draw’s verifiable random bits for Question #3. The Project published the initial draw to the ABB, committed the final summands to the ABB, and published the ballot authentication information to the VBB.
• Monday, November 6, 2017, 9:50 UTC – The Project queried the random beacon for the initial draw’s verifiable random bits for Question #4. The Project published the initial draw to the ABB, committed the final summands to the ABB, and published the ballot authentication information to the VBB.
• Monday, November 6, 2017, 9:58 UTC – The Project queried the random beacon for the initial draw’s verifiable random bits for Question #5. The Project published the initial draw to the ABB, committed the final summands to the ABB, and published the ballot authentication information to the VBB.
• Monday, November 6, 2017, 10:30 UTC – The Project sent the ballots to the Council of Europe printing service.
• Thursday, November 9, 2017, 10:00 UTC – The VBB opened the polls and began accepting votes. The field team began distributing ballots in parallel to the eight sessions; the ballots were all real and randomized to ensure uniform distribution across the attendees.
• Thursday, November 9, 2017, afternoon UTC – The field team began distributing decoy ballots to those who would take them.
• Thursday, November 9, 2017, 22:30 UTC – The VBB closed the polls and stopped accepting votes.
• Thursday, November 9, 2017, 22:42 UTC – The Project began the process of downloading the votes from the VBB, generating the print audit and tally information, and committing the data to the ABB in preparation for the final draw.
• Thursday, November 9, 2017, 22:49 UTC – The Project completed processing in preparation for the final draw.
• Friday, November 10, 2017, 05:00 UTC – The NIST random beacon began generating randomness to be used for the Final Draw for Question #1.
• Friday, November 10, 2017, 05:10 UTC – The NIST random beacon finished generating the random bits for the Final Draw query and the verifiable randomness for the Final Draw was determined for Question #1.
• Friday, November 10, 2017, 05:10 UTC – The NIST random beacon began generating randomness to be used for the Final Draw for Question #2.
• Friday, November 10, 2017, 05:20 UTC – The NIST random beacon finished generating the random bits for the Final Draw query and the verifiable randomness for the Final Draw was determined for Question #2.
• Friday, November 10, 2017, 05:20 UTC – The NIST random beacon began generating randomness to be used for the Final Draw for Question #3.
• Friday, November 10, 2017, 05:30 UTC – The NIST random beacon finished generating the random bits for the Final Draw query and the verifiable randomness for the Final Draw was determined for Question #3.
• Friday, November 10, 2017, 05:30 UTC – The NIST random beacon began generating randomness to be used for the Final Draw for Question #4.
• Friday, November 10, 2017, 05:40 UTC – The NIST random beacon finished generating the random bits for the Final Draw query and the verifiable randomness for the Final Draw was determined for Question #4.
• Friday, November 10, 2017, 05:40 UTC – The NIST random beacon began generating randomness to be used for the Final Draw for Question #5.
• Friday, November 10, 2017, 05:50 UTC – The NIST random beacon finished generating the random bits for the Final Draw query and the verifiable randomness for the Final Draw was determined for Question #5.
• Friday, November 10, 2017, 06:00 UTC – The Project began revealing the batches of auditable columns on the ABB based on the final draw.
• Friday, November 10, 2017, 06:18 UTC – The Project finished revealing the batches of auditable columns on the ABB based on the final draw.
• Friday, November 10, 2017, 08:00 UTC – The Project announced the results of the polls.

Election Results and Audit

A total of 834 real ballots and an unknown number of decoy ballots were distributed; the number of decoys being unknown as the entire stack of decoys went missing during the event. A total of 120 ballots were voted, of which 117 were real and 3 were decoys. The voting breakdown is as follows:

Question #	# Ballots Voted	# Decoys Voted	# Real Votes	Total For	Total Against
1	20	0	20	18 (90.0%)	2
2	26	0	26	20 (76.9%)	6
3	25	0	25	22 (88.0%)	3
4	24	0	24	22 (91.7%)	2
5	25	3	22	16 (72.7%)	6

A total of 72 duplicate ballots were generated from the election configurations. The roster size was chosen to be large enough to ensure with reasonable certainty that no duplicate ballots would be generated. The breakdown for the duplicates is as follows:

Question 1	Question 2	Question 3	Question 4	Question 5
12 (3.33%)	14 (3.89%)	14 (3.89%)	12 (3.33%)	20 (5.56%)

A cursory audit of the election was conducted by the Project, which entailed manual inspection that the data published to the ABB was consistent. An independent audit of the election has not yet been performed, as far as the Project is aware. The fully auditable election data derived from the VBB and ABB is available, allowing anyone to conduct a more complete, independent audit.

• Election Authority certificate
• Question 1 VBB data
• Question 1 ABB data
• Question 2 VBB data
• Question 2 ABB data
• Question 3 VBB data
• Question 3 ABB data
• Question 4 VBB data
• Question 4 ABB data
• Question 5 VBB data
• Question 5 ABB data

Caveats

This election did not employ all of the procedural controls one would expect of the full RSV system. In a complete deployment, for full privacy protection, we would handle ballot printing and distribution with appropriate controls to protect the secrecy of the vote codes and the privacy of the voter.

Comparison to Other World Forum Voting Systems

Random-sample voting was one of four votes to take place during the World Forum. The other systems included: evaluative voting on the same recommendations as the RSV vote (accessed through the same ballot), a push-button system to award a significant prize, and an official session sanctioned and organized by the Council of Europe. Although beyond the scope of our initial objectives, sharing the stage with a variety of voting systems provided an opportunity to do a comparison.

• The evaluative vote organized by Herrade Igersheim had a turnout of 67 voters (out of 834).
• The vote awarding a big prize was run using a push-button system. The total participation rate was 70-75% of the people present in the room despite it 1) having a negligible barrier to participate, 2) direct measurable impact (i.e., big prize), and 3) the voters being well-informed, responsibilized, and otherwise idle.
• The vote organized by the Council of Europe and advertised in the conference program was for the Citizen’s Assembly to create and vote on a variety of recommendations. The vote attracted 25 individuals of which 21 voted.

The vote organized by Herrade Igersheim had significantly lower turnout than the RSV system despite voters in the evaluative system and RSV system being encouraged to vote in both systems on their ballots and upon casting their votes. There are several possible reasons for this including voting fatigue, perception of redundancy, and voters preferring one system over the other. Based on prior trials, RSV had an average turnout. Other voting systems, to include the incentivized push-button system, had lower turnout than one would expect. This may indicate general voter apathy rather a fundamental problem with usability of each voting system.

Reception by the World forum

The technical side of the voting system appeared to work flawlessly. Several people expressed interest in RSV and the Project, represented by Nicolas K. Blanchard and David Chaum, who gave talks and multiple interviews for newspapers and radio shows.

Lessons Learned

This election operated on a significantly compressed schedule compared to prior trials. This was to fit into the schedule determined by the Council of Europe. The World Forum election had a repeat issue of low voter turnout due to loose integration with the event; but the World Forum election brought up several new challenges and lessons learned, especially regarding low voter turnout.

1. Like the 2016 Global Forum for Modern Direct Democracy (Global Forum), the World Forum election also suffered from low voter turnout. However, this was not due to technical difficulties as experienced at the Global Forum. The potential reasons included confusion over voter eligibility (i.e., rapporteurs, conference organizers, and minors), political sensitivities (i.e., high-ranking political officials), ballot distribution timing, quality of exposure over the course of the event, and apparent voter perception that the conclusions were self-evident (i.e., voter apathy). The Project further observed that even incentivized push-button polls received significantly lower voter turnout than expected. This shows that the updated design and interface were successful, but signals that voter turnout should still be a prioritized research area for the Project.
2. Despite improved integration with the World Forum over the Global Forum, voter awareness of the voting system was suboptimal. Inclusion of these votes into the formal proceedings may improve voter awareness and turnout.
3. Having a dedicated voting station with volunteers to assist voters may improve turnout and help reduce confusion regarding the RSV system. This may be particularly true of international attendees without data plans or other Internet access (it had been raised as a possibility early on but the Council of Europe could not handle the logistics)
4. The Project reaffirmed conventional wisdom in that individual voter awareness/education and rewarding participation improves voter turnout. During the Global Forum, individual discussions with participants by the field team increased voter turnout by 15 (out of 36). During the World Forum, the Project increased voter turnout by 30-40 (accounting for 25-35% of cast ballots). This style of campaigning, while effective, fails to scale without sufficient volunteers and activism.
5. Several features of the ballot continue to confuse voters. The double-ballot which has been used in each election thus far is still confusing to many users although none came to the field team and asked for an explanation (which had happened at the Global Forum).
6. The RSV system has garnered significant attention from an international audience. Part of the reason for low turnout was the unexpectedly large attendance by individuals who did not understand the three official languages of the World Forum. On the technical side, the RSV software will need to be reworked to better facilitate elections targeted for an international audience.

Future Work

The World Forum brought to light a few interesting lines of research.

1. Venues such as the World Forum are opportunities to test new voting systems. How does the presence of multiple voting systems at a single event affect voting patterns?
2. To discourage ballot trading, the Project did not reveal the ballot questions until polling time. It is unknown whether this affected voter turnout, ballot trading, or voter perception regarding election integrity.

Appendix: NIST Random Beacon

At the time of publication, the NIST random beacon uses two independent hardware true random number generators (TRNG) made with SP 800-90 compliant components. Together, these TRNGs generate 512 bits each minute. Each record is chained, timestamped, and signed by NIST.